1 | /* |
2 | * Copyright 2022 Advanced Micro Devices, Inc. |
3 | * |
4 | * Permission is hereby granted, free of charge, to any person obtaining a |
5 | * copy of this software and associated documentation files (the "Software"), |
6 | * to deal in the Software without restriction, including without limitation |
7 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
8 | * and/or sell copies of the Software, and to permit persons to whom the |
9 | * Software is furnished to do so, subject to the following conditions: |
10 | * |
11 | * The above copyright notice and this permission notice shall be included in |
12 | * all copies or substantial portions of the Software. |
13 | * |
14 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
15 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
16 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
17 | * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR |
18 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
19 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
20 | * OTHER DEALINGS IN THE SOFTWARE. |
21 | * |
22 | * Authors: AMD |
23 | * |
24 | */ |
25 | #include "display_mode_vba_util_32.h" |
26 | #include "../dml_inline_defs.h" |
27 | #include "display_mode_vba_32.h" |
28 | #include "../display_mode_lib.h" |
29 | |
30 | #define DCN32_MAX_FMT_420_BUFFER_WIDTH 4096 |
31 | |
32 | unsigned int dml32_dscceComputeDelay( |
33 | unsigned int bpc, |
34 | double BPP, |
35 | unsigned int sliceWidth, |
36 | unsigned int numSlices, |
37 | enum output_format_class pixelFormat, |
38 | enum output_encoder_class Output) |
39 | { |
40 | // valid bpc = source bits per component in the set of {8, 10, 12} |
41 | // valid bpp = increments of 1/16 of a bit |
42 | // min = 6/7/8 in N420/N422/444, respectively |
43 | // max = such that compression is 1:1 |
44 | //valid sliceWidth = number of pixels per slice line, |
45 | // must be less than or equal to 5184/numSlices (or 4096/numSlices in 420 mode) |
46 | //valid numSlices = number of slices in the horiziontal direction per DSC engine in the set of {1, 2, 3, 4} |
47 | //valid pixelFormat = pixel/color format in the set of {:N444_RGB, :S422, :N422, :N420} |
48 | |
49 | // fixed value |
50 | unsigned int rcModelSize = 8192; |
51 | |
52 | // N422/N420 operate at 2 pixels per clock |
53 | unsigned int pixelsPerClock, lstall, D, initalXmitDelay, w, s, ix, wx, p, l0, a, ax, L, |
54 | Delay, pixels; |
55 | |
56 | if (pixelFormat == dm_420) |
57 | pixelsPerClock = 2; |
58 | else if (pixelFormat == dm_n422) |
59 | pixelsPerClock = 2; |
60 | // #all other modes operate at 1 pixel per clock |
61 | else |
62 | pixelsPerClock = 1; |
63 | |
64 | //initial transmit delay as per PPS |
65 | initalXmitDelay = dml_round(a: rcModelSize / 2.0 / BPP / pixelsPerClock); |
66 | |
67 | //compute ssm delay |
68 | if (bpc == 8) |
69 | D = 81; |
70 | else if (bpc == 10) |
71 | D = 89; |
72 | else |
73 | D = 113; |
74 | |
75 | //divide by pixel per cycle to compute slice width as seen by DSC |
76 | w = sliceWidth / pixelsPerClock; |
77 | |
78 | //422 mode has an additional cycle of delay |
79 | if (pixelFormat == dm_420 || pixelFormat == dm_444 || pixelFormat == dm_n422) |
80 | s = 0; |
81 | else |
82 | s = 1; |
83 | |
84 | //main calculation for the dscce |
85 | ix = initalXmitDelay + 45; |
86 | wx = (w + 2) / 3; |
87 | p = 3 * wx - w; |
88 | l0 = ix / w; |
89 | a = ix + p * l0; |
90 | ax = (a + 2) / 3 + D + 6 + 1; |
91 | L = (ax + wx - 1) / wx; |
92 | if ((ix % w) == 0 && p != 0) |
93 | lstall = 1; |
94 | else |
95 | lstall = 0; |
96 | Delay = L * wx * (numSlices - 1) + ax + s + lstall + 22; |
97 | |
98 | //dsc processes 3 pixel containers per cycle and a container can contain 1 or 2 pixels |
99 | pixels = Delay * 3 * pixelsPerClock; |
100 | |
101 | #ifdef __DML_VBA_DEBUG__ |
102 | dml_print("DML::%s: bpc: %d\n" , __func__, bpc); |
103 | dml_print("DML::%s: BPP: %f\n" , __func__, BPP); |
104 | dml_print("DML::%s: sliceWidth: %d\n" , __func__, sliceWidth); |
105 | dml_print("DML::%s: numSlices: %d\n" , __func__, numSlices); |
106 | dml_print("DML::%s: pixelFormat: %d\n" , __func__, pixelFormat); |
107 | dml_print("DML::%s: Output: %d\n" , __func__, Output); |
108 | dml_print("DML::%s: pixels: %d\n" , __func__, pixels); |
109 | #endif |
110 | |
111 | return pixels; |
112 | } |
113 | |
114 | unsigned int dml32_dscComputeDelay(enum output_format_class pixelFormat, enum output_encoder_class Output) |
115 | { |
116 | unsigned int Delay = 0; |
117 | |
118 | if (pixelFormat == dm_420) { |
119 | // sfr |
120 | Delay = Delay + 2; |
121 | // dsccif |
122 | Delay = Delay + 0; |
123 | // dscc - input deserializer |
124 | Delay = Delay + 3; |
125 | // dscc gets pixels every other cycle |
126 | Delay = Delay + 2; |
127 | // dscc - input cdc fifo |
128 | Delay = Delay + 12; |
129 | // dscc gets pixels every other cycle |
130 | Delay = Delay + 13; |
131 | // dscc - cdc uncertainty |
132 | Delay = Delay + 2; |
133 | // dscc - output cdc fifo |
134 | Delay = Delay + 7; |
135 | // dscc gets pixels every other cycle |
136 | Delay = Delay + 3; |
137 | // dscc - cdc uncertainty |
138 | Delay = Delay + 2; |
139 | // dscc - output serializer |
140 | Delay = Delay + 1; |
141 | // sft |
142 | Delay = Delay + 1; |
143 | } else if (pixelFormat == dm_n422 || (pixelFormat != dm_444)) { |
144 | // sfr |
145 | Delay = Delay + 2; |
146 | // dsccif |
147 | Delay = Delay + 1; |
148 | // dscc - input deserializer |
149 | Delay = Delay + 5; |
150 | // dscc - input cdc fifo |
151 | Delay = Delay + 25; |
152 | // dscc - cdc uncertainty |
153 | Delay = Delay + 2; |
154 | // dscc - output cdc fifo |
155 | Delay = Delay + 10; |
156 | // dscc - cdc uncertainty |
157 | Delay = Delay + 2; |
158 | // dscc - output serializer |
159 | Delay = Delay + 1; |
160 | // sft |
161 | Delay = Delay + 1; |
162 | } else { |
163 | // sfr |
164 | Delay = Delay + 2; |
165 | // dsccif |
166 | Delay = Delay + 0; |
167 | // dscc - input deserializer |
168 | Delay = Delay + 3; |
169 | // dscc - input cdc fifo |
170 | Delay = Delay + 12; |
171 | // dscc - cdc uncertainty |
172 | Delay = Delay + 2; |
173 | // dscc - output cdc fifo |
174 | Delay = Delay + 7; |
175 | // dscc - output serializer |
176 | Delay = Delay + 1; |
177 | // dscc - cdc uncertainty |
178 | Delay = Delay + 2; |
179 | // sft |
180 | Delay = Delay + 1; |
181 | } |
182 | |
183 | return Delay; |
184 | } |
185 | |
186 | |
187 | bool IsVertical(enum dm_rotation_angle Scan) |
188 | { |
189 | bool is_vert = false; |
190 | |
191 | if (Scan == dm_rotation_90 || Scan == dm_rotation_90m || Scan == dm_rotation_270 || Scan == dm_rotation_270m) |
192 | is_vert = true; |
193 | else |
194 | is_vert = false; |
195 | return is_vert; |
196 | } |
197 | |
198 | void dml32_CalculateSinglePipeDPPCLKAndSCLThroughput( |
199 | double HRatio, |
200 | double HRatioChroma, |
201 | double VRatio, |
202 | double VRatioChroma, |
203 | double MaxDCHUBToPSCLThroughput, |
204 | double MaxPSCLToLBThroughput, |
205 | double PixelClock, |
206 | enum source_format_class SourcePixelFormat, |
207 | unsigned int HTaps, |
208 | unsigned int HTapsChroma, |
209 | unsigned int VTaps, |
210 | unsigned int VTapsChroma, |
211 | |
212 | /* output */ |
213 | double *PSCL_THROUGHPUT, |
214 | double *PSCL_THROUGHPUT_CHROMA, |
215 | double *DPPCLKUsingSingleDPP) |
216 | { |
217 | double DPPCLKUsingSingleDPPLuma; |
218 | double DPPCLKUsingSingleDPPChroma; |
219 | |
220 | if (HRatio > 1) { |
221 | *PSCL_THROUGHPUT = dml_min(a: MaxDCHUBToPSCLThroughput, b: MaxPSCLToLBThroughput * HRatio / |
222 | dml_ceil(a: (double) HTaps / 6.0, granularity: 1.0)); |
223 | } else { |
224 | *PSCL_THROUGHPUT = dml_min(a: MaxDCHUBToPSCLThroughput, b: MaxPSCLToLBThroughput); |
225 | } |
226 | |
227 | DPPCLKUsingSingleDPPLuma = PixelClock * dml_max3(a: VTaps / 6 * dml_min(a: 1, b: HRatio), b: HRatio * VRatio / |
228 | *PSCL_THROUGHPUT, c: 1); |
229 | |
230 | if ((HTaps > 6 || VTaps > 6) && DPPCLKUsingSingleDPPLuma < 2 * PixelClock) |
231 | DPPCLKUsingSingleDPPLuma = 2 * PixelClock; |
232 | |
233 | if ((SourcePixelFormat != dm_420_8 && SourcePixelFormat != dm_420_10 && SourcePixelFormat != dm_420_12 && |
234 | SourcePixelFormat != dm_rgbe_alpha)) { |
235 | *PSCL_THROUGHPUT_CHROMA = 0; |
236 | *DPPCLKUsingSingleDPP = DPPCLKUsingSingleDPPLuma; |
237 | } else { |
238 | if (HRatioChroma > 1) { |
239 | *PSCL_THROUGHPUT_CHROMA = dml_min(a: MaxDCHUBToPSCLThroughput, b: MaxPSCLToLBThroughput * |
240 | HRatioChroma / dml_ceil(a: (double) HTapsChroma / 6.0, granularity: 1.0)); |
241 | } else { |
242 | *PSCL_THROUGHPUT_CHROMA = dml_min(a: MaxDCHUBToPSCLThroughput, b: MaxPSCLToLBThroughput); |
243 | } |
244 | DPPCLKUsingSingleDPPChroma = PixelClock * dml_max3(a: VTapsChroma / 6 * dml_min(a: 1, b: HRatioChroma), |
245 | b: HRatioChroma * VRatioChroma / *PSCL_THROUGHPUT_CHROMA, c: 1); |
246 | if ((HTapsChroma > 6 || VTapsChroma > 6) && DPPCLKUsingSingleDPPChroma < 2 * PixelClock) |
247 | DPPCLKUsingSingleDPPChroma = 2 * PixelClock; |
248 | *DPPCLKUsingSingleDPP = dml_max(a: DPPCLKUsingSingleDPPLuma, b: DPPCLKUsingSingleDPPChroma); |
249 | } |
250 | } |
251 | |
252 | void dml32_CalculateBytePerPixelAndBlockSizes( |
253 | enum source_format_class SourcePixelFormat, |
254 | enum dm_swizzle_mode SurfaceTiling, |
255 | |
256 | /* Output */ |
257 | unsigned int *BytePerPixelY, |
258 | unsigned int *BytePerPixelC, |
259 | double *BytePerPixelDETY, |
260 | double *BytePerPixelDETC, |
261 | unsigned int *BlockHeight256BytesY, |
262 | unsigned int *BlockHeight256BytesC, |
263 | unsigned int *BlockWidth256BytesY, |
264 | unsigned int *BlockWidth256BytesC, |
265 | unsigned int *MacroTileHeightY, |
266 | unsigned int *MacroTileHeightC, |
267 | unsigned int *MacroTileWidthY, |
268 | unsigned int *MacroTileWidthC) |
269 | { |
270 | if (SourcePixelFormat == dm_444_64) { |
271 | *BytePerPixelDETY = 8; |
272 | *BytePerPixelDETC = 0; |
273 | *BytePerPixelY = 8; |
274 | *BytePerPixelC = 0; |
275 | } else if (SourcePixelFormat == dm_444_32 || SourcePixelFormat == dm_rgbe) { |
276 | *BytePerPixelDETY = 4; |
277 | *BytePerPixelDETC = 0; |
278 | *BytePerPixelY = 4; |
279 | *BytePerPixelC = 0; |
280 | } else if (SourcePixelFormat == dm_444_16) { |
281 | *BytePerPixelDETY = 2; |
282 | *BytePerPixelDETC = 0; |
283 | *BytePerPixelY = 2; |
284 | *BytePerPixelC = 0; |
285 | } else if (SourcePixelFormat == dm_444_8) { |
286 | *BytePerPixelDETY = 1; |
287 | *BytePerPixelDETC = 0; |
288 | *BytePerPixelY = 1; |
289 | *BytePerPixelC = 0; |
290 | } else if (SourcePixelFormat == dm_rgbe_alpha) { |
291 | *BytePerPixelDETY = 4; |
292 | *BytePerPixelDETC = 1; |
293 | *BytePerPixelY = 4; |
294 | *BytePerPixelC = 1; |
295 | } else if (SourcePixelFormat == dm_420_8) { |
296 | *BytePerPixelDETY = 1; |
297 | *BytePerPixelDETC = 2; |
298 | *BytePerPixelY = 1; |
299 | *BytePerPixelC = 2; |
300 | } else if (SourcePixelFormat == dm_420_12) { |
301 | *BytePerPixelDETY = 2; |
302 | *BytePerPixelDETC = 4; |
303 | *BytePerPixelY = 2; |
304 | *BytePerPixelC = 4; |
305 | } else { |
306 | *BytePerPixelDETY = 4.0 / 3; |
307 | *BytePerPixelDETC = 8.0 / 3; |
308 | *BytePerPixelY = 2; |
309 | *BytePerPixelC = 4; |
310 | } |
311 | #ifdef __DML_VBA_DEBUG__ |
312 | dml_print("DML::%s: SourcePixelFormat = %d\n" , __func__, SourcePixelFormat); |
313 | dml_print("DML::%s: BytePerPixelDETY = %f\n" , __func__, *BytePerPixelDETY); |
314 | dml_print("DML::%s: BytePerPixelDETC = %f\n" , __func__, *BytePerPixelDETC); |
315 | dml_print("DML::%s: BytePerPixelY = %d\n" , __func__, *BytePerPixelY); |
316 | dml_print("DML::%s: BytePerPixelC = %d\n" , __func__, *BytePerPixelC); |
317 | #endif |
318 | if ((SourcePixelFormat == dm_444_64 || SourcePixelFormat == dm_444_32 |
319 | || SourcePixelFormat == dm_444_16 |
320 | || SourcePixelFormat == dm_444_8 |
321 | || SourcePixelFormat == dm_mono_16 |
322 | || SourcePixelFormat == dm_mono_8 |
323 | || SourcePixelFormat == dm_rgbe)) { |
324 | if (SurfaceTiling == dm_sw_linear) |
325 | *BlockHeight256BytesY = 1; |
326 | else if (SourcePixelFormat == dm_444_64) |
327 | *BlockHeight256BytesY = 4; |
328 | else if (SourcePixelFormat == dm_444_8) |
329 | *BlockHeight256BytesY = 16; |
330 | else |
331 | *BlockHeight256BytesY = 8; |
332 | |
333 | *BlockWidth256BytesY = 256U / *BytePerPixelY / *BlockHeight256BytesY; |
334 | *BlockHeight256BytesC = 0; |
335 | *BlockWidth256BytesC = 0; |
336 | } else { |
337 | if (SurfaceTiling == dm_sw_linear) { |
338 | *BlockHeight256BytesY = 1; |
339 | *BlockHeight256BytesC = 1; |
340 | } else if (SourcePixelFormat == dm_rgbe_alpha) { |
341 | *BlockHeight256BytesY = 8; |
342 | *BlockHeight256BytesC = 16; |
343 | } else if (SourcePixelFormat == dm_420_8) { |
344 | *BlockHeight256BytesY = 16; |
345 | *BlockHeight256BytesC = 8; |
346 | } else { |
347 | *BlockHeight256BytesY = 8; |
348 | *BlockHeight256BytesC = 8; |
349 | } |
350 | *BlockWidth256BytesY = 256U / *BytePerPixelY / *BlockHeight256BytesY; |
351 | *BlockWidth256BytesC = 256U / *BytePerPixelC / *BlockHeight256BytesC; |
352 | } |
353 | #ifdef __DML_VBA_DEBUG__ |
354 | dml_print("DML::%s: BlockWidth256BytesY = %d\n" , __func__, *BlockWidth256BytesY); |
355 | dml_print("DML::%s: BlockHeight256BytesY = %d\n" , __func__, *BlockHeight256BytesY); |
356 | dml_print("DML::%s: BlockWidth256BytesC = %d\n" , __func__, *BlockWidth256BytesC); |
357 | dml_print("DML::%s: BlockHeight256BytesC = %d\n" , __func__, *BlockHeight256BytesC); |
358 | #endif |
359 | |
360 | if (SurfaceTiling == dm_sw_linear) { |
361 | *MacroTileHeightY = *BlockHeight256BytesY; |
362 | *MacroTileWidthY = 256 / *BytePerPixelY / *MacroTileHeightY; |
363 | *MacroTileHeightC = *BlockHeight256BytesC; |
364 | if (*MacroTileHeightC == 0) |
365 | *MacroTileWidthC = 0; |
366 | else |
367 | *MacroTileWidthC = 256 / *BytePerPixelC / *MacroTileHeightC; |
368 | } else if (SurfaceTiling == dm_sw_64kb_d || SurfaceTiling == dm_sw_64kb_d_t || |
369 | SurfaceTiling == dm_sw_64kb_d_x || SurfaceTiling == dm_sw_64kb_r_x) { |
370 | *MacroTileHeightY = 16 * *BlockHeight256BytesY; |
371 | *MacroTileWidthY = 65536 / *BytePerPixelY / *MacroTileHeightY; |
372 | *MacroTileHeightC = 16 * *BlockHeight256BytesC; |
373 | if (*MacroTileHeightC == 0) |
374 | *MacroTileWidthC = 0; |
375 | else |
376 | *MacroTileWidthC = 65536 / *BytePerPixelC / *MacroTileHeightC; |
377 | } else { |
378 | *MacroTileHeightY = 32 * *BlockHeight256BytesY; |
379 | *MacroTileWidthY = 65536 * 4 / *BytePerPixelY / *MacroTileHeightY; |
380 | *MacroTileHeightC = 32 * *BlockHeight256BytesC; |
381 | if (*MacroTileHeightC == 0) |
382 | *MacroTileWidthC = 0; |
383 | else |
384 | *MacroTileWidthC = 65536 * 4 / *BytePerPixelC / *MacroTileHeightC; |
385 | } |
386 | |
387 | #ifdef __DML_VBA_DEBUG__ |
388 | dml_print("DML::%s: MacroTileWidthY = %d\n" , __func__, *MacroTileWidthY); |
389 | dml_print("DML::%s: MacroTileHeightY = %d\n" , __func__, *MacroTileHeightY); |
390 | dml_print("DML::%s: MacroTileWidthC = %d\n" , __func__, *MacroTileWidthC); |
391 | dml_print("DML::%s: MacroTileHeightC = %d\n" , __func__, *MacroTileHeightC); |
392 | #endif |
393 | } // CalculateBytePerPixelAndBlockSizes |
394 | |
395 | void dml32_CalculateSwathAndDETConfiguration( |
396 | unsigned int DETSizeOverride[], |
397 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange[], |
398 | unsigned int ConfigReturnBufferSizeInKByte, |
399 | unsigned int MaxTotalDETInKByte, |
400 | unsigned int MinCompressedBufferSizeInKByte, |
401 | double ForceSingleDPP, |
402 | unsigned int NumberOfActiveSurfaces, |
403 | unsigned int nomDETInKByte, |
404 | enum unbounded_requesting_policy UseUnboundedRequestingFinal, |
405 | bool DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment, |
406 | unsigned int PixelChunkSizeKBytes, |
407 | unsigned int ROBSizeKBytes, |
408 | unsigned int CompressedBufferSegmentSizeInkByteFinal, |
409 | enum output_encoder_class Output[], |
410 | double ReadBandwidthLuma[], |
411 | double ReadBandwidthChroma[], |
412 | double MaximumSwathWidthLuma[], |
413 | double MaximumSwathWidthChroma[], |
414 | enum dm_rotation_angle SourceRotation[], |
415 | bool ViewportStationary[], |
416 | enum source_format_class SourcePixelFormat[], |
417 | enum dm_swizzle_mode SurfaceTiling[], |
418 | unsigned int ViewportWidth[], |
419 | unsigned int ViewportHeight[], |
420 | unsigned int ViewportXStart[], |
421 | unsigned int ViewportYStart[], |
422 | unsigned int ViewportXStartC[], |
423 | unsigned int ViewportYStartC[], |
424 | unsigned int SurfaceWidthY[], |
425 | unsigned int SurfaceWidthC[], |
426 | unsigned int SurfaceHeightY[], |
427 | unsigned int SurfaceHeightC[], |
428 | unsigned int Read256BytesBlockHeightY[], |
429 | unsigned int Read256BytesBlockHeightC[], |
430 | unsigned int Read256BytesBlockWidthY[], |
431 | unsigned int Read256BytesBlockWidthC[], |
432 | enum odm_combine_mode ODMMode[], |
433 | unsigned int BlendingAndTiming[], |
434 | unsigned int BytePerPixY[], |
435 | unsigned int BytePerPixC[], |
436 | double BytePerPixDETY[], |
437 | double BytePerPixDETC[], |
438 | unsigned int HActive[], |
439 | double HRatio[], |
440 | double HRatioChroma[], |
441 | unsigned int DPPPerSurface[], |
442 | |
443 | /* Output */ |
444 | unsigned int swath_width_luma_ub[], |
445 | unsigned int swath_width_chroma_ub[], |
446 | double SwathWidth[], |
447 | double SwathWidthChroma[], |
448 | unsigned int SwathHeightY[], |
449 | unsigned int SwathHeightC[], |
450 | unsigned int DETBufferSizeInKByte[], |
451 | unsigned int DETBufferSizeY[], |
452 | unsigned int DETBufferSizeC[], |
453 | bool *UnboundedRequestEnabled, |
454 | unsigned int *CompressedBufferSizeInkByte, |
455 | unsigned int *CompBufReservedSpaceKBytes, |
456 | bool *CompBufReservedSpaceNeedAdjustment, |
457 | bool ViewportSizeSupportPerSurface[], |
458 | bool *ViewportSizeSupport) |
459 | { |
460 | unsigned int MaximumSwathHeightY[DC__NUM_DPP__MAX]; |
461 | unsigned int MaximumSwathHeightC[DC__NUM_DPP__MAX]; |
462 | unsigned int RoundedUpMaxSwathSizeBytesY[DC__NUM_DPP__MAX]; |
463 | unsigned int RoundedUpMaxSwathSizeBytesC[DC__NUM_DPP__MAX]; |
464 | unsigned int RoundedUpSwathSizeBytesY; |
465 | unsigned int RoundedUpSwathSizeBytesC; |
466 | double SwathWidthdoubleDPP[DC__NUM_DPP__MAX]; |
467 | double SwathWidthdoubleDPPChroma[DC__NUM_DPP__MAX]; |
468 | unsigned int k; |
469 | unsigned int TotalActiveDPP = 0; |
470 | bool NoChromaSurfaces = true; |
471 | unsigned int DETBufferSizeInKByteForSwathCalculation; |
472 | |
473 | #ifdef __DML_VBA_DEBUG__ |
474 | dml_print("DML::%s: ForceSingleDPP = %d\n" , __func__, ForceSingleDPP); |
475 | dml_print("DML::%s: ROBSizeKBytes = %d\n" , __func__, ROBSizeKBytes); |
476 | dml_print("DML::%s: PixelChunkSizeKBytes = %d\n" , __func__, PixelChunkSizeKBytes); |
477 | #endif |
478 | dml32_CalculateSwathWidth(ForceSingleDPP, |
479 | NumberOfActiveSurfaces, |
480 | SourcePixelFormat, |
481 | SourceScan: SourceRotation, |
482 | ViewportStationary, |
483 | ViewportWidth, |
484 | ViewportHeight, |
485 | ViewportXStart, |
486 | ViewportYStart, |
487 | ViewportXStartC, |
488 | ViewportYStartC, |
489 | SurfaceWidthY, |
490 | SurfaceWidthC, |
491 | SurfaceHeightY, |
492 | SurfaceHeightC, |
493 | ODMMode, |
494 | BytePerPixY, |
495 | BytePerPixC, |
496 | Read256BytesBlockHeightY, |
497 | Read256BytesBlockHeightC, |
498 | Read256BytesBlockWidthY, |
499 | Read256BytesBlockWidthC, |
500 | BlendingAndTiming, |
501 | HActive, |
502 | HRatio, |
503 | DPPPerSurface, |
504 | |
505 | /* Output */ |
506 | SwathWidthdoubleDPPY: SwathWidthdoubleDPP, |
507 | SwathWidthdoubleDPPC: SwathWidthdoubleDPPChroma, |
508 | SwathWidthY: SwathWidth, |
509 | SwathWidthC: SwathWidthChroma, |
510 | MaximumSwathHeightY, |
511 | MaximumSwathHeightC, |
512 | swath_width_luma_ub, |
513 | swath_width_chroma_ub); |
514 | |
515 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
516 | RoundedUpMaxSwathSizeBytesY[k] = swath_width_luma_ub[k] * BytePerPixDETY[k] * MaximumSwathHeightY[k]; |
517 | RoundedUpMaxSwathSizeBytesC[k] = swath_width_chroma_ub[k] * BytePerPixDETC[k] * MaximumSwathHeightC[k]; |
518 | #ifdef __DML_VBA_DEBUG__ |
519 | dml_print("DML::%s: k=%0d DPPPerSurface = %d\n" , __func__, k, DPPPerSurface[k]); |
520 | dml_print("DML::%s: k=%0d swath_width_luma_ub = %d\n" , __func__, k, swath_width_luma_ub[k]); |
521 | dml_print("DML::%s: k=%0d BytePerPixDETY = %f\n" , __func__, k, BytePerPixDETY[k]); |
522 | dml_print("DML::%s: k=%0d MaximumSwathHeightY = %d\n" , __func__, k, MaximumSwathHeightY[k]); |
523 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesY = %d\n" , __func__, k, |
524 | RoundedUpMaxSwathSizeBytesY[k]); |
525 | dml_print("DML::%s: k=%0d swath_width_chroma_ub = %d\n" , __func__, k, swath_width_chroma_ub[k]); |
526 | dml_print("DML::%s: k=%0d BytePerPixDETC = %f\n" , __func__, k, BytePerPixDETC[k]); |
527 | dml_print("DML::%s: k=%0d MaximumSwathHeightC = %d\n" , __func__, k, MaximumSwathHeightC[k]); |
528 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesC = %d\n" , __func__, k, |
529 | RoundedUpMaxSwathSizeBytesC[k]); |
530 | #endif |
531 | |
532 | if (SourcePixelFormat[k] == dm_420_10) { |
533 | RoundedUpMaxSwathSizeBytesY[k] = dml_ceil(a: (unsigned int) RoundedUpMaxSwathSizeBytesY[k], granularity: 256); |
534 | RoundedUpMaxSwathSizeBytesC[k] = dml_ceil(a: (unsigned int) RoundedUpMaxSwathSizeBytesC[k], granularity: 256); |
535 | } |
536 | } |
537 | |
538 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
539 | TotalActiveDPP = TotalActiveDPP + (ForceSingleDPP ? 1 : DPPPerSurface[k]); |
540 | if (SourcePixelFormat[k] == dm_420_8 || SourcePixelFormat[k] == dm_420_10 || |
541 | SourcePixelFormat[k] == dm_420_12 || SourcePixelFormat[k] == dm_rgbe_alpha) { |
542 | NoChromaSurfaces = false; |
543 | } |
544 | } |
545 | |
546 | // By default, just set the reserved space to 2 pixel chunks size |
547 | *CompBufReservedSpaceKBytes = PixelChunkSizeKBytes * 2; |
548 | |
549 | // if unbounded req is enabled, program reserved space such that the ROB will not hold more than 8 swaths worth of data |
550 | // - assume worst-case compression rate of 4. [ROB size - 8 * swath_size / max_compression ratio] |
551 | // - assume for "narrow" vp case in which the ROB can fit 8 swaths, the DET should be big enough to do full size req |
552 | *CompBufReservedSpaceNeedAdjustment = ((int) ROBSizeKBytes - (int) *CompBufReservedSpaceKBytes) > (int) (RoundedUpMaxSwathSizeBytesY[0]/512); |
553 | |
554 | if (*CompBufReservedSpaceNeedAdjustment == 1) { |
555 | *CompBufReservedSpaceKBytes = ROBSizeKBytes - RoundedUpMaxSwathSizeBytesY[0]/512; |
556 | } |
557 | |
558 | #ifdef __DML_VBA_DEBUG__ |
559 | dml_print("DML::%s: CompBufReservedSpaceKBytes = %d\n" , __func__, *CompBufReservedSpaceKBytes); |
560 | dml_print("DML::%s: CompBufReservedSpaceNeedAdjustment = %d\n" , __func__, *CompBufReservedSpaceNeedAdjustment); |
561 | #endif |
562 | |
563 | *UnboundedRequestEnabled = dml32_UnboundedRequest(UseUnboundedRequestingFinal, TotalNumberOfActiveDPP: TotalActiveDPP, NoChroma: NoChromaSurfaces, Output: Output[0], SurfaceTiling: SurfaceTiling[0], CompBufReservedSpaceNeedAdjustment: *CompBufReservedSpaceNeedAdjustment, DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment); |
564 | |
565 | dml32_CalculateDETBufferSize(DETSizeOverride, |
566 | UseMALLForPStateChange, |
567 | ForceSingleDPP, |
568 | NumberOfActiveSurfaces, |
569 | UnboundedRequestEnabled: *UnboundedRequestEnabled, |
570 | nomDETInKByte, |
571 | MaxTotalDETInKByte, |
572 | ConfigReturnBufferSizeInKByte, |
573 | MinCompressedBufferSizeInKByte, |
574 | CompressedBufferSegmentSizeInkByteFinal, |
575 | SourcePixelFormat, |
576 | ReadBandwidthLuma, |
577 | ReadBandwidthChroma, |
578 | RoundedUpMaxSwathSizeBytesY, |
579 | RoundedUpMaxSwathSizeBytesC, |
580 | DPPPerSurface, |
581 | |
582 | /* Output */ |
583 | DETBufferSizeInKByte, // per hubp pipe |
584 | CompressedBufferSizeInkByte); |
585 | |
586 | #ifdef __DML_VBA_DEBUG__ |
587 | dml_print("DML::%s: TotalActiveDPP = %d\n" , __func__, TotalActiveDPP); |
588 | dml_print("DML::%s: nomDETInKByte = %d\n" , __func__, nomDETInKByte); |
589 | dml_print("DML::%s: ConfigReturnBufferSizeInKByte = %d\n" , __func__, ConfigReturnBufferSizeInKByte); |
590 | dml_print("DML::%s: UseUnboundedRequestingFinal = %d\n" , __func__, UseUnboundedRequestingFinal); |
591 | dml_print("DML::%s: UnboundedRequestEnabled = %d\n" , __func__, *UnboundedRequestEnabled); |
592 | dml_print("DML::%s: CompressedBufferSizeInkByte = %d\n" , __func__, *CompressedBufferSizeInkByte); |
593 | #endif |
594 | |
595 | *ViewportSizeSupport = true; |
596 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
597 | |
598 | DETBufferSizeInKByteForSwathCalculation = (UseMALLForPStateChange[k] == |
599 | dm_use_mall_pstate_change_phantom_pipe ? 1024 : DETBufferSizeInKByte[k]); |
600 | #ifdef __DML_VBA_DEBUG__ |
601 | dml_print("DML::%s: k=%0d DETBufferSizeInKByteForSwathCalculation = %d\n" , __func__, k, |
602 | DETBufferSizeInKByteForSwathCalculation); |
603 | #endif |
604 | |
605 | if (RoundedUpMaxSwathSizeBytesY[k] + RoundedUpMaxSwathSizeBytesC[k] <= |
606 | DETBufferSizeInKByteForSwathCalculation * 1024 / 2) { |
607 | SwathHeightY[k] = MaximumSwathHeightY[k]; |
608 | SwathHeightC[k] = MaximumSwathHeightC[k]; |
609 | RoundedUpSwathSizeBytesY = RoundedUpMaxSwathSizeBytesY[k]; |
610 | RoundedUpSwathSizeBytesC = RoundedUpMaxSwathSizeBytesC[k]; |
611 | } else if (RoundedUpMaxSwathSizeBytesY[k] >= 1.5 * RoundedUpMaxSwathSizeBytesC[k] && |
612 | RoundedUpMaxSwathSizeBytesY[k] / 2 + RoundedUpMaxSwathSizeBytesC[k] <= |
613 | DETBufferSizeInKByteForSwathCalculation * 1024 / 2) { |
614 | SwathHeightY[k] = MaximumSwathHeightY[k] / 2; |
615 | SwathHeightC[k] = MaximumSwathHeightC[k]; |
616 | RoundedUpSwathSizeBytesY = RoundedUpMaxSwathSizeBytesY[k] / 2; |
617 | RoundedUpSwathSizeBytesC = RoundedUpMaxSwathSizeBytesC[k]; |
618 | } else if (RoundedUpMaxSwathSizeBytesY[k] < 1.5 * RoundedUpMaxSwathSizeBytesC[k] && |
619 | RoundedUpMaxSwathSizeBytesY[k] + RoundedUpMaxSwathSizeBytesC[k] / 2 <= |
620 | DETBufferSizeInKByteForSwathCalculation * 1024 / 2) { |
621 | SwathHeightY[k] = MaximumSwathHeightY[k]; |
622 | SwathHeightC[k] = MaximumSwathHeightC[k] / 2; |
623 | RoundedUpSwathSizeBytesY = RoundedUpMaxSwathSizeBytesY[k]; |
624 | RoundedUpSwathSizeBytesC = RoundedUpMaxSwathSizeBytesC[k] / 2; |
625 | } else { |
626 | SwathHeightY[k] = MaximumSwathHeightY[k] / 2; |
627 | SwathHeightC[k] = MaximumSwathHeightC[k] / 2; |
628 | RoundedUpSwathSizeBytesY = RoundedUpMaxSwathSizeBytesY[k] / 2; |
629 | RoundedUpSwathSizeBytesC = RoundedUpMaxSwathSizeBytesC[k] / 2; |
630 | } |
631 | |
632 | if ((RoundedUpMaxSwathSizeBytesY[k] / 2 + RoundedUpMaxSwathSizeBytesC[k] / 2 > |
633 | DETBufferSizeInKByteForSwathCalculation * 1024 / 2) |
634 | || SwathWidth[k] > MaximumSwathWidthLuma[k] || (SwathHeightC[k] > 0 && |
635 | SwathWidthChroma[k] > MaximumSwathWidthChroma[k])) { |
636 | *ViewportSizeSupport = false; |
637 | ViewportSizeSupportPerSurface[k] = false; |
638 | } else { |
639 | ViewportSizeSupportPerSurface[k] = true; |
640 | } |
641 | |
642 | if (SwathHeightC[k] == 0) { |
643 | #ifdef __DML_VBA_DEBUG__ |
644 | dml_print("DML::%s: k=%0d All DET for plane0\n" , __func__, k); |
645 | #endif |
646 | DETBufferSizeY[k] = DETBufferSizeInKByte[k] * 1024; |
647 | DETBufferSizeC[k] = 0; |
648 | } else if (RoundedUpSwathSizeBytesY <= 1.5 * RoundedUpSwathSizeBytesC) { |
649 | #ifdef __DML_VBA_DEBUG__ |
650 | dml_print("DML::%s: k=%0d Half DET for plane0, half for plane1\n" , __func__, k); |
651 | #endif |
652 | DETBufferSizeY[k] = DETBufferSizeInKByte[k] * 1024 / 2; |
653 | DETBufferSizeC[k] = DETBufferSizeInKByte[k] * 1024 / 2; |
654 | } else { |
655 | #ifdef __DML_VBA_DEBUG__ |
656 | dml_print("DML::%s: k=%0d 2/3 DET for plane0, 1/3 for plane1\n" , __func__, k); |
657 | #endif |
658 | DETBufferSizeY[k] = dml_floor(a: DETBufferSizeInKByte[k] * 1024 * 2 / 3, granularity: 1024); |
659 | DETBufferSizeC[k] = DETBufferSizeInKByte[k] * 1024 - DETBufferSizeY[k]; |
660 | } |
661 | |
662 | #ifdef __DML_VBA_DEBUG__ |
663 | dml_print("DML::%s: k=%0d SwathHeightY = %d\n" , __func__, k, SwathHeightY[k]); |
664 | dml_print("DML::%s: k=%0d SwathHeightC = %d\n" , __func__, k, SwathHeightC[k]); |
665 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesY = %d\n" , __func__, |
666 | k, RoundedUpMaxSwathSizeBytesY[k]); |
667 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesC = %d\n" , __func__, |
668 | k, RoundedUpMaxSwathSizeBytesC[k]); |
669 | dml_print("DML::%s: k=%0d RoundedUpSwathSizeBytesY = %d\n" , __func__, k, RoundedUpSwathSizeBytesY); |
670 | dml_print("DML::%s: k=%0d RoundedUpSwathSizeBytesC = %d\n" , __func__, k, RoundedUpSwathSizeBytesC); |
671 | dml_print("DML::%s: k=%0d DETBufferSizeInKByte = %d\n" , __func__, k, DETBufferSizeInKByte[k]); |
672 | dml_print("DML::%s: k=%0d DETBufferSizeY = %d\n" , __func__, k, DETBufferSizeY[k]); |
673 | dml_print("DML::%s: k=%0d DETBufferSizeC = %d\n" , __func__, k, DETBufferSizeC[k]); |
674 | dml_print("DML::%s: k=%0d ViewportSizeSupportPerSurface = %d\n" , __func__, k, |
675 | ViewportSizeSupportPerSurface[k]); |
676 | #endif |
677 | |
678 | } |
679 | } // CalculateSwathAndDETConfiguration |
680 | |
681 | void dml32_CalculateSwathWidth( |
682 | bool ForceSingleDPP, |
683 | unsigned int NumberOfActiveSurfaces, |
684 | enum source_format_class SourcePixelFormat[], |
685 | enum dm_rotation_angle SourceRotation[], |
686 | bool ViewportStationary[], |
687 | unsigned int ViewportWidth[], |
688 | unsigned int ViewportHeight[], |
689 | unsigned int ViewportXStart[], |
690 | unsigned int ViewportYStart[], |
691 | unsigned int ViewportXStartC[], |
692 | unsigned int ViewportYStartC[], |
693 | unsigned int SurfaceWidthY[], |
694 | unsigned int SurfaceWidthC[], |
695 | unsigned int SurfaceHeightY[], |
696 | unsigned int SurfaceHeightC[], |
697 | enum odm_combine_mode ODMMode[], |
698 | unsigned int BytePerPixY[], |
699 | unsigned int BytePerPixC[], |
700 | unsigned int Read256BytesBlockHeightY[], |
701 | unsigned int Read256BytesBlockHeightC[], |
702 | unsigned int Read256BytesBlockWidthY[], |
703 | unsigned int Read256BytesBlockWidthC[], |
704 | unsigned int BlendingAndTiming[], |
705 | unsigned int HActive[], |
706 | double HRatio[], |
707 | unsigned int DPPPerSurface[], |
708 | |
709 | /* Output */ |
710 | double SwathWidthdoubleDPPY[], |
711 | double SwathWidthdoubleDPPC[], |
712 | double SwathWidthY[], // per-pipe |
713 | double SwathWidthC[], // per-pipe |
714 | unsigned int MaximumSwathHeightY[], |
715 | unsigned int MaximumSwathHeightC[], |
716 | unsigned int swath_width_luma_ub[], // per-pipe |
717 | unsigned int swath_width_chroma_ub[]) // per-pipe |
718 | { |
719 | unsigned int k, j; |
720 | enum odm_combine_mode MainSurfaceODMMode; |
721 | |
722 | unsigned int surface_width_ub_l; |
723 | unsigned int surface_height_ub_l; |
724 | unsigned int surface_width_ub_c = 0; |
725 | unsigned int surface_height_ub_c = 0; |
726 | |
727 | #ifdef __DML_VBA_DEBUG__ |
728 | dml_print("DML::%s: ForceSingleDPP = %d\n" , __func__, ForceSingleDPP); |
729 | dml_print("DML::%s: NumberOfActiveSurfaces = %d\n" , __func__, NumberOfActiveSurfaces); |
730 | #endif |
731 | |
732 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
733 | if (!IsVertical(Scan: SourceRotation[k])) |
734 | SwathWidthdoubleDPPY[k] = ViewportWidth[k]; |
735 | else |
736 | SwathWidthdoubleDPPY[k] = ViewportHeight[k]; |
737 | |
738 | #ifdef __DML_VBA_DEBUG__ |
739 | dml_print("DML::%s: k=%d ViewportWidth=%d\n" , __func__, k, ViewportWidth[k]); |
740 | dml_print("DML::%s: k=%d ViewportHeight=%d\n" , __func__, k, ViewportHeight[k]); |
741 | #endif |
742 | |
743 | MainSurfaceODMMode = ODMMode[k]; |
744 | for (j = 0; j < NumberOfActiveSurfaces; ++j) { |
745 | if (BlendingAndTiming[k] == j) |
746 | MainSurfaceODMMode = ODMMode[j]; |
747 | } |
748 | |
749 | if (ForceSingleDPP) { |
750 | SwathWidthY[k] = SwathWidthdoubleDPPY[k]; |
751 | } else { |
752 | if (MainSurfaceODMMode == dm_odm_combine_mode_4to1) { |
753 | SwathWidthY[k] = dml_min(a: SwathWidthdoubleDPPY[k], |
754 | b: dml_round(a: HActive[k] / 4.0 * HRatio[k])); |
755 | } else if (MainSurfaceODMMode == dm_odm_combine_mode_2to1) { |
756 | SwathWidthY[k] = dml_min(a: SwathWidthdoubleDPPY[k], |
757 | b: dml_round(a: HActive[k] / 2.0 * HRatio[k])); |
758 | } else if (DPPPerSurface[k] == 2) { |
759 | SwathWidthY[k] = SwathWidthdoubleDPPY[k] / 2; |
760 | } else { |
761 | SwathWidthY[k] = SwathWidthdoubleDPPY[k]; |
762 | } |
763 | } |
764 | |
765 | #ifdef __DML_VBA_DEBUG__ |
766 | dml_print("DML::%s: k=%d HActive=%d\n" , __func__, k, HActive[k]); |
767 | dml_print("DML::%s: k=%d HRatio=%f\n" , __func__, k, HRatio[k]); |
768 | dml_print("DML::%s: k=%d MainSurfaceODMMode=%d\n" , __func__, k, MainSurfaceODMMode); |
769 | dml_print("DML::%s: k=%d SwathWidthdoubleDPPY=%d\n" , __func__, k, SwathWidthdoubleDPPY[k]); |
770 | dml_print("DML::%s: k=%d SwathWidthY=%d\n" , __func__, k, SwathWidthY[k]); |
771 | #endif |
772 | |
773 | if (SourcePixelFormat[k] == dm_420_8 || SourcePixelFormat[k] == dm_420_10 || |
774 | SourcePixelFormat[k] == dm_420_12) { |
775 | SwathWidthC[k] = SwathWidthY[k] / 2; |
776 | SwathWidthdoubleDPPC[k] = SwathWidthdoubleDPPY[k] / 2; |
777 | } else { |
778 | SwathWidthC[k] = SwathWidthY[k]; |
779 | SwathWidthdoubleDPPC[k] = SwathWidthdoubleDPPY[k]; |
780 | } |
781 | |
782 | if (ForceSingleDPP == true) { |
783 | SwathWidthY[k] = SwathWidthdoubleDPPY[k]; |
784 | SwathWidthC[k] = SwathWidthdoubleDPPC[k]; |
785 | } |
786 | |
787 | surface_width_ub_l = dml_ceil(a: SurfaceWidthY[k], granularity: Read256BytesBlockWidthY[k]); |
788 | surface_height_ub_l = dml_ceil(a: SurfaceHeightY[k], granularity: Read256BytesBlockHeightY[k]); |
789 | |
790 | if (!IsVertical(Scan: SourceRotation[k])) { |
791 | MaximumSwathHeightY[k] = Read256BytesBlockHeightY[k]; |
792 | MaximumSwathHeightC[k] = Read256BytesBlockHeightC[k]; |
793 | if (ViewportStationary[k] && DPPPerSurface[k] == 1) { |
794 | swath_width_luma_ub[k] = dml_min(a: surface_width_ub_l, |
795 | b: dml_floor(a: ViewportXStart[k] + |
796 | SwathWidthY[k] + |
797 | Read256BytesBlockWidthY[k] - 1, |
798 | granularity: Read256BytesBlockWidthY[k]) - |
799 | dml_floor(a: ViewportXStart[k], |
800 | granularity: Read256BytesBlockWidthY[k])); |
801 | } else { |
802 | swath_width_luma_ub[k] = dml_min(a: surface_width_ub_l, |
803 | b: dml_ceil(a: SwathWidthY[k] - 1, |
804 | granularity: Read256BytesBlockWidthY[k]) + |
805 | Read256BytesBlockWidthY[k]); |
806 | } |
807 | if (BytePerPixC[k] > 0) { |
808 | surface_width_ub_c = dml_ceil(a: SurfaceWidthC[k], granularity: Read256BytesBlockWidthC[k]); |
809 | if (ViewportStationary[k] && DPPPerSurface[k] == 1) { |
810 | swath_width_chroma_ub[k] = dml_min(a: surface_width_ub_c, |
811 | b: dml_floor(a: ViewportXStartC[k] + SwathWidthC[k] + |
812 | Read256BytesBlockWidthC[k] - 1, |
813 | granularity: Read256BytesBlockWidthC[k]) - |
814 | dml_floor(a: ViewportXStartC[k], |
815 | granularity: Read256BytesBlockWidthC[k])); |
816 | } else { |
817 | swath_width_chroma_ub[k] = dml_min(a: surface_width_ub_c, |
818 | b: dml_ceil(a: SwathWidthC[k] - 1, |
819 | granularity: Read256BytesBlockWidthC[k]) + |
820 | Read256BytesBlockWidthC[k]); |
821 | } |
822 | } else { |
823 | swath_width_chroma_ub[k] = 0; |
824 | } |
825 | } else { |
826 | MaximumSwathHeightY[k] = Read256BytesBlockWidthY[k]; |
827 | MaximumSwathHeightC[k] = Read256BytesBlockWidthC[k]; |
828 | |
829 | if (ViewportStationary[k] && DPPPerSurface[k] == 1) { |
830 | swath_width_luma_ub[k] = dml_min(a: surface_height_ub_l, b: dml_floor(a: ViewportYStart[k] + |
831 | SwathWidthY[k] + Read256BytesBlockHeightY[k] - 1, |
832 | granularity: Read256BytesBlockHeightY[k]) - |
833 | dml_floor(a: ViewportYStart[k], granularity: Read256BytesBlockHeightY[k])); |
834 | } else { |
835 | swath_width_luma_ub[k] = dml_min(a: surface_height_ub_l, b: dml_ceil(a: SwathWidthY[k] - 1, |
836 | granularity: Read256BytesBlockHeightY[k]) + Read256BytesBlockHeightY[k]); |
837 | } |
838 | if (BytePerPixC[k] > 0) { |
839 | surface_height_ub_c = dml_ceil(a: SurfaceHeightC[k], granularity: Read256BytesBlockHeightC[k]); |
840 | if (ViewportStationary[k] && DPPPerSurface[k] == 1) { |
841 | swath_width_chroma_ub[k] = dml_min(a: surface_height_ub_c, |
842 | b: dml_floor(a: ViewportYStartC[k] + SwathWidthC[k] + |
843 | Read256BytesBlockHeightC[k] - 1, |
844 | granularity: Read256BytesBlockHeightC[k]) - |
845 | dml_floor(a: ViewportYStartC[k], |
846 | granularity: Read256BytesBlockHeightC[k])); |
847 | } else { |
848 | swath_width_chroma_ub[k] = dml_min(a: surface_height_ub_c, |
849 | b: dml_ceil(a: SwathWidthC[k] - 1, granularity: Read256BytesBlockHeightC[k]) + |
850 | Read256BytesBlockHeightC[k]); |
851 | } |
852 | } else { |
853 | swath_width_chroma_ub[k] = 0; |
854 | } |
855 | } |
856 | |
857 | #ifdef __DML_VBA_DEBUG__ |
858 | dml_print("DML::%s: k=%d surface_width_ub_l=%0d\n" , __func__, k, surface_width_ub_l); |
859 | dml_print("DML::%s: k=%d surface_height_ub_l=%0d\n" , __func__, k, surface_height_ub_l); |
860 | dml_print("DML::%s: k=%d surface_width_ub_c=%0d\n" , __func__, k, surface_width_ub_c); |
861 | dml_print("DML::%s: k=%d surface_height_ub_c=%0d\n" , __func__, k, surface_height_ub_c); |
862 | dml_print("DML::%s: k=%d Read256BytesBlockWidthY=%0d\n" , __func__, k, Read256BytesBlockWidthY[k]); |
863 | dml_print("DML::%s: k=%d Read256BytesBlockHeightY=%0d\n" , __func__, k, Read256BytesBlockHeightY[k]); |
864 | dml_print("DML::%s: k=%d Read256BytesBlockWidthC=%0d\n" , __func__, k, Read256BytesBlockWidthC[k]); |
865 | dml_print("DML::%s: k=%d Read256BytesBlockHeightC=%0d\n" , __func__, k, Read256BytesBlockHeightC[k]); |
866 | dml_print("DML::%s: k=%d ViewportStationary=%0d\n" , __func__, k, ViewportStationary[k]); |
867 | dml_print("DML::%s: k=%d DPPPerSurface=%0d\n" , __func__, k, DPPPerSurface[k]); |
868 | dml_print("DML::%s: k=%d swath_width_luma_ub=%0d\n" , __func__, k, swath_width_luma_ub[k]); |
869 | dml_print("DML::%s: k=%d swath_width_chroma_ub=%0d\n" , __func__, k, swath_width_chroma_ub[k]); |
870 | dml_print("DML::%s: k=%d MaximumSwathHeightY=%0d\n" , __func__, k, MaximumSwathHeightY[k]); |
871 | dml_print("DML::%s: k=%d MaximumSwathHeightC=%0d\n" , __func__, k, MaximumSwathHeightC[k]); |
872 | #endif |
873 | |
874 | } |
875 | } // CalculateSwathWidth |
876 | |
877 | bool dml32_UnboundedRequest(enum unbounded_requesting_policy UseUnboundedRequestingFinal, |
878 | unsigned int TotalNumberOfActiveDPP, |
879 | bool NoChroma, |
880 | enum output_encoder_class Output, |
881 | enum dm_swizzle_mode SurfaceTiling, |
882 | bool CompBufReservedSpaceNeedAdjustment, |
883 | bool DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment) |
884 | { |
885 | bool ret_val = false; |
886 | |
887 | ret_val = (UseUnboundedRequestingFinal != dm_unbounded_requesting_disable && |
888 | TotalNumberOfActiveDPP == 1 && NoChroma); |
889 | if (UseUnboundedRequestingFinal == dm_unbounded_requesting_edp_only && Output != dm_edp) |
890 | ret_val = false; |
891 | |
892 | if (SurfaceTiling == dm_sw_linear) |
893 | ret_val = false; |
894 | |
895 | if (CompBufReservedSpaceNeedAdjustment == 1 && DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment) |
896 | ret_val = false; |
897 | |
898 | #ifdef __DML_VBA_DEBUG__ |
899 | dml_print("DML::%s: CompBufReservedSpaceNeedAdjustment = %d\n" , __func__, CompBufReservedSpaceNeedAdjustment); |
900 | dml_print("DML::%s: DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment = %d\n" , __func__, DisableUnboundRequestIfCompBufReservedSpaceNeedAdjustment); |
901 | dml_print("DML::%s: ret_val = %d\n" , __func__, ret_val); |
902 | #endif |
903 | |
904 | return (ret_val); |
905 | } |
906 | |
907 | void dml32_CalculateDETBufferSize( |
908 | unsigned int DETSizeOverride[], |
909 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange[], |
910 | bool ForceSingleDPP, |
911 | unsigned int NumberOfActiveSurfaces, |
912 | bool UnboundedRequestEnabled, |
913 | unsigned int nomDETInKByte, |
914 | unsigned int MaxTotalDETInKByte, |
915 | unsigned int ConfigReturnBufferSizeInKByte, |
916 | unsigned int MinCompressedBufferSizeInKByte, |
917 | unsigned int CompressedBufferSegmentSizeInkByteFinal, |
918 | enum source_format_class SourcePixelFormat[], |
919 | double ReadBandwidthLuma[], |
920 | double ReadBandwidthChroma[], |
921 | unsigned int RoundedUpMaxSwathSizeBytesY[], |
922 | unsigned int RoundedUpMaxSwathSizeBytesC[], |
923 | unsigned int DPPPerSurface[], |
924 | /* Output */ |
925 | unsigned int DETBufferSizeInKByte[], |
926 | unsigned int *CompressedBufferSizeInkByte) |
927 | { |
928 | unsigned int DETBufferSizePoolInKByte; |
929 | unsigned int NextDETBufferPieceInKByte; |
930 | bool DETPieceAssignedToThisSurfaceAlready[DC__NUM_DPP__MAX]; |
931 | bool NextPotentialSurfaceToAssignDETPieceFound; |
932 | unsigned int NextSurfaceToAssignDETPiece; |
933 | double TotalBandwidth; |
934 | double BandwidthOfSurfacesNotAssignedDETPiece; |
935 | unsigned int max_minDET; |
936 | unsigned int minDET; |
937 | unsigned int minDET_pipe; |
938 | unsigned int j, k; |
939 | |
940 | #ifdef __DML_VBA_DEBUG__ |
941 | dml_print("DML::%s: ForceSingleDPP = %d\n" , __func__, ForceSingleDPP); |
942 | dml_print("DML::%s: nomDETInKByte = %d\n" , __func__, nomDETInKByte); |
943 | dml_print("DML::%s: NumberOfActiveSurfaces = %d\n" , __func__, NumberOfActiveSurfaces); |
944 | dml_print("DML::%s: UnboundedRequestEnabled = %d\n" , __func__, UnboundedRequestEnabled); |
945 | dml_print("DML::%s: MaxTotalDETInKByte = %d\n" , __func__, MaxTotalDETInKByte); |
946 | dml_print("DML::%s: ConfigReturnBufferSizeInKByte = %d\n" , __func__, ConfigReturnBufferSizeInKByte); |
947 | dml_print("DML::%s: MinCompressedBufferSizeInKByte = %d\n" , __func__, MinCompressedBufferSizeInKByte); |
948 | dml_print("DML::%s: CompressedBufferSegmentSizeInkByteFinal = %d\n" , __func__, |
949 | CompressedBufferSegmentSizeInkByteFinal); |
950 | #endif |
951 | |
952 | // Note: Will use default det size if that fits 2 swaths |
953 | if (UnboundedRequestEnabled) { |
954 | if (DETSizeOverride[0] > 0) { |
955 | DETBufferSizeInKByte[0] = DETSizeOverride[0]; |
956 | } else { |
957 | DETBufferSizeInKByte[0] = dml_max(a: nomDETInKByte, b: dml_ceil(a: 2.0 * |
958 | ((double) RoundedUpMaxSwathSizeBytesY[0] + |
959 | (double) RoundedUpMaxSwathSizeBytesC[0]) / 1024.0, granularity: 64.0)); |
960 | } |
961 | *CompressedBufferSizeInkByte = ConfigReturnBufferSizeInKByte - DETBufferSizeInKByte[0]; |
962 | } else { |
963 | DETBufferSizePoolInKByte = MaxTotalDETInKByte; |
964 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
965 | DETBufferSizeInKByte[k] = nomDETInKByte; |
966 | if (SourcePixelFormat[k] == dm_420_8 || SourcePixelFormat[k] == dm_420_10 || |
967 | SourcePixelFormat[k] == dm_420_12) { |
968 | max_minDET = nomDETInKByte - 64; |
969 | } else { |
970 | max_minDET = nomDETInKByte; |
971 | } |
972 | minDET = 128; |
973 | minDET_pipe = 0; |
974 | |
975 | // add DET resource until can hold 2 full swaths |
976 | while (minDET <= max_minDET && minDET_pipe == 0) { |
977 | if (2.0 * ((double) RoundedUpMaxSwathSizeBytesY[k] + |
978 | (double) RoundedUpMaxSwathSizeBytesC[k]) / 1024.0 <= minDET) |
979 | minDET_pipe = minDET; |
980 | minDET = minDET + 64; |
981 | } |
982 | |
983 | #ifdef __DML_VBA_DEBUG__ |
984 | dml_print("DML::%s: k=%0d minDET = %d\n" , __func__, k, minDET); |
985 | dml_print("DML::%s: k=%0d max_minDET = %d\n" , __func__, k, max_minDET); |
986 | dml_print("DML::%s: k=%0d minDET_pipe = %d\n" , __func__, k, minDET_pipe); |
987 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesY = %d\n" , __func__, k, |
988 | RoundedUpMaxSwathSizeBytesY[k]); |
989 | dml_print("DML::%s: k=%0d RoundedUpMaxSwathSizeBytesC = %d\n" , __func__, k, |
990 | RoundedUpMaxSwathSizeBytesC[k]); |
991 | #endif |
992 | |
993 | if (minDET_pipe == 0) { |
994 | minDET_pipe = dml_max(a: 128, b: dml_ceil(a: ((double)RoundedUpMaxSwathSizeBytesY[k] + |
995 | (double)RoundedUpMaxSwathSizeBytesC[k]) / 1024.0, granularity: 64)); |
996 | #ifdef __DML_VBA_DEBUG__ |
997 | dml_print("DML::%s: k=%0d minDET_pipe = %d (assume each plane take half DET)\n" , |
998 | __func__, k, minDET_pipe); |
999 | #endif |
1000 | } |
1001 | |
1002 | if (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) { |
1003 | DETBufferSizeInKByte[k] = 0; |
1004 | } else if (DETSizeOverride[k] > 0) { |
1005 | DETBufferSizeInKByte[k] = DETSizeOverride[k]; |
1006 | DETBufferSizePoolInKByte = DETBufferSizePoolInKByte - |
1007 | (ForceSingleDPP ? 1 : DPPPerSurface[k]) * DETSizeOverride[k]; |
1008 | } else if ((ForceSingleDPP ? 1 : DPPPerSurface[k]) * minDET_pipe <= DETBufferSizePoolInKByte) { |
1009 | DETBufferSizeInKByte[k] = minDET_pipe; |
1010 | DETBufferSizePoolInKByte = DETBufferSizePoolInKByte - |
1011 | (ForceSingleDPP ? 1 : DPPPerSurface[k]) * minDET_pipe; |
1012 | } |
1013 | |
1014 | #ifdef __DML_VBA_DEBUG__ |
1015 | dml_print("DML::%s: k=%d DPPPerSurface = %d\n" , __func__, k, DPPPerSurface[k]); |
1016 | dml_print("DML::%s: k=%d DETSizeOverride = %d\n" , __func__, k, DETSizeOverride[k]); |
1017 | dml_print("DML::%s: k=%d DETBufferSizeInKByte = %d\n" , __func__, k, DETBufferSizeInKByte[k]); |
1018 | dml_print("DML::%s: DETBufferSizePoolInKByte = %d\n" , __func__, DETBufferSizePoolInKByte); |
1019 | #endif |
1020 | } |
1021 | |
1022 | TotalBandwidth = 0; |
1023 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1024 | if (UseMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) |
1025 | TotalBandwidth = TotalBandwidth + ReadBandwidthLuma[k] + ReadBandwidthChroma[k]; |
1026 | } |
1027 | #ifdef __DML_VBA_DEBUG__ |
1028 | dml_print("DML::%s: --- Before bandwidth adjustment ---\n" , __func__); |
1029 | for (uint k = 0; k < NumberOfActiveSurfaces; ++k) |
1030 | dml_print("DML::%s: k=%d DETBufferSizeInKByte = %d\n" , __func__, k, DETBufferSizeInKByte[k]); |
1031 | dml_print("DML::%s: --- DET allocation with bandwidth ---\n" , __func__); |
1032 | dml_print("DML::%s: TotalBandwidth = %f\n" , __func__, TotalBandwidth); |
1033 | #endif |
1034 | BandwidthOfSurfacesNotAssignedDETPiece = TotalBandwidth; |
1035 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1036 | |
1037 | if (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) { |
1038 | DETPieceAssignedToThisSurfaceAlready[k] = true; |
1039 | } else if (DETSizeOverride[k] > 0 || (((double) (ForceSingleDPP ? 1 : DPPPerSurface[k]) * |
1040 | (double) DETBufferSizeInKByte[k] / (double) MaxTotalDETInKByte) >= |
1041 | ((ReadBandwidthLuma[k] + ReadBandwidthChroma[k]) / TotalBandwidth))) { |
1042 | DETPieceAssignedToThisSurfaceAlready[k] = true; |
1043 | BandwidthOfSurfacesNotAssignedDETPiece = BandwidthOfSurfacesNotAssignedDETPiece - |
1044 | ReadBandwidthLuma[k] - ReadBandwidthChroma[k]; |
1045 | } else { |
1046 | DETPieceAssignedToThisSurfaceAlready[k] = false; |
1047 | } |
1048 | #ifdef __DML_VBA_DEBUG__ |
1049 | dml_print("DML::%s: k=%d DETPieceAssignedToThisSurfaceAlready = %d\n" , __func__, k, |
1050 | DETPieceAssignedToThisSurfaceAlready[k]); |
1051 | dml_print("DML::%s: k=%d BandwidthOfSurfacesNotAssignedDETPiece = %f\n" , __func__, k, |
1052 | BandwidthOfSurfacesNotAssignedDETPiece); |
1053 | #endif |
1054 | } |
1055 | |
1056 | for (j = 0; j < NumberOfActiveSurfaces; ++j) { |
1057 | NextPotentialSurfaceToAssignDETPieceFound = false; |
1058 | NextSurfaceToAssignDETPiece = 0; |
1059 | |
1060 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1061 | #ifdef __DML_VBA_DEBUG__ |
1062 | dml_print("DML::%s: j=%d k=%d, ReadBandwidthLuma[k] = %f\n" , __func__, j, k, |
1063 | ReadBandwidthLuma[k]); |
1064 | dml_print("DML::%s: j=%d k=%d, ReadBandwidthChroma[k] = %f\n" , __func__, j, k, |
1065 | ReadBandwidthChroma[k]); |
1066 | dml_print("DML::%s: j=%d k=%d, ReadBandwidthLuma[Next] = %f\n" , __func__, j, k, |
1067 | ReadBandwidthLuma[NextSurfaceToAssignDETPiece]); |
1068 | dml_print("DML::%s: j=%d k=%d, ReadBandwidthChroma[Next] = %f\n" , __func__, j, k, |
1069 | ReadBandwidthChroma[NextSurfaceToAssignDETPiece]); |
1070 | dml_print("DML::%s: j=%d k=%d, NextSurfaceToAssignDETPiece = %d\n" , __func__, j, k, |
1071 | NextSurfaceToAssignDETPiece); |
1072 | #endif |
1073 | if (!DETPieceAssignedToThisSurfaceAlready[k] && |
1074 | (!NextPotentialSurfaceToAssignDETPieceFound || |
1075 | ReadBandwidthLuma[k] + ReadBandwidthChroma[k] < |
1076 | ReadBandwidthLuma[NextSurfaceToAssignDETPiece] + |
1077 | ReadBandwidthChroma[NextSurfaceToAssignDETPiece])) { |
1078 | NextSurfaceToAssignDETPiece = k; |
1079 | NextPotentialSurfaceToAssignDETPieceFound = true; |
1080 | } |
1081 | #ifdef __DML_VBA_DEBUG__ |
1082 | dml_print("DML::%s: j=%d k=%d, DETPieceAssignedToThisSurfaceAlready = %d\n" , |
1083 | __func__, j, k, DETPieceAssignedToThisSurfaceAlready[k]); |
1084 | dml_print("DML::%s: j=%d k=%d, NextPotentialSurfaceToAssignDETPieceFound = %d\n" , |
1085 | __func__, j, k, NextPotentialSurfaceToAssignDETPieceFound); |
1086 | #endif |
1087 | } |
1088 | |
1089 | if (NextPotentialSurfaceToAssignDETPieceFound) { |
1090 | // Note: To show the banker's rounding behavior in VBA and also the fact |
1091 | // that the DET buffer size varies due to precision issue |
1092 | // |
1093 | //double tmp1 = ((double) DETBufferSizePoolInKByte * |
1094 | // (ReadBandwidthLuma[NextSurfaceToAssignDETPiece] + |
1095 | // ReadBandwidthChroma[NextSurfaceToAssignDETPiece]) / |
1096 | // BandwidthOfSurfacesNotAssignedDETPiece / |
1097 | // ((ForceSingleDPP ? 1 : DPPPerSurface[NextSurfaceToAssignDETPiece]) * 64.0)); |
1098 | //double tmp2 = dml_round((double) DETBufferSizePoolInKByte * |
1099 | // (ReadBandwidthLuma[NextSurfaceToAssignDETPiece] + |
1100 | // ReadBandwidthChroma[NextSurfaceToAssignDETPiece]) / |
1101 | //BandwidthOfSurfacesNotAssignedDETPiece / |
1102 | // ((ForceSingleDPP ? 1 : DPPPerSurface[NextSurfaceToAssignDETPiece]) * 64.0)); |
1103 | // |
1104 | //dml_print("DML::%s: j=%d, tmp1 = %f\n", __func__, j, tmp1); |
1105 | //dml_print("DML::%s: j=%d, tmp2 = %f\n", __func__, j, tmp2); |
1106 | |
1107 | NextDETBufferPieceInKByte = dml_min( |
1108 | a: dml_round(a: (double) DETBufferSizePoolInKByte * |
1109 | (ReadBandwidthLuma[NextSurfaceToAssignDETPiece] + |
1110 | ReadBandwidthChroma[NextSurfaceToAssignDETPiece]) / |
1111 | BandwidthOfSurfacesNotAssignedDETPiece / |
1112 | ((ForceSingleDPP ? 1 : |
1113 | DPPPerSurface[NextSurfaceToAssignDETPiece]) * 64.0)) * |
1114 | (ForceSingleDPP ? 1 : |
1115 | DPPPerSurface[NextSurfaceToAssignDETPiece]) * 64.0, |
1116 | b: dml_floor(a: (double) DETBufferSizePoolInKByte, |
1117 | granularity: (ForceSingleDPP ? 1 : |
1118 | DPPPerSurface[NextSurfaceToAssignDETPiece]) * 64.0)); |
1119 | |
1120 | // Above calculation can assign the entire DET buffer allocation to a single pipe. |
1121 | // We should limit the per-pipe DET size to the nominal / max per pipe. |
1122 | if (NextDETBufferPieceInKByte > nomDETInKByte * (ForceSingleDPP ? 1 : DPPPerSurface[k])) { |
1123 | if (DETBufferSizeInKByte[NextSurfaceToAssignDETPiece] < |
1124 | nomDETInKByte * (ForceSingleDPP ? 1 : DPPPerSurface[k])) { |
1125 | NextDETBufferPieceInKByte = nomDETInKByte * (ForceSingleDPP ? 1 : DPPPerSurface[k]) - |
1126 | DETBufferSizeInKByte[NextSurfaceToAssignDETPiece]; |
1127 | } else { |
1128 | // Case where DETBufferSizeInKByte[NextSurfaceToAssignDETPiece] |
1129 | // already has the max per-pipe value |
1130 | NextDETBufferPieceInKByte = 0; |
1131 | } |
1132 | } |
1133 | |
1134 | #ifdef __DML_VBA_DEBUG__ |
1135 | dml_print("DML::%s: j=%0d, DETBufferSizePoolInKByte = %d\n" , __func__, j, |
1136 | DETBufferSizePoolInKByte); |
1137 | dml_print("DML::%s: j=%0d, NextSurfaceToAssignDETPiece = %d\n" , __func__, j, |
1138 | NextSurfaceToAssignDETPiece); |
1139 | dml_print("DML::%s: j=%0d, ReadBandwidthLuma[%0d] = %f\n" , __func__, j, |
1140 | NextSurfaceToAssignDETPiece, ReadBandwidthLuma[NextSurfaceToAssignDETPiece]); |
1141 | dml_print("DML::%s: j=%0d, ReadBandwidthChroma[%0d] = %f\n" , __func__, j, |
1142 | NextSurfaceToAssignDETPiece, ReadBandwidthChroma[NextSurfaceToAssignDETPiece]); |
1143 | dml_print("DML::%s: j=%0d, BandwidthOfSurfacesNotAssignedDETPiece = %f\n" , |
1144 | __func__, j, BandwidthOfSurfacesNotAssignedDETPiece); |
1145 | dml_print("DML::%s: j=%0d, NextDETBufferPieceInKByte = %d\n" , __func__, j, |
1146 | NextDETBufferPieceInKByte); |
1147 | dml_print("DML::%s: j=%0d, DETBufferSizeInKByte[%0d] increases from %0d " , |
1148 | __func__, j, NextSurfaceToAssignDETPiece, |
1149 | DETBufferSizeInKByte[NextSurfaceToAssignDETPiece]); |
1150 | #endif |
1151 | |
1152 | DETBufferSizeInKByte[NextSurfaceToAssignDETPiece] = |
1153 | DETBufferSizeInKByte[NextSurfaceToAssignDETPiece] |
1154 | + NextDETBufferPieceInKByte |
1155 | / (ForceSingleDPP ? 1 : DPPPerSurface[NextSurfaceToAssignDETPiece]); |
1156 | #ifdef __DML_VBA_DEBUG__ |
1157 | dml_print("to %0d\n" , DETBufferSizeInKByte[NextSurfaceToAssignDETPiece]); |
1158 | #endif |
1159 | |
1160 | DETBufferSizePoolInKByte = DETBufferSizePoolInKByte - NextDETBufferPieceInKByte; |
1161 | DETPieceAssignedToThisSurfaceAlready[NextSurfaceToAssignDETPiece] = true; |
1162 | BandwidthOfSurfacesNotAssignedDETPiece = BandwidthOfSurfacesNotAssignedDETPiece - |
1163 | (ReadBandwidthLuma[NextSurfaceToAssignDETPiece] + |
1164 | ReadBandwidthChroma[NextSurfaceToAssignDETPiece]); |
1165 | } |
1166 | } |
1167 | *CompressedBufferSizeInkByte = MinCompressedBufferSizeInKByte; |
1168 | } |
1169 | *CompressedBufferSizeInkByte = *CompressedBufferSizeInkByte * CompressedBufferSegmentSizeInkByteFinal / 64; |
1170 | |
1171 | #ifdef __DML_VBA_DEBUG__ |
1172 | dml_print("DML::%s: --- After bandwidth adjustment ---\n" , __func__); |
1173 | dml_print("DML::%s: CompressedBufferSizeInkByte = %d\n" , __func__, *CompressedBufferSizeInkByte); |
1174 | for (uint k = 0; k < NumberOfActiveSurfaces; ++k) { |
1175 | dml_print("DML::%s: k=%d DETBufferSizeInKByte = %d (TotalReadBandWidth=%f)\n" , |
1176 | __func__, k, DETBufferSizeInKByte[k], ReadBandwidthLuma[k] + ReadBandwidthChroma[k]); |
1177 | } |
1178 | #endif |
1179 | } // CalculateDETBufferSize |
1180 | |
1181 | void dml32_CalculateODMMode( |
1182 | unsigned int MaximumPixelsPerLinePerDSCUnit, |
1183 | unsigned int HActive, |
1184 | enum output_format_class OutFormat, |
1185 | enum output_encoder_class Output, |
1186 | enum odm_combine_policy ODMUse, |
1187 | double StateDispclk, |
1188 | double MaxDispclk, |
1189 | bool DSCEnable, |
1190 | unsigned int TotalNumberOfActiveDPP, |
1191 | unsigned int MaxNumDPP, |
1192 | double PixelClock, |
1193 | double DISPCLKDPPCLKDSCCLKDownSpreading, |
1194 | double DISPCLKRampingMargin, |
1195 | double DISPCLKDPPCLKVCOSpeed, |
1196 | unsigned int NumberOfDSCSlices, |
1197 | |
1198 | /* Output */ |
1199 | bool *TotalAvailablePipesSupport, |
1200 | unsigned int *NumberOfDPP, |
1201 | enum odm_combine_mode *ODMMode, |
1202 | double *RequiredDISPCLKPerSurface) |
1203 | { |
1204 | |
1205 | double SurfaceRequiredDISPCLKWithoutODMCombine; |
1206 | double SurfaceRequiredDISPCLKWithODMCombineTwoToOne; |
1207 | double SurfaceRequiredDISPCLKWithODMCombineFourToOne; |
1208 | |
1209 | SurfaceRequiredDISPCLKWithoutODMCombine = dml32_CalculateRequiredDispclk(ODMMode: dm_odm_combine_mode_disabled, |
1210 | PixelClock, DISPCLKDPPCLKDSCCLKDownSpreading, DISPCLKRampingMargin, DISPCLKDPPCLKVCOSpeed, |
1211 | MaxDispclk); |
1212 | SurfaceRequiredDISPCLKWithODMCombineTwoToOne = dml32_CalculateRequiredDispclk(ODMMode: dm_odm_combine_mode_2to1, |
1213 | PixelClock, DISPCLKDPPCLKDSCCLKDownSpreading, DISPCLKRampingMargin, DISPCLKDPPCLKVCOSpeed, |
1214 | MaxDispclk); |
1215 | SurfaceRequiredDISPCLKWithODMCombineFourToOne = dml32_CalculateRequiredDispclk(ODMMode: dm_odm_combine_mode_4to1, |
1216 | PixelClock, DISPCLKDPPCLKDSCCLKDownSpreading, DISPCLKRampingMargin, DISPCLKDPPCLKVCOSpeed, |
1217 | MaxDispclk); |
1218 | *TotalAvailablePipesSupport = true; |
1219 | *ODMMode = dm_odm_combine_mode_disabled; // initialize as disable |
1220 | |
1221 | if (ODMUse == dm_odm_combine_policy_none) |
1222 | *ODMMode = dm_odm_combine_mode_disabled; |
1223 | |
1224 | *RequiredDISPCLKPerSurface = SurfaceRequiredDISPCLKWithoutODMCombine; |
1225 | *NumberOfDPP = 0; |
1226 | |
1227 | // FIXME check ODMUse == "" condition does it mean bypass or Gabriel means something like don't care?? |
1228 | // (ODMUse == "" || ODMUse == "CombineAsNeeded") |
1229 | |
1230 | if (!(Output == dm_hdmi || Output == dm_dp || Output == dm_edp) && (ODMUse == dm_odm_combine_policy_4to1 || |
1231 | ((SurfaceRequiredDISPCLKWithODMCombineTwoToOne > StateDispclk || |
1232 | (DSCEnable && (HActive > 2 * MaximumPixelsPerLinePerDSCUnit)) |
1233 | || NumberOfDSCSlices > 8)))) { |
1234 | if (TotalNumberOfActiveDPP + 4 <= MaxNumDPP) { |
1235 | *ODMMode = dm_odm_combine_mode_4to1; |
1236 | *RequiredDISPCLKPerSurface = SurfaceRequiredDISPCLKWithODMCombineFourToOne; |
1237 | *NumberOfDPP = 4; |
1238 | } else { |
1239 | *TotalAvailablePipesSupport = false; |
1240 | } |
1241 | } else if (Output != dm_hdmi && (ODMUse == dm_odm_combine_policy_2to1 || |
1242 | (((SurfaceRequiredDISPCLKWithoutODMCombine > StateDispclk && |
1243 | SurfaceRequiredDISPCLKWithODMCombineTwoToOne <= StateDispclk) || |
1244 | (DSCEnable && (HActive > MaximumPixelsPerLinePerDSCUnit)) |
1245 | || (NumberOfDSCSlices <= 8 && NumberOfDSCSlices > 4))))) { |
1246 | if (TotalNumberOfActiveDPP + 2 <= MaxNumDPP) { |
1247 | *ODMMode = dm_odm_combine_mode_2to1; |
1248 | *RequiredDISPCLKPerSurface = SurfaceRequiredDISPCLKWithODMCombineTwoToOne; |
1249 | *NumberOfDPP = 2; |
1250 | } else { |
1251 | *TotalAvailablePipesSupport = false; |
1252 | } |
1253 | } else { |
1254 | if (TotalNumberOfActiveDPP + 1 <= MaxNumDPP) |
1255 | *NumberOfDPP = 1; |
1256 | else |
1257 | *TotalAvailablePipesSupport = false; |
1258 | } |
1259 | if (OutFormat == dm_420 && HActive > DCN32_MAX_FMT_420_BUFFER_WIDTH && |
1260 | ODMUse != dm_odm_combine_policy_4to1) { |
1261 | if (HActive > DCN32_MAX_FMT_420_BUFFER_WIDTH * 4) { |
1262 | *ODMMode = dm_odm_combine_mode_disabled; |
1263 | *NumberOfDPP = 0; |
1264 | *TotalAvailablePipesSupport = false; |
1265 | } else if (HActive > DCN32_MAX_FMT_420_BUFFER_WIDTH * 2 || |
1266 | *ODMMode == dm_odm_combine_mode_4to1) { |
1267 | *ODMMode = dm_odm_combine_mode_4to1; |
1268 | *RequiredDISPCLKPerSurface = SurfaceRequiredDISPCLKWithODMCombineFourToOne; |
1269 | *NumberOfDPP = 4; |
1270 | } else { |
1271 | *ODMMode = dm_odm_combine_mode_2to1; |
1272 | *RequiredDISPCLKPerSurface = SurfaceRequiredDISPCLKWithODMCombineTwoToOne; |
1273 | *NumberOfDPP = 2; |
1274 | } |
1275 | } |
1276 | if (Output == dm_hdmi && OutFormat == dm_420 && |
1277 | HActive > DCN32_MAX_FMT_420_BUFFER_WIDTH) { |
1278 | *ODMMode = dm_odm_combine_mode_disabled; |
1279 | *NumberOfDPP = 0; |
1280 | *TotalAvailablePipesSupport = false; |
1281 | } |
1282 | } |
1283 | |
1284 | double dml32_CalculateRequiredDispclk( |
1285 | enum odm_combine_mode ODMMode, |
1286 | double PixelClock, |
1287 | double DISPCLKDPPCLKDSCCLKDownSpreading, |
1288 | double DISPCLKRampingMargin, |
1289 | double DISPCLKDPPCLKVCOSpeed, |
1290 | double MaxDispclk) |
1291 | { |
1292 | double RequiredDispclk = 0.; |
1293 | double PixelClockAfterODM; |
1294 | double DISPCLKWithRampingRoundedToDFSGranularity; |
1295 | double DISPCLKWithoutRampingRoundedToDFSGranularity; |
1296 | double MaxDispclkRoundedDownToDFSGranularity; |
1297 | |
1298 | if (ODMMode == dm_odm_combine_mode_4to1) |
1299 | PixelClockAfterODM = PixelClock / 4; |
1300 | else if (ODMMode == dm_odm_combine_mode_2to1) |
1301 | PixelClockAfterODM = PixelClock / 2; |
1302 | else |
1303 | PixelClockAfterODM = PixelClock; |
1304 | |
1305 | |
1306 | DISPCLKWithRampingRoundedToDFSGranularity = dml32_RoundToDFSGranularity( |
1307 | Clock: PixelClockAfterODM * (1 + DISPCLKDPPCLKDSCCLKDownSpreading / 100) |
1308 | * (1 + DISPCLKRampingMargin / 100), round_up: 1, VCOSpeed: DISPCLKDPPCLKVCOSpeed); |
1309 | |
1310 | DISPCLKWithoutRampingRoundedToDFSGranularity = dml32_RoundToDFSGranularity( |
1311 | Clock: PixelClockAfterODM * (1 + DISPCLKDPPCLKDSCCLKDownSpreading / 100), round_up: 1, VCOSpeed: DISPCLKDPPCLKVCOSpeed); |
1312 | |
1313 | MaxDispclkRoundedDownToDFSGranularity = dml32_RoundToDFSGranularity(Clock: MaxDispclk, round_up: 0, VCOSpeed: DISPCLKDPPCLKVCOSpeed); |
1314 | |
1315 | if (DISPCLKWithoutRampingRoundedToDFSGranularity > MaxDispclkRoundedDownToDFSGranularity) |
1316 | RequiredDispclk = DISPCLKWithoutRampingRoundedToDFSGranularity; |
1317 | else if (DISPCLKWithRampingRoundedToDFSGranularity > MaxDispclkRoundedDownToDFSGranularity) |
1318 | RequiredDispclk = MaxDispclkRoundedDownToDFSGranularity; |
1319 | else |
1320 | RequiredDispclk = DISPCLKWithRampingRoundedToDFSGranularity; |
1321 | |
1322 | return RequiredDispclk; |
1323 | } |
1324 | |
1325 | double dml32_RoundToDFSGranularity(double Clock, bool round_up, double VCOSpeed) |
1326 | { |
1327 | if (Clock <= 0.0) |
1328 | return 0.0; |
1329 | |
1330 | if (round_up) |
1331 | return VCOSpeed * 4.0 / dml_floor(a: VCOSpeed * 4.0 / Clock, granularity: 1.0); |
1332 | else |
1333 | return VCOSpeed * 4.0 / dml_ceil(a: VCOSpeed * 4.0 / Clock, granularity: 1.0); |
1334 | } |
1335 | |
1336 | void dml32_CalculateOutputLink( |
1337 | double PHYCLKPerState, |
1338 | double PHYCLKD18PerState, |
1339 | double PHYCLKD32PerState, |
1340 | double Downspreading, |
1341 | bool IsMainSurfaceUsingTheIndicatedTiming, |
1342 | enum output_encoder_class Output, |
1343 | enum output_format_class OutputFormat, |
1344 | unsigned int HTotal, |
1345 | unsigned int HActive, |
1346 | double PixelClockBackEnd, |
1347 | double ForcedOutputLinkBPP, |
1348 | unsigned int DSCInputBitPerComponent, |
1349 | unsigned int NumberOfDSCSlices, |
1350 | double AudioSampleRate, |
1351 | unsigned int AudioSampleLayout, |
1352 | enum odm_combine_mode ODMModeNoDSC, |
1353 | enum odm_combine_mode ODMModeDSC, |
1354 | bool DSCEnable, |
1355 | unsigned int OutputLinkDPLanes, |
1356 | enum dm_output_link_dp_rate OutputLinkDPRate, |
1357 | |
1358 | /* Output */ |
1359 | bool *RequiresDSC, |
1360 | double *RequiresFEC, |
1361 | double *OutBpp, |
1362 | enum dm_output_type *OutputType, |
1363 | enum dm_output_rate *OutputRate, |
1364 | unsigned int *RequiredSlots) |
1365 | { |
1366 | bool LinkDSCEnable; |
1367 | unsigned int dummy; |
1368 | *RequiresDSC = false; |
1369 | *RequiresFEC = false; |
1370 | *OutBpp = 0; |
1371 | *OutputType = dm_output_type_unknown; |
1372 | *OutputRate = dm_output_rate_unknown; |
1373 | |
1374 | if (IsMainSurfaceUsingTheIndicatedTiming) { |
1375 | if (Output == dm_hdmi) { |
1376 | *RequiresDSC = false; |
1377 | *RequiresFEC = false; |
1378 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: dml_min(a: 600, b: PHYCLKPerState) * 10, Lanes: 3, HTotal, HActive, |
1379 | PixelClock: PixelClockBackEnd, DesiredBPP: ForcedOutputLinkBPP, DSCEnable: false, Output, Format: OutputFormat, |
1380 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1381 | ODMModeNoDSC, ODMModeDSC, RequiredSlots: &dummy); |
1382 | //OutputTypeAndRate = "HDMI"; |
1383 | *OutputType = dm_output_type_hdmi; |
1384 | |
1385 | } else if (Output == dm_dp || Output == dm_dp2p0 || Output == dm_edp) { |
1386 | if (DSCEnable == true) { |
1387 | *RequiresDSC = true; |
1388 | LinkDSCEnable = true; |
1389 | if (Output == dm_dp || Output == dm_dp2p0) |
1390 | *RequiresFEC = true; |
1391 | else |
1392 | *RequiresFEC = false; |
1393 | } else { |
1394 | *RequiresDSC = false; |
1395 | LinkDSCEnable = false; |
1396 | if (Output == dm_dp2p0) |
1397 | *RequiresFEC = true; |
1398 | else |
1399 | *RequiresFEC = false; |
1400 | } |
1401 | if (Output == dm_dp2p0) { |
1402 | *OutBpp = 0; |
1403 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_uhbr10) && |
1404 | PHYCLKD32PerState >= 10000 / 32) { |
1405 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 10000, |
1406 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1407 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, Format: OutputFormat, |
1408 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, |
1409 | AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1410 | if (*OutBpp == 0 && PHYCLKD32PerState < 13500 / 32 && DSCEnable == true && |
1411 | ForcedOutputLinkBPP == 0) { |
1412 | *RequiresDSC = true; |
1413 | LinkDSCEnable = true; |
1414 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 10000, |
1415 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1416 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1417 | Format: OutputFormat, DSCInputBitPerComponent, |
1418 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1419 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1420 | } |
1421 | //OutputTypeAndRate = Output & " UHBR10"; |
1422 | *OutputType = dm_output_type_dp2p0; |
1423 | *OutputRate = dm_output_rate_dp_rate_uhbr10; |
1424 | } |
1425 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_uhbr13p5) && |
1426 | *OutBpp == 0 && PHYCLKD32PerState >= 13500 / 32) { |
1427 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 13500, |
1428 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1429 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, Format: OutputFormat, |
1430 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, |
1431 | AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1432 | |
1433 | if (*OutBpp == 0 && PHYCLKD32PerState < 20000 / 32 && DSCEnable == true && |
1434 | ForcedOutputLinkBPP == 0) { |
1435 | *RequiresDSC = true; |
1436 | LinkDSCEnable = true; |
1437 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 13500, |
1438 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1439 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1440 | Format: OutputFormat, DSCInputBitPerComponent, |
1441 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1442 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1443 | } |
1444 | //OutputTypeAndRate = Output & " UHBR13p5"; |
1445 | *OutputType = dm_output_type_dp2p0; |
1446 | *OutputRate = dm_output_rate_dp_rate_uhbr13p5; |
1447 | } |
1448 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_uhbr20) && |
1449 | *OutBpp == 0 && PHYCLKD32PerState >= 20000 / 32) { |
1450 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 20000, |
1451 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1452 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, Format: OutputFormat, |
1453 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, |
1454 | AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1455 | if (*OutBpp == 0 && DSCEnable == true && ForcedOutputLinkBPP == 0) { |
1456 | *RequiresDSC = true; |
1457 | LinkDSCEnable = true; |
1458 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 20000, |
1459 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1460 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1461 | Format: OutputFormat, DSCInputBitPerComponent, |
1462 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1463 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1464 | } |
1465 | //OutputTypeAndRate = Output & " UHBR20"; |
1466 | *OutputType = dm_output_type_dp2p0; |
1467 | *OutputRate = dm_output_rate_dp_rate_uhbr20; |
1468 | } |
1469 | } else { |
1470 | *OutBpp = 0; |
1471 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_hbr) && |
1472 | PHYCLKPerState >= 270) { |
1473 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 2700, |
1474 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1475 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, Format: OutputFormat, |
1476 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, |
1477 | AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1478 | if (*OutBpp == 0 && PHYCLKPerState < 540 && DSCEnable == true && |
1479 | ForcedOutputLinkBPP == 0) { |
1480 | *RequiresDSC = true; |
1481 | LinkDSCEnable = true; |
1482 | if (Output == dm_dp) |
1483 | *RequiresFEC = true; |
1484 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 2700, |
1485 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1486 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1487 | Format: OutputFormat, DSCInputBitPerComponent, |
1488 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1489 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1490 | } |
1491 | //OutputTypeAndRate = Output & " HBR"; |
1492 | *OutputType = (Output == dm_dp) ? dm_output_type_dp : dm_output_type_edp; |
1493 | *OutputRate = dm_output_rate_dp_rate_hbr; |
1494 | } |
1495 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_hbr2) && |
1496 | *OutBpp == 0 && PHYCLKPerState >= 540) { |
1497 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 5400, |
1498 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1499 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, Format: OutputFormat, |
1500 | DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, |
1501 | AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1502 | |
1503 | if (*OutBpp == 0 && PHYCLKPerState < 810 && DSCEnable == true && |
1504 | ForcedOutputLinkBPP == 0) { |
1505 | *RequiresDSC = true; |
1506 | LinkDSCEnable = true; |
1507 | if (Output == dm_dp) |
1508 | *RequiresFEC = true; |
1509 | |
1510 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 5400, |
1511 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1512 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1513 | Format: OutputFormat, DSCInputBitPerComponent, |
1514 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1515 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1516 | } |
1517 | //OutputTypeAndRate = Output & " HBR2"; |
1518 | *OutputType = (Output == dm_dp) ? dm_output_type_dp : dm_output_type_edp; |
1519 | *OutputRate = dm_output_rate_dp_rate_hbr2; |
1520 | } |
1521 | if ((OutputLinkDPRate == dm_dp_rate_na || OutputLinkDPRate == dm_dp_rate_hbr3) && *OutBpp == 0 && PHYCLKPerState >= 810) { |
1522 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 8100, |
1523 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1524 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1525 | Format: OutputFormat, DSCInputBitPerComponent, DSCSlices: NumberOfDSCSlices, |
1526 | AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, ODMModeNoDSC, ODMModeDSC, |
1527 | RequiredSlots); |
1528 | |
1529 | if (*OutBpp == 0 && DSCEnable == true && ForcedOutputLinkBPP == 0) { |
1530 | *RequiresDSC = true; |
1531 | LinkDSCEnable = true; |
1532 | if (Output == dm_dp) |
1533 | *RequiresFEC = true; |
1534 | |
1535 | *OutBpp = dml32_TruncToValidBPP(LinkBitRate: (1 - Downspreading / 100) * 8100, |
1536 | Lanes: OutputLinkDPLanes, HTotal, HActive, PixelClock: PixelClockBackEnd, |
1537 | DesiredBPP: ForcedOutputLinkBPP, DSCEnable: LinkDSCEnable, Output, |
1538 | Format: OutputFormat, DSCInputBitPerComponent, |
1539 | DSCSlices: NumberOfDSCSlices, AudioRate: AudioSampleRate, AudioLayout: AudioSampleLayout, |
1540 | ODMModeNoDSC, ODMModeDSC, RequiredSlots); |
1541 | } |
1542 | //OutputTypeAndRate = Output & " HBR3"; |
1543 | *OutputType = (Output == dm_dp) ? dm_output_type_dp : dm_output_type_edp; |
1544 | *OutputRate = dm_output_rate_dp_rate_hbr3; |
1545 | } |
1546 | } |
1547 | } |
1548 | } |
1549 | } |
1550 | |
1551 | void dml32_CalculateDPPCLK( |
1552 | unsigned int NumberOfActiveSurfaces, |
1553 | double DISPCLKDPPCLKDSCCLKDownSpreading, |
1554 | double DISPCLKDPPCLKVCOSpeed, |
1555 | double DPPCLKUsingSingleDPP[], |
1556 | unsigned int DPPPerSurface[], |
1557 | |
1558 | /* output */ |
1559 | double *GlobalDPPCLK, |
1560 | double Dppclk[]) |
1561 | { |
1562 | unsigned int k; |
1563 | *GlobalDPPCLK = 0; |
1564 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1565 | Dppclk[k] = DPPCLKUsingSingleDPP[k] / DPPPerSurface[k] * (1 + DISPCLKDPPCLKDSCCLKDownSpreading / 100); |
1566 | *GlobalDPPCLK = dml_max(a: *GlobalDPPCLK, b: Dppclk[k]); |
1567 | } |
1568 | *GlobalDPPCLK = dml32_RoundToDFSGranularity(Clock: *GlobalDPPCLK, round_up: 1, VCOSpeed: DISPCLKDPPCLKVCOSpeed); |
1569 | for (k = 0; k < NumberOfActiveSurfaces; ++k) |
1570 | Dppclk[k] = *GlobalDPPCLK / 255 * dml_ceil(a: Dppclk[k] * 255.0 / *GlobalDPPCLK, granularity: 1.0); |
1571 | } |
1572 | |
1573 | double dml32_TruncToValidBPP( |
1574 | double LinkBitRate, |
1575 | unsigned int Lanes, |
1576 | unsigned int HTotal, |
1577 | unsigned int HActive, |
1578 | double PixelClock, |
1579 | double DesiredBPP, |
1580 | bool DSCEnable, |
1581 | enum output_encoder_class Output, |
1582 | enum output_format_class Format, |
1583 | unsigned int DSCInputBitPerComponent, |
1584 | unsigned int DSCSlices, |
1585 | unsigned int AudioRate, |
1586 | unsigned int AudioLayout, |
1587 | enum odm_combine_mode ODMModeNoDSC, |
1588 | enum odm_combine_mode ODMModeDSC, |
1589 | /* Output */ |
1590 | unsigned int *RequiredSlots) |
1591 | { |
1592 | double MaxLinkBPP; |
1593 | unsigned int MinDSCBPP; |
1594 | double MaxDSCBPP; |
1595 | unsigned int NonDSCBPP0; |
1596 | unsigned int NonDSCBPP1; |
1597 | unsigned int NonDSCBPP2; |
1598 | |
1599 | if (Format == dm_420) { |
1600 | NonDSCBPP0 = 12; |
1601 | NonDSCBPP1 = 15; |
1602 | NonDSCBPP2 = 18; |
1603 | MinDSCBPP = 6; |
1604 | MaxDSCBPP = 1.5 * DSCInputBitPerComponent - 1 / 16; |
1605 | } else if (Format == dm_444) { |
1606 | NonDSCBPP0 = 24; |
1607 | NonDSCBPP1 = 30; |
1608 | NonDSCBPP2 = 36; |
1609 | MinDSCBPP = 8; |
1610 | MaxDSCBPP = 3 * DSCInputBitPerComponent - 1.0 / 16; |
1611 | } else { |
1612 | if (Output == dm_hdmi) { |
1613 | NonDSCBPP0 = 24; |
1614 | NonDSCBPP1 = 24; |
1615 | NonDSCBPP2 = 24; |
1616 | } else { |
1617 | NonDSCBPP0 = 16; |
1618 | NonDSCBPP1 = 20; |
1619 | NonDSCBPP2 = 24; |
1620 | } |
1621 | if (Format == dm_n422) { |
1622 | MinDSCBPP = 7; |
1623 | MaxDSCBPP = 2 * DSCInputBitPerComponent - 1.0 / 16.0; |
1624 | } else { |
1625 | MinDSCBPP = 8; |
1626 | MaxDSCBPP = 3 * DSCInputBitPerComponent - 1.0 / 16.0; |
1627 | } |
1628 | } |
1629 | if (Output == dm_dp2p0) { |
1630 | MaxLinkBPP = LinkBitRate * Lanes / PixelClock * 128 / 132 * 383 / 384 * 65536 / 65540; |
1631 | } else if (DSCEnable && Output == dm_dp) { |
1632 | MaxLinkBPP = LinkBitRate / 10 * 8 * Lanes / PixelClock * (1 - 2.4 / 100); |
1633 | } else { |
1634 | MaxLinkBPP = LinkBitRate / 10 * 8 * Lanes / PixelClock; |
1635 | } |
1636 | |
1637 | if (DSCEnable) { |
1638 | if (ODMModeDSC == dm_odm_combine_mode_4to1) |
1639 | MaxLinkBPP = dml_min(a: MaxLinkBPP, b: 16); |
1640 | else if (ODMModeDSC == dm_odm_combine_mode_2to1) |
1641 | MaxLinkBPP = dml_min(a: MaxLinkBPP, b: 32); |
1642 | else if (ODMModeDSC == dm_odm_split_mode_1to2) |
1643 | MaxLinkBPP = 2 * MaxLinkBPP; |
1644 | } else { |
1645 | if (ODMModeNoDSC == dm_odm_combine_mode_4to1) |
1646 | MaxLinkBPP = dml_min(a: MaxLinkBPP, b: 16); |
1647 | else if (ODMModeNoDSC == dm_odm_combine_mode_2to1) |
1648 | MaxLinkBPP = dml_min(a: MaxLinkBPP, b: 32); |
1649 | else if (ODMModeNoDSC == dm_odm_split_mode_1to2) |
1650 | MaxLinkBPP = 2 * MaxLinkBPP; |
1651 | } |
1652 | |
1653 | if (DesiredBPP == 0) { |
1654 | if (DSCEnable) { |
1655 | if (MaxLinkBPP < MinDSCBPP) |
1656 | return BPP_INVALID; |
1657 | else if (MaxLinkBPP >= MaxDSCBPP) |
1658 | return MaxDSCBPP; |
1659 | else |
1660 | return dml_floor(a: 16.0 * MaxLinkBPP, granularity: 1.0) / 16.0; |
1661 | } else { |
1662 | if (MaxLinkBPP >= NonDSCBPP2) |
1663 | return NonDSCBPP2; |
1664 | else if (MaxLinkBPP >= NonDSCBPP1) |
1665 | return NonDSCBPP1; |
1666 | else if (MaxLinkBPP >= NonDSCBPP0) |
1667 | return 16.0; |
1668 | else |
1669 | return BPP_INVALID; |
1670 | } |
1671 | } else { |
1672 | if (!((DSCEnable == false && (DesiredBPP == NonDSCBPP2 || DesiredBPP == NonDSCBPP1 || |
1673 | DesiredBPP <= NonDSCBPP0)) || |
1674 | (DSCEnable && DesiredBPP >= MinDSCBPP && DesiredBPP <= MaxDSCBPP))) |
1675 | return BPP_INVALID; |
1676 | else |
1677 | return DesiredBPP; |
1678 | } |
1679 | |
1680 | *RequiredSlots = dml_ceil(a: DesiredBPP / MaxLinkBPP * 64, granularity: 1); |
1681 | |
1682 | return BPP_INVALID; |
1683 | } // TruncToValidBPP |
1684 | |
1685 | double dml32_RequiredDTBCLK( |
1686 | bool DSCEnable, |
1687 | double PixelClock, |
1688 | enum output_format_class OutputFormat, |
1689 | double OutputBpp, |
1690 | unsigned int DSCSlices, |
1691 | unsigned int HTotal, |
1692 | unsigned int HActive, |
1693 | unsigned int AudioRate, |
1694 | unsigned int AudioLayout) |
1695 | { |
1696 | double PixelWordRate; |
1697 | double HCActive; |
1698 | double HCBlank; |
1699 | double AverageTribyteRate; |
1700 | double HActiveTribyteRate; |
1701 | |
1702 | if (DSCEnable != true) |
1703 | return dml_max(a: PixelClock / 4.0 * OutputBpp / 24.0, b: 25.0); |
1704 | |
1705 | PixelWordRate = PixelClock / (OutputFormat == dm_444 ? 1 : 2); |
1706 | HCActive = dml_ceil(a: DSCSlices * dml_ceil(a: OutputBpp * |
1707 | dml_ceil(a: HActive / DSCSlices, granularity: 1) / 8.0, granularity: 1) / 3.0, granularity: 1); |
1708 | HCBlank = 64 + 32 * |
1709 | dml_ceil(a: AudioRate * (AudioLayout == 1 ? 1 : 0.25) * HTotal / (PixelClock * 1000), granularity: 1); |
1710 | AverageTribyteRate = PixelWordRate * (HCActive + HCBlank) / HTotal; |
1711 | HActiveTribyteRate = PixelWordRate * HCActive / HActive; |
1712 | return dml_max4(a: PixelWordRate / 4.0, b: AverageTribyteRate / 4.0, c: HActiveTribyteRate / 4.0, d: 25.0) * 1.002; |
1713 | } |
1714 | |
1715 | unsigned int dml32_DSCDelayRequirement(bool DSCEnabled, |
1716 | enum odm_combine_mode ODMMode, |
1717 | unsigned int DSCInputBitPerComponent, |
1718 | double OutputBpp, |
1719 | unsigned int HActive, |
1720 | unsigned int HTotal, |
1721 | unsigned int NumberOfDSCSlices, |
1722 | enum output_format_class OutputFormat, |
1723 | enum output_encoder_class Output, |
1724 | double PixelClock, |
1725 | double PixelClockBackEnd, |
1726 | double dsc_delay_factor_wa) |
1727 | { |
1728 | unsigned int DSCDelayRequirement_val; |
1729 | |
1730 | if (DSCEnabled == true && OutputBpp != 0) { |
1731 | if (ODMMode == dm_odm_combine_mode_4to1) { |
1732 | DSCDelayRequirement_val = 4 * (dml32_dscceComputeDelay(bpc: DSCInputBitPerComponent, BPP: OutputBpp, |
1733 | sliceWidth: dml_ceil(a: HActive / NumberOfDSCSlices, granularity: 1), numSlices: NumberOfDSCSlices / 4, |
1734 | pixelFormat: OutputFormat, Output) + dml32_dscComputeDelay(pixelFormat: OutputFormat, Output)); |
1735 | } else if (ODMMode == dm_odm_combine_mode_2to1) { |
1736 | DSCDelayRequirement_val = 2 * (dml32_dscceComputeDelay(bpc: DSCInputBitPerComponent, BPP: OutputBpp, |
1737 | sliceWidth: dml_ceil(a: HActive / NumberOfDSCSlices, granularity: 1), numSlices: NumberOfDSCSlices / 2, |
1738 | pixelFormat: OutputFormat, Output) + dml32_dscComputeDelay(pixelFormat: OutputFormat, Output)); |
1739 | } else { |
1740 | DSCDelayRequirement_val = dml32_dscceComputeDelay(bpc: DSCInputBitPerComponent, BPP: OutputBpp, |
1741 | sliceWidth: dml_ceil(a: HActive / NumberOfDSCSlices, granularity: 1), numSlices: NumberOfDSCSlices, |
1742 | pixelFormat: OutputFormat, Output) + dml32_dscComputeDelay(pixelFormat: OutputFormat, Output); |
1743 | } |
1744 | |
1745 | DSCDelayRequirement_val = DSCDelayRequirement_val + (HTotal - HActive) * |
1746 | dml_ceil(a: (double)DSCDelayRequirement_val / HActive, granularity: 1); |
1747 | |
1748 | DSCDelayRequirement_val = DSCDelayRequirement_val * PixelClock / PixelClockBackEnd; |
1749 | |
1750 | } else { |
1751 | DSCDelayRequirement_val = 0; |
1752 | } |
1753 | |
1754 | #ifdef __DML_VBA_DEBUG__ |
1755 | dml_print("DML::%s: DSCEnabled = %d\n" , __func__, DSCEnabled); |
1756 | dml_print("DML::%s: OutputBpp = %f\n" , __func__, OutputBpp); |
1757 | dml_print("DML::%s: HActive = %d\n" , __func__, HActive); |
1758 | dml_print("DML::%s: OutputFormat = %d\n" , __func__, OutputFormat); |
1759 | dml_print("DML::%s: DSCInputBitPerComponent = %d\n" , __func__, DSCInputBitPerComponent); |
1760 | dml_print("DML::%s: NumberOfDSCSlices = %d\n" , __func__, NumberOfDSCSlices); |
1761 | dml_print("DML::%s: DSCDelayRequirement_val = %d\n" , __func__, DSCDelayRequirement_val); |
1762 | #endif |
1763 | |
1764 | return dml_ceil(a: DSCDelayRequirement_val * dsc_delay_factor_wa, granularity: 1); |
1765 | } |
1766 | |
1767 | void dml32_CalculateSurfaceSizeInMall( |
1768 | unsigned int NumberOfActiveSurfaces, |
1769 | unsigned int MALLAllocatedForDCN, |
1770 | enum dm_use_mall_for_static_screen_mode UseMALLForStaticScreen[], |
1771 | enum dm_use_mall_for_pstate_change_mode UsesMALLForPStateChange[], |
1772 | bool DCCEnable[], |
1773 | bool ViewportStationary[], |
1774 | unsigned int ViewportXStartY[], |
1775 | unsigned int ViewportYStartY[], |
1776 | unsigned int ViewportXStartC[], |
1777 | unsigned int ViewportYStartC[], |
1778 | unsigned int ViewportWidthY[], |
1779 | unsigned int ViewportHeightY[], |
1780 | unsigned int BytesPerPixelY[], |
1781 | unsigned int ViewportWidthC[], |
1782 | unsigned int ViewportHeightC[], |
1783 | unsigned int BytesPerPixelC[], |
1784 | unsigned int SurfaceWidthY[], |
1785 | unsigned int SurfaceWidthC[], |
1786 | unsigned int SurfaceHeightY[], |
1787 | unsigned int SurfaceHeightC[], |
1788 | unsigned int Read256BytesBlockWidthY[], |
1789 | unsigned int Read256BytesBlockWidthC[], |
1790 | unsigned int Read256BytesBlockHeightY[], |
1791 | unsigned int Read256BytesBlockHeightC[], |
1792 | unsigned int ReadBlockWidthY[], |
1793 | unsigned int ReadBlockWidthC[], |
1794 | unsigned int ReadBlockHeightY[], |
1795 | unsigned int ReadBlockHeightC[], |
1796 | unsigned int DCCMetaPitchY[], |
1797 | unsigned int DCCMetaPitchC[], |
1798 | |
1799 | /* Output */ |
1800 | unsigned int SurfaceSizeInMALL[], |
1801 | bool *ExceededMALLSize) |
1802 | { |
1803 | unsigned int k; |
1804 | unsigned int = 0; |
1805 | unsigned int TotalSurfaceSizeInMALLForSubVP = 0; |
1806 | unsigned int MALLAllocatedForDCNInBytes = MALLAllocatedForDCN * 1024 * 1024; |
1807 | |
1808 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1809 | if (ViewportStationary[k]) { |
1810 | SurfaceSizeInMALL[k] = dml_min(a: dml_ceil(a: SurfaceWidthY[k], granularity: ReadBlockWidthY[k]), |
1811 | b: dml_floor(a: ViewportXStartY[k] + ViewportWidthY[k] + ReadBlockWidthY[k] - 1, |
1812 | granularity: ReadBlockWidthY[k]) - dml_floor(a: ViewportXStartY[k], |
1813 | granularity: ReadBlockWidthY[k])) * dml_min(a: dml_ceil(a: SurfaceHeightY[k], |
1814 | granularity: ReadBlockHeightY[k]), b: dml_floor(a: ViewportYStartY[k] + |
1815 | ViewportHeightY[k] + ReadBlockHeightY[k] - 1, granularity: ReadBlockHeightY[k]) - |
1816 | dml_floor(a: ViewportYStartY[k], granularity: ReadBlockHeightY[k])) * BytesPerPixelY[k]; |
1817 | |
1818 | if (ReadBlockWidthC[k] > 0) { |
1819 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1820 | dml_min(a: dml_ceil(a: SurfaceWidthC[k], granularity: ReadBlockWidthC[k]), |
1821 | b: dml_floor(a: ViewportXStartC[k] + ViewportWidthC[k] + |
1822 | ReadBlockWidthC[k] - 1, granularity: ReadBlockWidthC[k]) - |
1823 | dml_floor(a: ViewportXStartC[k], granularity: ReadBlockWidthC[k])) * |
1824 | dml_min(a: dml_ceil(a: SurfaceHeightC[k], granularity: ReadBlockHeightC[k]), |
1825 | b: dml_floor(a: ViewportYStartC[k] + ViewportHeightC[k] + |
1826 | ReadBlockHeightC[k] - 1, granularity: ReadBlockHeightC[k]) - |
1827 | dml_floor(a: ViewportYStartC[k], granularity: ReadBlockHeightC[k])) * |
1828 | BytesPerPixelC[k]; |
1829 | } |
1830 | if (DCCEnable[k] == true) { |
1831 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1832 | (dml_min(a: dml_ceil(a: DCCMetaPitchY[k], granularity: 8 * Read256BytesBlockWidthY[k]), |
1833 | b: dml_floor(a: ViewportXStartY[k] + ViewportWidthY[k] + 8 * |
1834 | Read256BytesBlockWidthY[k] - 1, granularity: 8 * Read256BytesBlockWidthY[k]) |
1835 | - dml_floor(a: ViewportXStartY[k], granularity: 8 * Read256BytesBlockWidthY[k])) |
1836 | * dml_min(a: dml_ceil(a: SurfaceHeightY[k], granularity: 8 * |
1837 | Read256BytesBlockHeightY[k]), b: dml_floor(a: ViewportYStartY[k] + |
1838 | ViewportHeightY[k] + 8 * Read256BytesBlockHeightY[k] - 1, granularity: 8 * |
1839 | Read256BytesBlockHeightY[k]) - dml_floor(a: ViewportYStartY[k], granularity: 8 * |
1840 | Read256BytesBlockHeightY[k])) * BytesPerPixelY[k] / 256) + (64 * 1024); |
1841 | if (Read256BytesBlockWidthC[k] > 0) { |
1842 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1843 | dml_min(a: dml_ceil(a: DCCMetaPitchC[k], granularity: 8 * |
1844 | Read256BytesBlockWidthC[k]), |
1845 | b: dml_floor(a: ViewportXStartC[k] + ViewportWidthC[k] + 8 |
1846 | * Read256BytesBlockWidthC[k] - 1, granularity: 8 * |
1847 | Read256BytesBlockWidthC[k]) - |
1848 | dml_floor(a: ViewportXStartC[k], granularity: 8 * |
1849 | Read256BytesBlockWidthC[k])) * |
1850 | dml_min(a: dml_ceil(a: SurfaceHeightC[k], granularity: 8 * |
1851 | Read256BytesBlockHeightC[k]), |
1852 | b: dml_floor(a: ViewportYStartC[k] + ViewportHeightC[k] + |
1853 | 8 * Read256BytesBlockHeightC[k] - 1, granularity: 8 * |
1854 | Read256BytesBlockHeightC[k]) - |
1855 | dml_floor(a: ViewportYStartC[k], granularity: 8 * |
1856 | Read256BytesBlockHeightC[k])) * |
1857 | BytesPerPixelC[k] / 256; |
1858 | } |
1859 | } |
1860 | } else { |
1861 | SurfaceSizeInMALL[k] = dml_ceil(a: dml_min(a: SurfaceWidthY[k], b: ViewportWidthY[k] + |
1862 | ReadBlockWidthY[k] - 1), granularity: ReadBlockWidthY[k]) * |
1863 | dml_ceil(a: dml_min(a: SurfaceHeightY[k], b: ViewportHeightY[k] + |
1864 | ReadBlockHeightY[k] - 1), granularity: ReadBlockHeightY[k]) * |
1865 | BytesPerPixelY[k]; |
1866 | if (ReadBlockWidthC[k] > 0) { |
1867 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1868 | dml_ceil(a: dml_min(a: SurfaceWidthC[k], b: ViewportWidthC[k] + |
1869 | ReadBlockWidthC[k] - 1), granularity: ReadBlockWidthC[k]) * |
1870 | dml_ceil(a: dml_min(a: SurfaceHeightC[k], b: ViewportHeightC[k] + |
1871 | ReadBlockHeightC[k] - 1), granularity: ReadBlockHeightC[k]) * |
1872 | BytesPerPixelC[k]; |
1873 | } |
1874 | if (DCCEnable[k] == true) { |
1875 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1876 | (dml_ceil(a: dml_min(a: DCCMetaPitchY[k], b: ViewportWidthY[k] + 8 * |
1877 | Read256BytesBlockWidthY[k] - 1), granularity: 8 * |
1878 | Read256BytesBlockWidthY[k]) * |
1879 | dml_ceil(a: dml_min(a: SurfaceHeightY[k], b: ViewportHeightY[k] + 8 * |
1880 | Read256BytesBlockHeightY[k] - 1), granularity: 8 * |
1881 | Read256BytesBlockHeightY[k]) * BytesPerPixelY[k] / 256) + (64 * 1024); |
1882 | |
1883 | if (Read256BytesBlockWidthC[k] > 0) { |
1884 | SurfaceSizeInMALL[k] = SurfaceSizeInMALL[k] + |
1885 | dml_ceil(a: dml_min(a: DCCMetaPitchC[k], b: ViewportWidthC[k] + 8 * |
1886 | Read256BytesBlockWidthC[k] - 1), granularity: 8 * |
1887 | Read256BytesBlockWidthC[k]) * |
1888 | dml_ceil(a: dml_min(a: SurfaceHeightC[k], b: ViewportHeightC[k] + 8 * |
1889 | Read256BytesBlockHeightC[k] - 1), granularity: 8 * |
1890 | Read256BytesBlockHeightC[k]) * |
1891 | BytesPerPixelC[k] / 256; |
1892 | } |
1893 | } |
1894 | } |
1895 | } |
1896 | |
1897 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1898 | /* SS and Subvp counted separate as they are never used at the same time */ |
1899 | if (UsesMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) |
1900 | TotalSurfaceSizeInMALLForSubVP = TotalSurfaceSizeInMALLForSubVP + SurfaceSizeInMALL[k]; |
1901 | else if (UseMALLForStaticScreen[k] == dm_use_mall_static_screen_enable) |
1902 | TotalSurfaceSizeInMALLForSS = TotalSurfaceSizeInMALLForSS + SurfaceSizeInMALL[k]; |
1903 | } |
1904 | *ExceededMALLSize = (TotalSurfaceSizeInMALLForSS > MALLAllocatedForDCNInBytes) || |
1905 | (TotalSurfaceSizeInMALLForSubVP > MALLAllocatedForDCNInBytes); |
1906 | } // CalculateSurfaceSizeInMall |
1907 | |
1908 | void dml32_CalculateVMRowAndSwath( |
1909 | unsigned int NumberOfActiveSurfaces, |
1910 | DmlPipe myPipe[], |
1911 | unsigned int SurfaceSizeInMALL[], |
1912 | unsigned int PTEBufferSizeInRequestsLuma, |
1913 | unsigned int PTEBufferSizeInRequestsChroma, |
1914 | unsigned int DCCMetaBufferSizeBytes, |
1915 | enum dm_use_mall_for_static_screen_mode UseMALLForStaticScreen[], |
1916 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange[], |
1917 | unsigned int MALLAllocatedForDCN, |
1918 | double SwathWidthY[], |
1919 | double SwathWidthC[], |
1920 | bool GPUVMEnable, |
1921 | bool HostVMEnable, |
1922 | unsigned int HostVMMaxNonCachedPageTableLevels, |
1923 | unsigned int GPUVMMaxPageTableLevels, |
1924 | unsigned int GPUVMMinPageSizeKBytes[], |
1925 | unsigned int HostVMMinPageSize, |
1926 | |
1927 | /* Output */ |
1928 | bool PTEBufferSizeNotExceeded[], |
1929 | bool DCCMetaBufferSizeNotExceeded[], |
1930 | unsigned int dpte_row_width_luma_ub[], |
1931 | unsigned int dpte_row_width_chroma_ub[], |
1932 | unsigned int dpte_row_height_luma[], |
1933 | unsigned int dpte_row_height_chroma[], |
1934 | unsigned int dpte_row_height_linear_luma[], // VBA_DELTA |
1935 | unsigned int dpte_row_height_linear_chroma[], // VBA_DELTA |
1936 | unsigned int meta_req_width[], |
1937 | unsigned int meta_req_width_chroma[], |
1938 | unsigned int meta_req_height[], |
1939 | unsigned int meta_req_height_chroma[], |
1940 | unsigned int meta_row_width[], |
1941 | unsigned int meta_row_width_chroma[], |
1942 | unsigned int meta_row_height[], |
1943 | unsigned int meta_row_height_chroma[], |
1944 | unsigned int vm_group_bytes[], |
1945 | unsigned int dpte_group_bytes[], |
1946 | unsigned int PixelPTEReqWidthY[], |
1947 | unsigned int PixelPTEReqHeightY[], |
1948 | unsigned int PTERequestSizeY[], |
1949 | unsigned int PixelPTEReqWidthC[], |
1950 | unsigned int PixelPTEReqHeightC[], |
1951 | unsigned int PTERequestSizeC[], |
1952 | unsigned int dpde0_bytes_per_frame_ub_l[], |
1953 | unsigned int meta_pte_bytes_per_frame_ub_l[], |
1954 | unsigned int dpde0_bytes_per_frame_ub_c[], |
1955 | unsigned int meta_pte_bytes_per_frame_ub_c[], |
1956 | double PrefetchSourceLinesY[], |
1957 | double PrefetchSourceLinesC[], |
1958 | double VInitPreFillY[], |
1959 | double VInitPreFillC[], |
1960 | unsigned int MaxNumSwathY[], |
1961 | unsigned int MaxNumSwathC[], |
1962 | double meta_row_bw[], |
1963 | double dpte_row_bw[], |
1964 | double PixelPTEBytesPerRow[], |
1965 | double PDEAndMetaPTEBytesFrame[], |
1966 | double MetaRowByte[], |
1967 | bool use_one_row_for_frame[], |
1968 | bool use_one_row_for_frame_flip[], |
1969 | bool UsesMALLForStaticScreen[], |
1970 | bool PTE_BUFFER_MODE[], |
1971 | unsigned int BIGK_FRAGMENT_SIZE[]) |
1972 | { |
1973 | unsigned int k; |
1974 | unsigned int PTEBufferSizeInRequestsForLuma[DC__NUM_DPP__MAX]; |
1975 | unsigned int PTEBufferSizeInRequestsForChroma[DC__NUM_DPP__MAX]; |
1976 | unsigned int PDEAndMetaPTEBytesFrameY; |
1977 | unsigned int PDEAndMetaPTEBytesFrameC; |
1978 | unsigned int MetaRowByteY[DC__NUM_DPP__MAX]; |
1979 | unsigned int MetaRowByteC[DC__NUM_DPP__MAX]; |
1980 | unsigned int PixelPTEBytesPerRowY[DC__NUM_DPP__MAX]; |
1981 | unsigned int PixelPTEBytesPerRowC[DC__NUM_DPP__MAX]; |
1982 | unsigned int PixelPTEBytesPerRowY_one_row_per_frame[DC__NUM_DPP__MAX]; |
1983 | unsigned int PixelPTEBytesPerRowC_one_row_per_frame[DC__NUM_DPP__MAX]; |
1984 | unsigned int dpte_row_width_luma_ub_one_row_per_frame[DC__NUM_DPP__MAX]; |
1985 | unsigned int dpte_row_height_luma_one_row_per_frame[DC__NUM_DPP__MAX]; |
1986 | unsigned int dpte_row_width_chroma_ub_one_row_per_frame[DC__NUM_DPP__MAX]; |
1987 | unsigned int dpte_row_height_chroma_one_row_per_frame[DC__NUM_DPP__MAX]; |
1988 | bool one_row_per_frame_fits_in_buffer[DC__NUM_DPP__MAX]; |
1989 | |
1990 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
1991 | if (HostVMEnable == true) { |
1992 | vm_group_bytes[k] = 512; |
1993 | dpte_group_bytes[k] = 512; |
1994 | } else if (GPUVMEnable == true) { |
1995 | vm_group_bytes[k] = 2048; |
1996 | if (GPUVMMinPageSizeKBytes[k] >= 64 && IsVertical(Scan: myPipe[k].SourceRotation)) |
1997 | dpte_group_bytes[k] = 512; |
1998 | else |
1999 | dpte_group_bytes[k] = 2048; |
2000 | } else { |
2001 | vm_group_bytes[k] = 0; |
2002 | dpte_group_bytes[k] = 0; |
2003 | } |
2004 | |
2005 | if (myPipe[k].SourcePixelFormat == dm_420_8 || myPipe[k].SourcePixelFormat == dm_420_10 || |
2006 | myPipe[k].SourcePixelFormat == dm_420_12 || |
2007 | myPipe[k].SourcePixelFormat == dm_rgbe_alpha) { |
2008 | if ((myPipe[k].SourcePixelFormat == dm_420_10 || myPipe[k].SourcePixelFormat == dm_420_12) && |
2009 | !IsVertical(Scan: myPipe[k].SourceRotation)) { |
2010 | PTEBufferSizeInRequestsForLuma[k] = |
2011 | (PTEBufferSizeInRequestsLuma + PTEBufferSizeInRequestsChroma) / 2; |
2012 | PTEBufferSizeInRequestsForChroma[k] = PTEBufferSizeInRequestsForLuma[k]; |
2013 | } else { |
2014 | PTEBufferSizeInRequestsForLuma[k] = PTEBufferSizeInRequestsLuma; |
2015 | PTEBufferSizeInRequestsForChroma[k] = PTEBufferSizeInRequestsChroma; |
2016 | } |
2017 | |
2018 | PDEAndMetaPTEBytesFrameC = dml32_CalculateVMAndRowBytes( |
2019 | ViewportStationary: myPipe[k].ViewportStationary, |
2020 | DCCEnable: myPipe[k].DCCEnable, |
2021 | NumberOfDPPs: myPipe[k].DPPPerSurface, |
2022 | BlockHeight256Bytes: myPipe[k].BlockHeight256BytesC, |
2023 | BlockWidth256Bytes: myPipe[k].BlockWidth256BytesC, |
2024 | SourcePixelFormat: myPipe[k].SourcePixelFormat, |
2025 | SurfaceTiling: myPipe[k].SurfaceTiling, |
2026 | BytePerPixel: myPipe[k].BytePerPixelC, |
2027 | SourceScan: myPipe[k].SourceRotation, |
2028 | SwathWidth: SwathWidthC[k], |
2029 | ViewportHeight: myPipe[k].ViewportHeightChroma, |
2030 | ViewportXStart: myPipe[k].ViewportXStartC, |
2031 | ViewportYStart: myPipe[k].ViewportYStartC, |
2032 | GPUVMEnable, |
2033 | HostVMEnable, |
2034 | HostVMMaxNonCachedPageTableLevels, |
2035 | GPUVMMaxPageTableLevels, |
2036 | GPUVMMinPageSizeKBytes: GPUVMMinPageSizeKBytes[k], |
2037 | HostVMMinPageSize, |
2038 | PTEBufferSizeInRequests: PTEBufferSizeInRequestsForChroma[k], |
2039 | Pitch: myPipe[k].PitchC, |
2040 | DCCMetaPitch: myPipe[k].DCCMetaPitchC, |
2041 | MacroTileWidth: myPipe[k].BlockWidthC, |
2042 | MacroTileHeight: myPipe[k].BlockHeightC, |
2043 | |
2044 | /* Output */ |
2045 | MetaRowByte: &MetaRowByteC[k], |
2046 | PixelPTEBytesPerRow: &PixelPTEBytesPerRowC[k], |
2047 | dpte_row_width_ub: &dpte_row_width_chroma_ub[k], |
2048 | dpte_row_height: &dpte_row_height_chroma[k], |
2049 | dpte_row_height_linear: &dpte_row_height_linear_chroma[k], |
2050 | PixelPTEBytesPerRow_one_row_per_frame: &PixelPTEBytesPerRowC_one_row_per_frame[k], |
2051 | dpte_row_width_ub_one_row_per_frame: &dpte_row_width_chroma_ub_one_row_per_frame[k], |
2052 | dpte_row_height_one_row_per_frame: &dpte_row_height_chroma_one_row_per_frame[k], |
2053 | MetaRequestWidth: &meta_req_width_chroma[k], |
2054 | MetaRequestHeight: &meta_req_height_chroma[k], |
2055 | meta_row_width: &meta_row_width_chroma[k], |
2056 | meta_row_height: &meta_row_height_chroma[k], |
2057 | PixelPTEReqWidth: &PixelPTEReqWidthC[k], |
2058 | PixelPTEReqHeight: &PixelPTEReqHeightC[k], |
2059 | PTERequestSize: &PTERequestSizeC[k], |
2060 | DPDE0BytesFrame: &dpde0_bytes_per_frame_ub_c[k], |
2061 | MetaPTEBytesFrame: &meta_pte_bytes_per_frame_ub_c[k]); |
2062 | |
2063 | PrefetchSourceLinesC[k] = dml32_CalculatePrefetchSourceLines( |
2064 | VRatio: myPipe[k].VRatioChroma, |
2065 | VTaps: myPipe[k].VTapsChroma, |
2066 | Interlace: myPipe[k].InterlaceEnable, |
2067 | ProgressiveToInterlaceUnitInOPP: myPipe[k].ProgressiveToInterlaceUnitInOPP, |
2068 | SwathHeight: myPipe[k].SwathHeightC, |
2069 | SourceRotation: myPipe[k].SourceRotation, |
2070 | ViewportStationary: myPipe[k].ViewportStationary, |
2071 | SwathWidth: SwathWidthC[k], |
2072 | ViewportHeight: myPipe[k].ViewportHeightChroma, |
2073 | ViewportXStart: myPipe[k].ViewportXStartC, |
2074 | ViewportYStart: myPipe[k].ViewportYStartC, |
2075 | |
2076 | /* Output */ |
2077 | VInitPreFill: &VInitPreFillC[k], |
2078 | MaxNumSwath: &MaxNumSwathC[k]); |
2079 | } else { |
2080 | PTEBufferSizeInRequestsForLuma[k] = PTEBufferSizeInRequestsLuma + PTEBufferSizeInRequestsChroma; |
2081 | PTEBufferSizeInRequestsForChroma[k] = 0; |
2082 | PixelPTEBytesPerRowC[k] = 0; |
2083 | PDEAndMetaPTEBytesFrameC = 0; |
2084 | MetaRowByteC[k] = 0; |
2085 | MaxNumSwathC[k] = 0; |
2086 | PrefetchSourceLinesC[k] = 0; |
2087 | dpte_row_height_chroma_one_row_per_frame[k] = 0; |
2088 | dpte_row_width_chroma_ub_one_row_per_frame[k] = 0; |
2089 | PixelPTEBytesPerRowC_one_row_per_frame[k] = 0; |
2090 | } |
2091 | |
2092 | PDEAndMetaPTEBytesFrameY = dml32_CalculateVMAndRowBytes( |
2093 | ViewportStationary: myPipe[k].ViewportStationary, |
2094 | DCCEnable: myPipe[k].DCCEnable, |
2095 | NumberOfDPPs: myPipe[k].DPPPerSurface, |
2096 | BlockHeight256Bytes: myPipe[k].BlockHeight256BytesY, |
2097 | BlockWidth256Bytes: myPipe[k].BlockWidth256BytesY, |
2098 | SourcePixelFormat: myPipe[k].SourcePixelFormat, |
2099 | SurfaceTiling: myPipe[k].SurfaceTiling, |
2100 | BytePerPixel: myPipe[k].BytePerPixelY, |
2101 | SourceScan: myPipe[k].SourceRotation, |
2102 | SwathWidth: SwathWidthY[k], |
2103 | ViewportHeight: myPipe[k].ViewportHeight, |
2104 | ViewportXStart: myPipe[k].ViewportXStart, |
2105 | ViewportYStart: myPipe[k].ViewportYStart, |
2106 | GPUVMEnable, |
2107 | HostVMEnable, |
2108 | HostVMMaxNonCachedPageTableLevels, |
2109 | GPUVMMaxPageTableLevels, |
2110 | GPUVMMinPageSizeKBytes: GPUVMMinPageSizeKBytes[k], |
2111 | HostVMMinPageSize, |
2112 | PTEBufferSizeInRequests: PTEBufferSizeInRequestsForLuma[k], |
2113 | Pitch: myPipe[k].PitchY, |
2114 | DCCMetaPitch: myPipe[k].DCCMetaPitchY, |
2115 | MacroTileWidth: myPipe[k].BlockWidthY, |
2116 | MacroTileHeight: myPipe[k].BlockHeightY, |
2117 | |
2118 | /* Output */ |
2119 | MetaRowByte: &MetaRowByteY[k], |
2120 | PixelPTEBytesPerRow: &PixelPTEBytesPerRowY[k], |
2121 | dpte_row_width_ub: &dpte_row_width_luma_ub[k], |
2122 | dpte_row_height: &dpte_row_height_luma[k], |
2123 | dpte_row_height_linear: &dpte_row_height_linear_luma[k], |
2124 | PixelPTEBytesPerRow_one_row_per_frame: &PixelPTEBytesPerRowY_one_row_per_frame[k], |
2125 | dpte_row_width_ub_one_row_per_frame: &dpte_row_width_luma_ub_one_row_per_frame[k], |
2126 | dpte_row_height_one_row_per_frame: &dpte_row_height_luma_one_row_per_frame[k], |
2127 | MetaRequestWidth: &meta_req_width[k], |
2128 | MetaRequestHeight: &meta_req_height[k], |
2129 | meta_row_width: &meta_row_width[k], |
2130 | meta_row_height: &meta_row_height[k], |
2131 | PixelPTEReqWidth: &PixelPTEReqWidthY[k], |
2132 | PixelPTEReqHeight: &PixelPTEReqHeightY[k], |
2133 | PTERequestSize: &PTERequestSizeY[k], |
2134 | DPDE0BytesFrame: &dpde0_bytes_per_frame_ub_l[k], |
2135 | MetaPTEBytesFrame: &meta_pte_bytes_per_frame_ub_l[k]); |
2136 | |
2137 | PrefetchSourceLinesY[k] = dml32_CalculatePrefetchSourceLines( |
2138 | VRatio: myPipe[k].VRatio, |
2139 | VTaps: myPipe[k].VTaps, |
2140 | Interlace: myPipe[k].InterlaceEnable, |
2141 | ProgressiveToInterlaceUnitInOPP: myPipe[k].ProgressiveToInterlaceUnitInOPP, |
2142 | SwathHeight: myPipe[k].SwathHeightY, |
2143 | SourceRotation: myPipe[k].SourceRotation, |
2144 | ViewportStationary: myPipe[k].ViewportStationary, |
2145 | SwathWidth: SwathWidthY[k], |
2146 | ViewportHeight: myPipe[k].ViewportHeight, |
2147 | ViewportXStart: myPipe[k].ViewportXStart, |
2148 | ViewportYStart: myPipe[k].ViewportYStart, |
2149 | |
2150 | /* Output */ |
2151 | VInitPreFill: &VInitPreFillY[k], |
2152 | MaxNumSwath: &MaxNumSwathY[k]); |
2153 | |
2154 | PDEAndMetaPTEBytesFrame[k] = PDEAndMetaPTEBytesFrameY + PDEAndMetaPTEBytesFrameC; |
2155 | MetaRowByte[k] = MetaRowByteY[k] + MetaRowByteC[k]; |
2156 | |
2157 | if (PixelPTEBytesPerRowY[k] <= 64 * PTEBufferSizeInRequestsForLuma[k] && |
2158 | PixelPTEBytesPerRowC[k] <= 64 * PTEBufferSizeInRequestsForChroma[k]) { |
2159 | PTEBufferSizeNotExceeded[k] = true; |
2160 | } else { |
2161 | PTEBufferSizeNotExceeded[k] = false; |
2162 | } |
2163 | |
2164 | one_row_per_frame_fits_in_buffer[k] = (PixelPTEBytesPerRowY_one_row_per_frame[k] <= 64 * 2 * |
2165 | PTEBufferSizeInRequestsForLuma[k] && |
2166 | PixelPTEBytesPerRowC_one_row_per_frame[k] <= 64 * 2 * PTEBufferSizeInRequestsForChroma[k]); |
2167 | } |
2168 | |
2169 | dml32_CalculateMALLUseForStaticScreen( |
2170 | NumberOfActiveSurfaces, |
2171 | MALLAllocatedForDCNFinal: MALLAllocatedForDCN, |
2172 | UseMALLForStaticScreen, // mode |
2173 | SurfaceSizeInMALL, |
2174 | one_row_per_frame_fits_in_buffer, |
2175 | /* Output */ |
2176 | UsesMALLForStaticScreen); // boolen |
2177 | |
2178 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
2179 | PTE_BUFFER_MODE[k] = myPipe[k].FORCE_ONE_ROW_FOR_FRAME || UsesMALLForStaticScreen[k] || |
2180 | (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_sub_viewport) || |
2181 | (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) || |
2182 | (GPUVMMinPageSizeKBytes[k] > 64); |
2183 | BIGK_FRAGMENT_SIZE[k] = dml_log2(x: GPUVMMinPageSizeKBytes[k] * 1024) - 12; |
2184 | } |
2185 | |
2186 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
2187 | #ifdef __DML_VBA_DEBUG__ |
2188 | dml_print("DML::%s: k=%d, SurfaceSizeInMALL = %d\n" , __func__, k, SurfaceSizeInMALL[k]); |
2189 | dml_print("DML::%s: k=%d, UsesMALLForStaticScreen = %d\n" , __func__, k, UsesMALLForStaticScreen[k]); |
2190 | #endif |
2191 | use_one_row_for_frame[k] = myPipe[k].FORCE_ONE_ROW_FOR_FRAME || UsesMALLForStaticScreen[k] || |
2192 | (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_sub_viewport) || |
2193 | (UseMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) || |
2194 | (GPUVMMinPageSizeKBytes[k] > 64 && IsVertical(Scan: myPipe[k].SourceRotation)); |
2195 | |
2196 | use_one_row_for_frame_flip[k] = use_one_row_for_frame[k] && |
2197 | !(UseMALLForPStateChange[k] == dm_use_mall_pstate_change_full_frame); |
2198 | |
2199 | if (use_one_row_for_frame[k]) { |
2200 | dpte_row_height_luma[k] = dpte_row_height_luma_one_row_per_frame[k]; |
2201 | dpte_row_width_luma_ub[k] = dpte_row_width_luma_ub_one_row_per_frame[k]; |
2202 | PixelPTEBytesPerRowY[k] = PixelPTEBytesPerRowY_one_row_per_frame[k]; |
2203 | dpte_row_height_chroma[k] = dpte_row_height_chroma_one_row_per_frame[k]; |
2204 | dpte_row_width_chroma_ub[k] = dpte_row_width_chroma_ub_one_row_per_frame[k]; |
2205 | PixelPTEBytesPerRowC[k] = PixelPTEBytesPerRowC_one_row_per_frame[k]; |
2206 | PTEBufferSizeNotExceeded[k] = one_row_per_frame_fits_in_buffer[k]; |
2207 | } |
2208 | |
2209 | if (MetaRowByte[k] <= DCCMetaBufferSizeBytes) |
2210 | DCCMetaBufferSizeNotExceeded[k] = true; |
2211 | else |
2212 | DCCMetaBufferSizeNotExceeded[k] = false; |
2213 | |
2214 | PixelPTEBytesPerRow[k] = PixelPTEBytesPerRowY[k] + PixelPTEBytesPerRowC[k]; |
2215 | if (use_one_row_for_frame[k]) |
2216 | PixelPTEBytesPerRow[k] = PixelPTEBytesPerRow[k] / 2; |
2217 | |
2218 | dml32_CalculateRowBandwidth( |
2219 | GPUVMEnable, |
2220 | SourcePixelFormat: myPipe[k].SourcePixelFormat, |
2221 | VRatio: myPipe[k].VRatio, |
2222 | VRatioChroma: myPipe[k].VRatioChroma, |
2223 | DCCEnable: myPipe[k].DCCEnable, |
2224 | LineTime: myPipe[k].HTotal / myPipe[k].PixelClock, |
2225 | MetaRowByteLuma: MetaRowByteY[k], MetaRowByteChroma: MetaRowByteC[k], |
2226 | meta_row_height_luma: meta_row_height[k], |
2227 | meta_row_height_chroma: meta_row_height_chroma[k], |
2228 | PixelPTEBytesPerRowLuma: PixelPTEBytesPerRowY[k], |
2229 | PixelPTEBytesPerRowChroma: PixelPTEBytesPerRowC[k], |
2230 | dpte_row_height_luma: dpte_row_height_luma[k], |
2231 | dpte_row_height_chroma: dpte_row_height_chroma[k], |
2232 | |
2233 | /* Output */ |
2234 | meta_row_bw: &meta_row_bw[k], |
2235 | dpte_row_bw: &dpte_row_bw[k]); |
2236 | #ifdef __DML_VBA_DEBUG__ |
2237 | dml_print("DML::%s: k=%d, use_one_row_for_frame = %d\n" , __func__, k, use_one_row_for_frame[k]); |
2238 | dml_print("DML::%s: k=%d, use_one_row_for_frame_flip = %d\n" , |
2239 | __func__, k, use_one_row_for_frame_flip[k]); |
2240 | dml_print("DML::%s: k=%d, UseMALLForPStateChange = %d\n" , |
2241 | __func__, k, UseMALLForPStateChange[k]); |
2242 | dml_print("DML::%s: k=%d, dpte_row_height_luma = %d\n" , __func__, k, dpte_row_height_luma[k]); |
2243 | dml_print("DML::%s: k=%d, dpte_row_width_luma_ub = %d\n" , |
2244 | __func__, k, dpte_row_width_luma_ub[k]); |
2245 | dml_print("DML::%s: k=%d, PixelPTEBytesPerRowY = %d\n" , __func__, k, PixelPTEBytesPerRowY[k]); |
2246 | dml_print("DML::%s: k=%d, dpte_row_height_chroma = %d\n" , |
2247 | __func__, k, dpte_row_height_chroma[k]); |
2248 | dml_print("DML::%s: k=%d, dpte_row_width_chroma_ub = %d\n" , |
2249 | __func__, k, dpte_row_width_chroma_ub[k]); |
2250 | dml_print("DML::%s: k=%d, PixelPTEBytesPerRowC = %d\n" , __func__, k, PixelPTEBytesPerRowC[k]); |
2251 | dml_print("DML::%s: k=%d, PixelPTEBytesPerRow = %d\n" , __func__, k, PixelPTEBytesPerRow[k]); |
2252 | dml_print("DML::%s: k=%d, PTEBufferSizeNotExceeded = %d\n" , |
2253 | __func__, k, PTEBufferSizeNotExceeded[k]); |
2254 | dml_print("DML::%s: k=%d, PTE_BUFFER_MODE = %d\n" , __func__, k, PTE_BUFFER_MODE[k]); |
2255 | dml_print("DML::%s: k=%d, BIGK_FRAGMENT_SIZE = %d\n" , __func__, k, BIGK_FRAGMENT_SIZE[k]); |
2256 | #endif |
2257 | } |
2258 | } // CalculateVMRowAndSwath |
2259 | |
2260 | unsigned int dml32_CalculateVMAndRowBytes( |
2261 | bool ViewportStationary, |
2262 | bool DCCEnable, |
2263 | unsigned int NumberOfDPPs, |
2264 | unsigned int BlockHeight256Bytes, |
2265 | unsigned int BlockWidth256Bytes, |
2266 | enum source_format_class SourcePixelFormat, |
2267 | unsigned int SurfaceTiling, |
2268 | unsigned int BytePerPixel, |
2269 | enum dm_rotation_angle SourceRotation, |
2270 | double SwathWidth, |
2271 | unsigned int ViewportHeight, |
2272 | unsigned int ViewportXStart, |
2273 | unsigned int ViewportYStart, |
2274 | bool GPUVMEnable, |
2275 | bool HostVMEnable, |
2276 | unsigned int HostVMMaxNonCachedPageTableLevels, |
2277 | unsigned int GPUVMMaxPageTableLevels, |
2278 | unsigned int GPUVMMinPageSizeKBytes, |
2279 | unsigned int HostVMMinPageSize, |
2280 | unsigned int PTEBufferSizeInRequests, |
2281 | unsigned int Pitch, |
2282 | unsigned int DCCMetaPitch, |
2283 | unsigned int MacroTileWidth, |
2284 | unsigned int MacroTileHeight, |
2285 | |
2286 | /* Output */ |
2287 | unsigned int *MetaRowByte, |
2288 | unsigned int *PixelPTEBytesPerRow, |
2289 | unsigned int *dpte_row_width_ub, |
2290 | unsigned int *dpte_row_height, |
2291 | unsigned int *dpte_row_height_linear, |
2292 | unsigned int *PixelPTEBytesPerRow_one_row_per_frame, |
2293 | unsigned int *dpte_row_width_ub_one_row_per_frame, |
2294 | unsigned int *dpte_row_height_one_row_per_frame, |
2295 | unsigned int *MetaRequestWidth, |
2296 | unsigned int *MetaRequestHeight, |
2297 | unsigned int *meta_row_width, |
2298 | unsigned int *meta_row_height, |
2299 | unsigned int *PixelPTEReqWidth, |
2300 | unsigned int *PixelPTEReqHeight, |
2301 | unsigned int *PTERequestSize, |
2302 | unsigned int *DPDE0BytesFrame, |
2303 | unsigned int *MetaPTEBytesFrame) |
2304 | { |
2305 | unsigned int MPDEBytesFrame; |
2306 | unsigned int DCCMetaSurfaceBytes; |
2307 | unsigned int ; |
2308 | unsigned int PDEAndMetaPTEBytesFrame; |
2309 | unsigned int HostVMDynamicLevels = 0; |
2310 | unsigned int MacroTileSizeBytes; |
2311 | unsigned int vp_height_meta_ub; |
2312 | unsigned int vp_height_dpte_ub; |
2313 | unsigned int PixelPTEReqWidth_linear = 0; // VBA_DELTA. VBA doesn't calculate this |
2314 | |
2315 | if (GPUVMEnable == true && HostVMEnable == true) { |
2316 | if (HostVMMinPageSize < 2048) |
2317 | HostVMDynamicLevels = HostVMMaxNonCachedPageTableLevels; |
2318 | else if (HostVMMinPageSize >= 2048 && HostVMMinPageSize < 1048576) |
2319 | HostVMDynamicLevels = dml_max(a: 0, b: (int) HostVMMaxNonCachedPageTableLevels - 1); |
2320 | else |
2321 | HostVMDynamicLevels = dml_max(a: 0, b: (int) HostVMMaxNonCachedPageTableLevels - 2); |
2322 | } |
2323 | |
2324 | *MetaRequestHeight = 8 * BlockHeight256Bytes; |
2325 | *MetaRequestWidth = 8 * BlockWidth256Bytes; |
2326 | if (SurfaceTiling == dm_sw_linear) { |
2327 | *meta_row_height = 32; |
2328 | *meta_row_width = dml_floor(a: ViewportXStart + SwathWidth + *MetaRequestWidth - 1, granularity: *MetaRequestWidth) |
2329 | - dml_floor(a: ViewportXStart, granularity: *MetaRequestWidth); |
2330 | } else if (!IsVertical(Scan: SourceRotation)) { |
2331 | *meta_row_height = *MetaRequestHeight; |
2332 | if (ViewportStationary && NumberOfDPPs == 1) { |
2333 | *meta_row_width = dml_floor(a: ViewportXStart + SwathWidth + *MetaRequestWidth - 1, |
2334 | granularity: *MetaRequestWidth) - dml_floor(a: ViewportXStart, granularity: *MetaRequestWidth); |
2335 | } else { |
2336 | *meta_row_width = dml_ceil(a: SwathWidth - 1, granularity: *MetaRequestWidth) + *MetaRequestWidth; |
2337 | } |
2338 | *MetaRowByte = *meta_row_width * *MetaRequestHeight * BytePerPixel / 256.0; |
2339 | } else { |
2340 | *meta_row_height = *MetaRequestWidth; |
2341 | if (ViewportStationary && NumberOfDPPs == 1) { |
2342 | *meta_row_width = dml_floor(a: ViewportYStart + ViewportHeight + *MetaRequestHeight - 1, |
2343 | granularity: *MetaRequestHeight) - dml_floor(a: ViewportYStart, granularity: *MetaRequestHeight); |
2344 | } else { |
2345 | *meta_row_width = dml_ceil(a: SwathWidth - 1, granularity: *MetaRequestHeight) + *MetaRequestHeight; |
2346 | } |
2347 | *MetaRowByte = *meta_row_width * *MetaRequestWidth * BytePerPixel / 256.0; |
2348 | } |
2349 | |
2350 | if (ViewportStationary && (NumberOfDPPs == 1 || !IsVertical(Scan: SourceRotation))) { |
2351 | vp_height_meta_ub = dml_floor(a: ViewportYStart + ViewportHeight + 64 * BlockHeight256Bytes - 1, |
2352 | granularity: 64 * BlockHeight256Bytes) - dml_floor(a: ViewportYStart, granularity: 64 * BlockHeight256Bytes); |
2353 | } else if (!IsVertical(Scan: SourceRotation)) { |
2354 | vp_height_meta_ub = dml_ceil(a: ViewportHeight - 1, granularity: 64 * BlockHeight256Bytes) + 64 * BlockHeight256Bytes; |
2355 | } else { |
2356 | vp_height_meta_ub = dml_ceil(a: SwathWidth - 1, granularity: 64 * BlockHeight256Bytes) + 64 * BlockHeight256Bytes; |
2357 | } |
2358 | |
2359 | DCCMetaSurfaceBytes = DCCMetaPitch * vp_height_meta_ub * BytePerPixel / 256.0; |
2360 | |
2361 | if (GPUVMEnable == true) { |
2362 | *MetaPTEBytesFrame = (dml_ceil(a: (double) (DCCMetaSurfaceBytes - 4.0 * 1024.0) / |
2363 | (8 * 4.0 * 1024), granularity: 1) + 1) * 64; |
2364 | MPDEBytesFrame = 128 * (GPUVMMaxPageTableLevels - 1); |
2365 | } else { |
2366 | *MetaPTEBytesFrame = 0; |
2367 | MPDEBytesFrame = 0; |
2368 | } |
2369 | |
2370 | if (DCCEnable != true) { |
2371 | *MetaPTEBytesFrame = 0; |
2372 | MPDEBytesFrame = 0; |
2373 | *MetaRowByte = 0; |
2374 | } |
2375 | |
2376 | MacroTileSizeBytes = MacroTileWidth * BytePerPixel * MacroTileHeight; |
2377 | |
2378 | if (GPUVMEnable == true && GPUVMMaxPageTableLevels > 1) { |
2379 | if (ViewportStationary && (NumberOfDPPs == 1 || !IsVertical(Scan: SourceRotation))) { |
2380 | vp_height_dpte_ub = dml_floor(a: ViewportYStart + ViewportHeight + |
2381 | MacroTileHeight - 1, granularity: MacroTileHeight) - |
2382 | dml_floor(a: ViewportYStart, granularity: MacroTileHeight); |
2383 | } else if (!IsVertical(Scan: SourceRotation)) { |
2384 | vp_height_dpte_ub = dml_ceil(a: ViewportHeight - 1, granularity: MacroTileHeight) + MacroTileHeight; |
2385 | } else { |
2386 | vp_height_dpte_ub = dml_ceil(a: SwathWidth - 1, granularity: MacroTileHeight) + MacroTileHeight; |
2387 | } |
2388 | *DPDE0BytesFrame = 64 * (dml_ceil(a: (Pitch * vp_height_dpte_ub * BytePerPixel - MacroTileSizeBytes) / |
2389 | (8 * 2097152), granularity: 1) + 1); |
2390 | ExtraDPDEBytesFrame = 128 * (GPUVMMaxPageTableLevels - 2); |
2391 | } else { |
2392 | *DPDE0BytesFrame = 0; |
2393 | ExtraDPDEBytesFrame = 0; |
2394 | vp_height_dpte_ub = 0; |
2395 | } |
2396 | |
2397 | PDEAndMetaPTEBytesFrame = *MetaPTEBytesFrame + MPDEBytesFrame + *DPDE0BytesFrame + ExtraDPDEBytesFrame; |
2398 | |
2399 | #ifdef __DML_VBA_DEBUG__ |
2400 | dml_print("DML::%s: DCCEnable = %d\n" , __func__, DCCEnable); |
2401 | dml_print("DML::%s: GPUVMEnable = %d\n" , __func__, GPUVMEnable); |
2402 | dml_print("DML::%s: SwModeLinear = %d\n" , __func__, SurfaceTiling == dm_sw_linear); |
2403 | dml_print("DML::%s: BytePerPixel = %d\n" , __func__, BytePerPixel); |
2404 | dml_print("DML::%s: GPUVMMaxPageTableLevels = %d\n" , __func__, GPUVMMaxPageTableLevels); |
2405 | dml_print("DML::%s: BlockHeight256Bytes = %d\n" , __func__, BlockHeight256Bytes); |
2406 | dml_print("DML::%s: BlockWidth256Bytes = %d\n" , __func__, BlockWidth256Bytes); |
2407 | dml_print("DML::%s: MacroTileHeight = %d\n" , __func__, MacroTileHeight); |
2408 | dml_print("DML::%s: MacroTileWidth = %d\n" , __func__, MacroTileWidth); |
2409 | dml_print("DML::%s: MetaPTEBytesFrame = %d\n" , __func__, *MetaPTEBytesFrame); |
2410 | dml_print("DML::%s: MPDEBytesFrame = %d\n" , __func__, MPDEBytesFrame); |
2411 | dml_print("DML::%s: DPDE0BytesFrame = %d\n" , __func__, *DPDE0BytesFrame); |
2412 | dml_print("DML::%s: ExtraDPDEBytesFrame= %d\n" , __func__, ExtraDPDEBytesFrame); |
2413 | dml_print("DML::%s: PDEAndMetaPTEBytesFrame = %d\n" , __func__, PDEAndMetaPTEBytesFrame); |
2414 | dml_print("DML::%s: ViewportHeight = %d\n" , __func__, ViewportHeight); |
2415 | dml_print("DML::%s: SwathWidth = %d\n" , __func__, SwathWidth); |
2416 | dml_print("DML::%s: vp_height_dpte_ub = %d\n" , __func__, vp_height_dpte_ub); |
2417 | #endif |
2418 | |
2419 | if (HostVMEnable == true) |
2420 | PDEAndMetaPTEBytesFrame = PDEAndMetaPTEBytesFrame * (1 + 8 * HostVMDynamicLevels); |
2421 | |
2422 | if (SurfaceTiling == dm_sw_linear) { |
2423 | *PixelPTEReqHeight = 1; |
2424 | *PixelPTEReqWidth = GPUVMMinPageSizeKBytes * 1024 * 8 / BytePerPixel; |
2425 | PixelPTEReqWidth_linear = GPUVMMinPageSizeKBytes * 1024 * 8 / BytePerPixel; |
2426 | *PTERequestSize = 64; |
2427 | } else if (GPUVMMinPageSizeKBytes == 4) { |
2428 | *PixelPTEReqHeight = 16 * BlockHeight256Bytes; |
2429 | *PixelPTEReqWidth = 16 * BlockWidth256Bytes; |
2430 | *PTERequestSize = 128; |
2431 | } else { |
2432 | *PixelPTEReqHeight = MacroTileHeight; |
2433 | *PixelPTEReqWidth = 8 * 1024 * GPUVMMinPageSizeKBytes / (MacroTileHeight * BytePerPixel); |
2434 | *PTERequestSize = 64; |
2435 | } |
2436 | #ifdef __DML_VBA_DEBUG__ |
2437 | dml_print("DML::%s: GPUVMMinPageSizeKBytes = %d\n" , __func__, GPUVMMinPageSizeKBytes); |
2438 | dml_print("DML::%s: PDEAndMetaPTEBytesFrame = %d (after HostVM factor)\n" , __func__, PDEAndMetaPTEBytesFrame); |
2439 | dml_print("DML::%s: PixelPTEReqHeight = %d\n" , __func__, *PixelPTEReqHeight); |
2440 | dml_print("DML::%s: PixelPTEReqWidth = %d\n" , __func__, *PixelPTEReqWidth); |
2441 | dml_print("DML::%s: PixelPTEReqWidth_linear = %d\n" , __func__, PixelPTEReqWidth_linear); |
2442 | dml_print("DML::%s: PTERequestSize = %d\n" , __func__, *PTERequestSize); |
2443 | dml_print("DML::%s: Pitch = %d\n" , __func__, Pitch); |
2444 | #endif |
2445 | |
2446 | *dpte_row_height_one_row_per_frame = vp_height_dpte_ub; |
2447 | *dpte_row_width_ub_one_row_per_frame = (dml_ceil(a: ((double)Pitch * (double)*dpte_row_height_one_row_per_frame / |
2448 | (double) *PixelPTEReqHeight - 1) / (double) *PixelPTEReqWidth, granularity: 1) + 1) * |
2449 | (double) *PixelPTEReqWidth; |
2450 | *PixelPTEBytesPerRow_one_row_per_frame = *dpte_row_width_ub_one_row_per_frame / *PixelPTEReqWidth * |
2451 | *PTERequestSize; |
2452 | |
2453 | if (SurfaceTiling == dm_sw_linear) { |
2454 | *dpte_row_height = dml_min(a: 128, b: 1 << (unsigned int) dml_floor(a: dml_log2(x: PTEBufferSizeInRequests * |
2455 | *PixelPTEReqWidth / Pitch), granularity: 1)); |
2456 | #ifdef __DML_VBA_DEBUG__ |
2457 | dml_print("DML::%s: dpte_row_height = %d (1)\n" , __func__, |
2458 | PTEBufferSizeInRequests * *PixelPTEReqWidth / Pitch); |
2459 | dml_print("DML::%s: dpte_row_height = %f (2)\n" , __func__, |
2460 | dml_log2(PTEBufferSizeInRequests * *PixelPTEReqWidth / Pitch)); |
2461 | dml_print("DML::%s: dpte_row_height = %f (3)\n" , __func__, |
2462 | dml_floor(dml_log2(PTEBufferSizeInRequests * *PixelPTEReqWidth / Pitch), 1)); |
2463 | dml_print("DML::%s: dpte_row_height = %d (4)\n" , __func__, |
2464 | 1 << (unsigned int) dml_floor(dml_log2(PTEBufferSizeInRequests * |
2465 | *PixelPTEReqWidth / Pitch), 1)); |
2466 | dml_print("DML::%s: dpte_row_height = %d\n" , __func__, *dpte_row_height); |
2467 | #endif |
2468 | *dpte_row_width_ub = dml_ceil(a: ((double) Pitch * (double) *dpte_row_height - 1), |
2469 | granularity: (double) *PixelPTEReqWidth) + *PixelPTEReqWidth; |
2470 | *PixelPTEBytesPerRow = *dpte_row_width_ub / (double)*PixelPTEReqWidth * (double)*PTERequestSize; |
2471 | |
2472 | // VBA_DELTA, VBA doesn't have programming value for pte row height linear. |
2473 | *dpte_row_height_linear = 1 << (unsigned int) dml_floor(a: dml_log2(x: PTEBufferSizeInRequests * |
2474 | PixelPTEReqWidth_linear / Pitch), granularity: 1); |
2475 | if (*dpte_row_height_linear > 128) |
2476 | *dpte_row_height_linear = 128; |
2477 | |
2478 | } else if (!IsVertical(Scan: SourceRotation)) { |
2479 | *dpte_row_height = *PixelPTEReqHeight; |
2480 | |
2481 | if (GPUVMMinPageSizeKBytes > 64) { |
2482 | *dpte_row_width_ub = (dml_ceil(a: (Pitch * *dpte_row_height / *PixelPTEReqHeight - 1) / |
2483 | *PixelPTEReqWidth, granularity: 1) + 1) * *PixelPTEReqWidth; |
2484 | } else if (ViewportStationary && (NumberOfDPPs == 1)) { |
2485 | *dpte_row_width_ub = dml_floor(a: ViewportXStart + SwathWidth + |
2486 | *PixelPTEReqWidth - 1, granularity: *PixelPTEReqWidth) - |
2487 | dml_floor(a: ViewportXStart, granularity: *PixelPTEReqWidth); |
2488 | } else { |
2489 | *dpte_row_width_ub = (dml_ceil(a: (SwathWidth - 1) / *PixelPTEReqWidth, granularity: 1) + 1) * |
2490 | *PixelPTEReqWidth; |
2491 | } |
2492 | |
2493 | *PixelPTEBytesPerRow = *dpte_row_width_ub / *PixelPTEReqWidth * *PTERequestSize; |
2494 | } else { |
2495 | *dpte_row_height = dml_min(a: *PixelPTEReqWidth, b: MacroTileWidth); |
2496 | |
2497 | if (ViewportStationary && (NumberOfDPPs == 1)) { |
2498 | *dpte_row_width_ub = dml_floor(a: ViewportYStart + ViewportHeight + *PixelPTEReqHeight - 1, |
2499 | granularity: *PixelPTEReqHeight) - dml_floor(a: ViewportYStart, granularity: *PixelPTEReqHeight); |
2500 | } else { |
2501 | *dpte_row_width_ub = (dml_ceil(a: (SwathWidth - 1) / *PixelPTEReqHeight, granularity: 1) + 1) |
2502 | * *PixelPTEReqHeight; |
2503 | } |
2504 | |
2505 | *PixelPTEBytesPerRow = *dpte_row_width_ub / *PixelPTEReqHeight * *PTERequestSize; |
2506 | } |
2507 | |
2508 | if (GPUVMEnable != true) |
2509 | *PixelPTEBytesPerRow = 0; |
2510 | if (HostVMEnable == true) |
2511 | *PixelPTEBytesPerRow = *PixelPTEBytesPerRow * (1 + 8 * HostVMDynamicLevels); |
2512 | |
2513 | #ifdef __DML_VBA_DEBUG__ |
2514 | dml_print("DML::%s: GPUVMMinPageSizeKBytes = %d\n" , __func__, GPUVMMinPageSizeKBytes); |
2515 | dml_print("DML::%s: dpte_row_height = %d\n" , __func__, *dpte_row_height); |
2516 | dml_print("DML::%s: dpte_row_height_linear = %d\n" , __func__, *dpte_row_height_linear); |
2517 | dml_print("DML::%s: dpte_row_width_ub = %d\n" , __func__, *dpte_row_width_ub); |
2518 | dml_print("DML::%s: PixelPTEBytesPerRow = %d\n" , __func__, *PixelPTEBytesPerRow); |
2519 | dml_print("DML::%s: PTEBufferSizeInRequests = %d\n" , __func__, PTEBufferSizeInRequests); |
2520 | dml_print("DML::%s: dpte_row_height_one_row_per_frame = %d\n" , __func__, *dpte_row_height_one_row_per_frame); |
2521 | dml_print("DML::%s: dpte_row_width_ub_one_row_per_frame = %d\n" , |
2522 | __func__, *dpte_row_width_ub_one_row_per_frame); |
2523 | dml_print("DML::%s: PixelPTEBytesPerRow_one_row_per_frame = %d\n" , |
2524 | __func__, *PixelPTEBytesPerRow_one_row_per_frame); |
2525 | dml_print("DML: vm_bytes = meta_pte_bytes_per_frame (per_pipe) = MetaPTEBytesFrame = : %i\n" , |
2526 | *MetaPTEBytesFrame); |
2527 | #endif |
2528 | |
2529 | return PDEAndMetaPTEBytesFrame; |
2530 | } // CalculateVMAndRowBytes |
2531 | |
2532 | double dml32_CalculatePrefetchSourceLines( |
2533 | double VRatio, |
2534 | unsigned int VTaps, |
2535 | bool Interlace, |
2536 | bool ProgressiveToInterlaceUnitInOPP, |
2537 | unsigned int SwathHeight, |
2538 | enum dm_rotation_angle SourceRotation, |
2539 | bool ViewportStationary, |
2540 | double SwathWidth, |
2541 | unsigned int ViewportHeight, |
2542 | unsigned int ViewportXStart, |
2543 | unsigned int ViewportYStart, |
2544 | |
2545 | /* Output */ |
2546 | double *VInitPreFill, |
2547 | unsigned int *MaxNumSwath) |
2548 | { |
2549 | |
2550 | unsigned int vp_start_rot; |
2551 | unsigned int sw0_tmp; |
2552 | unsigned int MaxPartialSwath; |
2553 | double numLines; |
2554 | |
2555 | #ifdef __DML_VBA_DEBUG__ |
2556 | dml_print("DML::%s: VRatio = %f\n" , __func__, VRatio); |
2557 | dml_print("DML::%s: VTaps = %d\n" , __func__, VTaps); |
2558 | dml_print("DML::%s: ViewportXStart = %d\n" , __func__, ViewportXStart); |
2559 | dml_print("DML::%s: ViewportYStart = %d\n" , __func__, ViewportYStart); |
2560 | dml_print("DML::%s: ViewportStationary = %d\n" , __func__, ViewportStationary); |
2561 | dml_print("DML::%s: SwathHeight = %d\n" , __func__, SwathHeight); |
2562 | #endif |
2563 | if (ProgressiveToInterlaceUnitInOPP) |
2564 | *VInitPreFill = dml_floor(a: (VRatio + (double) VTaps + 1) / 2.0, granularity: 1); |
2565 | else |
2566 | *VInitPreFill = dml_floor(a: (VRatio + (double) VTaps + 1 + Interlace * 0.5 * VRatio) / 2.0, granularity: 1); |
2567 | |
2568 | if (ViewportStationary) { |
2569 | if (SourceRotation == dm_rotation_180 || SourceRotation == dm_rotation_180m) { |
2570 | vp_start_rot = SwathHeight - |
2571 | (((unsigned int) (ViewportYStart + ViewportHeight - 1) % SwathHeight) + 1); |
2572 | } else if (SourceRotation == dm_rotation_270 || SourceRotation == dm_rotation_90m) { |
2573 | vp_start_rot = ViewportXStart; |
2574 | } else if (SourceRotation == dm_rotation_90 || SourceRotation == dm_rotation_270m) { |
2575 | vp_start_rot = SwathHeight - |
2576 | (((unsigned int)(ViewportYStart + SwathWidth - 1) % SwathHeight) + 1); |
2577 | } else { |
2578 | vp_start_rot = ViewportYStart; |
2579 | } |
2580 | sw0_tmp = SwathHeight - (vp_start_rot % SwathHeight); |
2581 | if (sw0_tmp < *VInitPreFill) |
2582 | *MaxNumSwath = dml_ceil(a: (*VInitPreFill - sw0_tmp) / SwathHeight, granularity: 1) + 1; |
2583 | else |
2584 | *MaxNumSwath = 1; |
2585 | MaxPartialSwath = dml_max(a: 1, b: (unsigned int) (vp_start_rot + *VInitPreFill - 1) % SwathHeight); |
2586 | } else { |
2587 | *MaxNumSwath = dml_ceil(a: (*VInitPreFill - 1.0) / SwathHeight, granularity: 1) + 1; |
2588 | if (*VInitPreFill > 1) |
2589 | MaxPartialSwath = dml_max(a: 1, b: (unsigned int) (*VInitPreFill - 2) % SwathHeight); |
2590 | else |
2591 | MaxPartialSwath = dml_max(a: 1, b: (unsigned int) (*VInitPreFill + SwathHeight - 2) % SwathHeight); |
2592 | } |
2593 | numLines = *MaxNumSwath * SwathHeight + MaxPartialSwath; |
2594 | |
2595 | #ifdef __DML_VBA_DEBUG__ |
2596 | dml_print("DML::%s: vp_start_rot = %d\n" , __func__, vp_start_rot); |
2597 | dml_print("DML::%s: VInitPreFill = %d\n" , __func__, *VInitPreFill); |
2598 | dml_print("DML::%s: MaxPartialSwath = %d\n" , __func__, MaxPartialSwath); |
2599 | dml_print("DML::%s: MaxNumSwath = %d\n" , __func__, *MaxNumSwath); |
2600 | dml_print("DML::%s: Prefetch source lines = %3.2f\n" , __func__, numLines); |
2601 | #endif |
2602 | return numLines; |
2603 | |
2604 | } // CalculatePrefetchSourceLines |
2605 | |
2606 | void dml32_CalculateMALLUseForStaticScreen( |
2607 | unsigned int NumberOfActiveSurfaces, |
2608 | unsigned int MALLAllocatedForDCNFinal, |
2609 | enum dm_use_mall_for_static_screen_mode *UseMALLForStaticScreen, |
2610 | unsigned int SurfaceSizeInMALL[], |
2611 | bool one_row_per_frame_fits_in_buffer[], |
2612 | |
2613 | /* output */ |
2614 | bool UsesMALLForStaticScreen[]) |
2615 | { |
2616 | unsigned int k; |
2617 | unsigned int SurfaceToAddToMALL; |
2618 | bool CanAddAnotherSurfaceToMALL; |
2619 | unsigned int TotalSurfaceSizeInMALL; |
2620 | |
2621 | TotalSurfaceSizeInMALL = 0; |
2622 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
2623 | UsesMALLForStaticScreen[k] = (UseMALLForStaticScreen[k] == dm_use_mall_static_screen_enable); |
2624 | if (UsesMALLForStaticScreen[k]) |
2625 | TotalSurfaceSizeInMALL = TotalSurfaceSizeInMALL + SurfaceSizeInMALL[k]; |
2626 | #ifdef __DML_VBA_DEBUG__ |
2627 | dml_print("DML::%s: k=%d, UsesMALLForStaticScreen = %d\n" , __func__, k, UsesMALLForStaticScreen[k]); |
2628 | dml_print("DML::%s: k=%d, TotalSurfaceSizeInMALL = %d\n" , __func__, k, TotalSurfaceSizeInMALL); |
2629 | #endif |
2630 | } |
2631 | |
2632 | SurfaceToAddToMALL = 0; |
2633 | CanAddAnotherSurfaceToMALL = true; |
2634 | while (CanAddAnotherSurfaceToMALL) { |
2635 | CanAddAnotherSurfaceToMALL = false; |
2636 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
2637 | if (TotalSurfaceSizeInMALL + SurfaceSizeInMALL[k] <= MALLAllocatedForDCNFinal * 1024 * 1024 && |
2638 | !UsesMALLForStaticScreen[k] && |
2639 | UseMALLForStaticScreen[k] != dm_use_mall_static_screen_disable && |
2640 | one_row_per_frame_fits_in_buffer[k] && |
2641 | (!CanAddAnotherSurfaceToMALL || |
2642 | SurfaceSizeInMALL[k] < SurfaceSizeInMALL[SurfaceToAddToMALL])) { |
2643 | CanAddAnotherSurfaceToMALL = true; |
2644 | SurfaceToAddToMALL = k; |
2645 | #ifdef __DML_VBA_DEBUG__ |
2646 | dml_print("DML::%s: k=%d, UseMALLForStaticScreen = %d (dis, en, optimize)\n" , |
2647 | __func__, k, UseMALLForStaticScreen[k]); |
2648 | #endif |
2649 | } |
2650 | } |
2651 | if (CanAddAnotherSurfaceToMALL) { |
2652 | UsesMALLForStaticScreen[SurfaceToAddToMALL] = true; |
2653 | TotalSurfaceSizeInMALL = TotalSurfaceSizeInMALL + SurfaceSizeInMALL[SurfaceToAddToMALL]; |
2654 | |
2655 | #ifdef __DML_VBA_DEBUG__ |
2656 | dml_print("DML::%s: SurfaceToAddToMALL = %d\n" , __func__, SurfaceToAddToMALL); |
2657 | dml_print("DML::%s: TotalSurfaceSizeInMALL = %d\n" , __func__, TotalSurfaceSizeInMALL); |
2658 | #endif |
2659 | |
2660 | } |
2661 | } |
2662 | } |
2663 | |
2664 | void dml32_CalculateRowBandwidth( |
2665 | bool GPUVMEnable, |
2666 | enum source_format_class SourcePixelFormat, |
2667 | double VRatio, |
2668 | double VRatioChroma, |
2669 | bool DCCEnable, |
2670 | double LineTime, |
2671 | unsigned int MetaRowByteLuma, |
2672 | unsigned int MetaRowByteChroma, |
2673 | unsigned int meta_row_height_luma, |
2674 | unsigned int meta_row_height_chroma, |
2675 | unsigned int PixelPTEBytesPerRowLuma, |
2676 | unsigned int PixelPTEBytesPerRowChroma, |
2677 | unsigned int dpte_row_height_luma, |
2678 | unsigned int dpte_row_height_chroma, |
2679 | /* Output */ |
2680 | double *meta_row_bw, |
2681 | double *dpte_row_bw) |
2682 | { |
2683 | if (DCCEnable != true) { |
2684 | *meta_row_bw = 0; |
2685 | } else if (SourcePixelFormat == dm_420_8 || SourcePixelFormat == dm_420_10 || SourcePixelFormat == dm_420_12 || |
2686 | SourcePixelFormat == dm_rgbe_alpha) { |
2687 | *meta_row_bw = VRatio * MetaRowByteLuma / (meta_row_height_luma * LineTime) + VRatioChroma * |
2688 | MetaRowByteChroma / (meta_row_height_chroma * LineTime); |
2689 | } else { |
2690 | *meta_row_bw = VRatio * MetaRowByteLuma / (meta_row_height_luma * LineTime); |
2691 | } |
2692 | |
2693 | if (GPUVMEnable != true) { |
2694 | *dpte_row_bw = 0; |
2695 | } else if (SourcePixelFormat == dm_420_8 || SourcePixelFormat == dm_420_10 || SourcePixelFormat == dm_420_12 || |
2696 | SourcePixelFormat == dm_rgbe_alpha) { |
2697 | *dpte_row_bw = VRatio * PixelPTEBytesPerRowLuma / (dpte_row_height_luma * LineTime) + |
2698 | VRatioChroma * PixelPTEBytesPerRowChroma / (dpte_row_height_chroma * LineTime); |
2699 | } else { |
2700 | *dpte_row_bw = VRatio * PixelPTEBytesPerRowLuma / (dpte_row_height_luma * LineTime); |
2701 | } |
2702 | } |
2703 | |
2704 | double dml32_CalculateUrgentLatency( |
2705 | double UrgentLatencyPixelDataOnly, |
2706 | double UrgentLatencyPixelMixedWithVMData, |
2707 | double UrgentLatencyVMDataOnly, |
2708 | bool DoUrgentLatencyAdjustment, |
2709 | double UrgentLatencyAdjustmentFabricClockComponent, |
2710 | double UrgentLatencyAdjustmentFabricClockReference, |
2711 | double FabricClock) |
2712 | { |
2713 | double ret; |
2714 | |
2715 | ret = dml_max3(a: UrgentLatencyPixelDataOnly, b: UrgentLatencyPixelMixedWithVMData, c: UrgentLatencyVMDataOnly); |
2716 | if (DoUrgentLatencyAdjustment == true) { |
2717 | ret = ret + UrgentLatencyAdjustmentFabricClockComponent * |
2718 | (UrgentLatencyAdjustmentFabricClockReference / FabricClock - 1); |
2719 | } |
2720 | return ret; |
2721 | } |
2722 | |
2723 | void dml32_CalculateUrgentBurstFactor( |
2724 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange, |
2725 | unsigned int swath_width_luma_ub, |
2726 | unsigned int swath_width_chroma_ub, |
2727 | unsigned int SwathHeightY, |
2728 | unsigned int SwathHeightC, |
2729 | double LineTime, |
2730 | double UrgentLatency, |
2731 | double CursorBufferSize, |
2732 | unsigned int CursorWidth, |
2733 | unsigned int CursorBPP, |
2734 | double VRatio, |
2735 | double VRatioC, |
2736 | double BytePerPixelInDETY, |
2737 | double BytePerPixelInDETC, |
2738 | unsigned int DETBufferSizeY, |
2739 | unsigned int DETBufferSizeC, |
2740 | /* Output */ |
2741 | double *UrgentBurstFactorCursor, |
2742 | double *UrgentBurstFactorLuma, |
2743 | double *UrgentBurstFactorChroma, |
2744 | bool *NotEnoughUrgentLatencyHiding) |
2745 | { |
2746 | double LinesInDETLuma; |
2747 | double LinesInDETChroma; |
2748 | unsigned int LinesInCursorBuffer; |
2749 | double CursorBufferSizeInTime; |
2750 | double DETBufferSizeInTimeLuma; |
2751 | double DETBufferSizeInTimeChroma; |
2752 | |
2753 | *NotEnoughUrgentLatencyHiding = 0; |
2754 | |
2755 | if (CursorWidth > 0) { |
2756 | LinesInCursorBuffer = 1 << (unsigned int) dml_floor(a: dml_log2(x: CursorBufferSize * 1024.0 / |
2757 | (CursorWidth * CursorBPP / 8.0)), granularity: 1.0); |
2758 | if (VRatio > 0) { |
2759 | CursorBufferSizeInTime = LinesInCursorBuffer * LineTime / VRatio; |
2760 | if (CursorBufferSizeInTime - UrgentLatency <= 0) { |
2761 | *NotEnoughUrgentLatencyHiding = 1; |
2762 | *UrgentBurstFactorCursor = 0; |
2763 | } else { |
2764 | *UrgentBurstFactorCursor = CursorBufferSizeInTime / |
2765 | (CursorBufferSizeInTime - UrgentLatency); |
2766 | } |
2767 | } else { |
2768 | *UrgentBurstFactorCursor = 1; |
2769 | } |
2770 | } |
2771 | |
2772 | LinesInDETLuma = (UseMALLForPStateChange == dm_use_mall_pstate_change_phantom_pipe ? 1024*1024 : |
2773 | DETBufferSizeY) / BytePerPixelInDETY / swath_width_luma_ub; |
2774 | |
2775 | if (VRatio > 0) { |
2776 | DETBufferSizeInTimeLuma = dml_floor(a: LinesInDETLuma, granularity: SwathHeightY) * LineTime / VRatio; |
2777 | if (DETBufferSizeInTimeLuma - UrgentLatency <= 0) { |
2778 | *NotEnoughUrgentLatencyHiding = 1; |
2779 | *UrgentBurstFactorLuma = 0; |
2780 | } else { |
2781 | *UrgentBurstFactorLuma = DETBufferSizeInTimeLuma / (DETBufferSizeInTimeLuma - UrgentLatency); |
2782 | } |
2783 | } else { |
2784 | *UrgentBurstFactorLuma = 1; |
2785 | } |
2786 | |
2787 | if (BytePerPixelInDETC > 0) { |
2788 | LinesInDETChroma = (UseMALLForPStateChange == dm_use_mall_pstate_change_phantom_pipe ? |
2789 | 1024 * 1024 : DETBufferSizeC) / BytePerPixelInDETC |
2790 | / swath_width_chroma_ub; |
2791 | |
2792 | if (VRatio > 0) { |
2793 | DETBufferSizeInTimeChroma = dml_floor(a: LinesInDETChroma, granularity: SwathHeightC) * LineTime / VRatio; |
2794 | if (DETBufferSizeInTimeChroma - UrgentLatency <= 0) { |
2795 | *NotEnoughUrgentLatencyHiding = 1; |
2796 | *UrgentBurstFactorChroma = 0; |
2797 | } else { |
2798 | *UrgentBurstFactorChroma = DETBufferSizeInTimeChroma |
2799 | / (DETBufferSizeInTimeChroma - UrgentLatency); |
2800 | } |
2801 | } else { |
2802 | *UrgentBurstFactorChroma = 1; |
2803 | } |
2804 | } |
2805 | } // CalculateUrgentBurstFactor |
2806 | |
2807 | void dml32_CalculateDCFCLKDeepSleep( |
2808 | unsigned int NumberOfActiveSurfaces, |
2809 | unsigned int BytePerPixelY[], |
2810 | unsigned int BytePerPixelC[], |
2811 | double VRatio[], |
2812 | double VRatioChroma[], |
2813 | double SwathWidthY[], |
2814 | double SwathWidthC[], |
2815 | unsigned int DPPPerSurface[], |
2816 | double HRatio[], |
2817 | double HRatioChroma[], |
2818 | double PixelClock[], |
2819 | double PSCL_THROUGHPUT[], |
2820 | double PSCL_THROUGHPUT_CHROMA[], |
2821 | double Dppclk[], |
2822 | double ReadBandwidthLuma[], |
2823 | double ReadBandwidthChroma[], |
2824 | unsigned int ReturnBusWidth, |
2825 | |
2826 | /* Output */ |
2827 | double *DCFClkDeepSleep) |
2828 | { |
2829 | unsigned int k; |
2830 | double DisplayPipeLineDeliveryTimeLuma; |
2831 | double DisplayPipeLineDeliveryTimeChroma; |
2832 | double DCFClkDeepSleepPerSurface[DC__NUM_DPP__MAX]; |
2833 | double ReadBandwidth = 0.0; |
2834 | |
2835 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
2836 | |
2837 | if (VRatio[k] <= 1) { |
2838 | DisplayPipeLineDeliveryTimeLuma = SwathWidthY[k] * DPPPerSurface[k] / HRatio[k] |
2839 | / PixelClock[k]; |
2840 | } else { |
2841 | DisplayPipeLineDeliveryTimeLuma = SwathWidthY[k] / PSCL_THROUGHPUT[k] / Dppclk[k]; |
2842 | } |
2843 | if (BytePerPixelC[k] == 0) { |
2844 | DisplayPipeLineDeliveryTimeChroma = 0; |
2845 | } else { |
2846 | if (VRatioChroma[k] <= 1) { |
2847 | DisplayPipeLineDeliveryTimeChroma = SwathWidthC[k] * |
2848 | DPPPerSurface[k] / HRatioChroma[k] / PixelClock[k]; |
2849 | } else { |
2850 | DisplayPipeLineDeliveryTimeChroma = SwathWidthC[k] / PSCL_THROUGHPUT_CHROMA[k] |
2851 | / Dppclk[k]; |
2852 | } |
2853 | } |
2854 | |
2855 | if (BytePerPixelC[k] > 0) { |
2856 | DCFClkDeepSleepPerSurface[k] = dml_max(__DML_MIN_DCFCLK_FACTOR__ * SwathWidthY[k] * |
2857 | BytePerPixelY[k] / 32.0 / DisplayPipeLineDeliveryTimeLuma, |
2858 | __DML_MIN_DCFCLK_FACTOR__ * SwathWidthC[k] * BytePerPixelC[k] / |
2859 | 32.0 / DisplayPipeLineDeliveryTimeChroma); |
2860 | } else { |
2861 | DCFClkDeepSleepPerSurface[k] = __DML_MIN_DCFCLK_FACTOR__ * SwathWidthY[k] * BytePerPixelY[k] / |
2862 | 64.0 / DisplayPipeLineDeliveryTimeLuma; |
2863 | } |
2864 | DCFClkDeepSleepPerSurface[k] = dml_max(a: DCFClkDeepSleepPerSurface[k], b: PixelClock[k] / 16); |
2865 | |
2866 | #ifdef __DML_VBA_DEBUG__ |
2867 | dml_print("DML::%s: k=%d, PixelClock = %f\n" , __func__, k, PixelClock[k]); |
2868 | dml_print("DML::%s: k=%d, DCFClkDeepSleepPerSurface = %f\n" , __func__, k, DCFClkDeepSleepPerSurface[k]); |
2869 | #endif |
2870 | } |
2871 | |
2872 | for (k = 0; k < NumberOfActiveSurfaces; ++k) |
2873 | ReadBandwidth = ReadBandwidth + ReadBandwidthLuma[k] + ReadBandwidthChroma[k]; |
2874 | |
2875 | *DCFClkDeepSleep = dml_max(a: 8.0, __DML_MIN_DCFCLK_FACTOR__ * ReadBandwidth / (double) ReturnBusWidth); |
2876 | |
2877 | #ifdef __DML_VBA_DEBUG__ |
2878 | dml_print("DML::%s: __DML_MIN_DCFCLK_FACTOR__ = %f\n" , __func__, __DML_MIN_DCFCLK_FACTOR__); |
2879 | dml_print("DML::%s: ReadBandwidth = %f\n" , __func__, ReadBandwidth); |
2880 | dml_print("DML::%s: ReturnBusWidth = %d\n" , __func__, ReturnBusWidth); |
2881 | dml_print("DML::%s: DCFClkDeepSleep = %f\n" , __func__, *DCFClkDeepSleep); |
2882 | #endif |
2883 | |
2884 | for (k = 0; k < NumberOfActiveSurfaces; ++k) |
2885 | *DCFClkDeepSleep = dml_max(a: *DCFClkDeepSleep, b: DCFClkDeepSleepPerSurface[k]); |
2886 | #ifdef __DML_VBA_DEBUG__ |
2887 | dml_print("DML::%s: DCFClkDeepSleep = %f (final)\n" , __func__, *DCFClkDeepSleep); |
2888 | #endif |
2889 | } // CalculateDCFCLKDeepSleep |
2890 | |
2891 | double dml32_CalculateWriteBackDelay( |
2892 | enum source_format_class WritebackPixelFormat, |
2893 | double WritebackHRatio, |
2894 | double WritebackVRatio, |
2895 | unsigned int WritebackVTaps, |
2896 | unsigned int WritebackDestinationWidth, |
2897 | unsigned int WritebackDestinationHeight, |
2898 | unsigned int WritebackSourceHeight, |
2899 | unsigned int HTotal) |
2900 | { |
2901 | double CalculateWriteBackDelay; |
2902 | double Line_length; |
2903 | double Output_lines_last_notclamped; |
2904 | double WritebackVInit; |
2905 | |
2906 | WritebackVInit = (WritebackVRatio + WritebackVTaps + 1) / 2; |
2907 | Line_length = dml_max(a: (double) WritebackDestinationWidth, |
2908 | b: dml_ceil(a: (double)WritebackDestinationWidth / 6.0, granularity: 1.0) * WritebackVTaps); |
2909 | Output_lines_last_notclamped = WritebackDestinationHeight - 1 - |
2910 | dml_ceil(a: ((double)WritebackSourceHeight - |
2911 | (double) WritebackVInit) / (double)WritebackVRatio, granularity: 1.0); |
2912 | if (Output_lines_last_notclamped < 0) { |
2913 | CalculateWriteBackDelay = 0; |
2914 | } else { |
2915 | CalculateWriteBackDelay = Output_lines_last_notclamped * Line_length + |
2916 | (HTotal - WritebackDestinationWidth) + 80; |
2917 | } |
2918 | return CalculateWriteBackDelay; |
2919 | } |
2920 | |
2921 | void dml32_UseMinimumDCFCLK( |
2922 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange[], |
2923 | bool DRRDisplay[], |
2924 | bool SynchronizeDRRDisplaysForUCLKPStateChangeFinal, |
2925 | unsigned int MaxInterDCNTileRepeaters, |
2926 | unsigned int MaxPrefetchMode, |
2927 | double DRAMClockChangeLatencyFinal, |
2928 | double FCLKChangeLatency, |
2929 | double SREnterPlusExitTime, |
2930 | unsigned int ReturnBusWidth, |
2931 | unsigned int RoundTripPingLatencyCycles, |
2932 | unsigned int ReorderingBytes, |
2933 | unsigned int PixelChunkSizeInKByte, |
2934 | unsigned int MetaChunkSize, |
2935 | bool GPUVMEnable, |
2936 | unsigned int GPUVMMaxPageTableLevels, |
2937 | bool HostVMEnable, |
2938 | unsigned int NumberOfActiveSurfaces, |
2939 | double HostVMMinPageSize, |
2940 | unsigned int HostVMMaxNonCachedPageTableLevels, |
2941 | bool DynamicMetadataVMEnabled, |
2942 | bool ImmediateFlipRequirement, |
2943 | bool ProgressiveToInterlaceUnitInOPP, |
2944 | double MaxAveragePercentOfIdealSDPPortBWDisplayCanUseInNormalSystemOperation, |
2945 | double PercentOfIdealSDPPortBWReceivedAfterUrgLatency, |
2946 | unsigned int VTotal[], |
2947 | unsigned int VActive[], |
2948 | unsigned int DynamicMetadataTransmittedBytes[], |
2949 | unsigned int DynamicMetadataLinesBeforeActiveRequired[], |
2950 | bool Interlace[], |
2951 | double RequiredDPPCLKPerSurface[][2][DC__NUM_DPP__MAX], |
2952 | double RequiredDISPCLK[][2], |
2953 | double UrgLatency[], |
2954 | unsigned int NoOfDPP[][2][DC__NUM_DPP__MAX], |
2955 | double ProjectedDCFClkDeepSleep[][2], |
2956 | double MaximumVStartup[][2][DC__NUM_DPP__MAX], |
2957 | unsigned int TotalNumberOfActiveDPP[][2], |
2958 | unsigned int TotalNumberOfDCCActiveDPP[][2], |
2959 | unsigned int dpte_group_bytes[], |
2960 | double PrefetchLinesY[][2][DC__NUM_DPP__MAX], |
2961 | double PrefetchLinesC[][2][DC__NUM_DPP__MAX], |
2962 | unsigned int swath_width_luma_ub_all_states[][2][DC__NUM_DPP__MAX], |
2963 | unsigned int swath_width_chroma_ub_all_states[][2][DC__NUM_DPP__MAX], |
2964 | unsigned int BytePerPixelY[], |
2965 | unsigned int BytePerPixelC[], |
2966 | unsigned int HTotal[], |
2967 | double PixelClock[], |
2968 | double PDEAndMetaPTEBytesPerFrame[][2][DC__NUM_DPP__MAX], |
2969 | double DPTEBytesPerRow[][2][DC__NUM_DPP__MAX], |
2970 | double MetaRowBytes[][2][DC__NUM_DPP__MAX], |
2971 | bool DynamicMetadataEnable[], |
2972 | double ReadBandwidthLuma[], |
2973 | double ReadBandwidthChroma[], |
2974 | double DCFCLKPerState[], |
2975 | /* Output */ |
2976 | double DCFCLKState[][2]) |
2977 | { |
2978 | unsigned int i, j, k; |
2979 | unsigned int dummy1; |
2980 | double dummy2, dummy3; |
2981 | double NormalEfficiency; |
2982 | double TotalMaxPrefetchFlipDPTERowBandwidth[DC__VOLTAGE_STATES][2]; |
2983 | |
2984 | NormalEfficiency = PercentOfIdealSDPPortBWReceivedAfterUrgLatency / 100.0; |
2985 | for (i = 0; i < DC__VOLTAGE_STATES; ++i) { |
2986 | for (j = 0; j <= 1; ++j) { |
2987 | double PixelDCFCLKCyclesRequiredInPrefetch[DC__NUM_DPP__MAX]; |
2988 | double PrefetchPixelLinesTime[DC__NUM_DPP__MAX]; |
2989 | double DCFCLKRequiredForPeakBandwidthPerSurface[DC__NUM_DPP__MAX]; |
2990 | double [DC__NUM_DPP__MAX]; |
2991 | double MinimumTWait = 0.0; |
2992 | double DPTEBandwidth; |
2993 | double DCFCLKRequiredForAverageBandwidth; |
2994 | unsigned int ; |
2995 | double ; |
2996 | double DCFCLKRequiredForPeakBandwidth; |
2997 | unsigned int NoOfDPPState[DC__NUM_DPP__MAX]; |
2998 | double MinimumTvmPlus2Tr0; |
2999 | |
3000 | TotalMaxPrefetchFlipDPTERowBandwidth[i][j] = 0; |
3001 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
3002 | TotalMaxPrefetchFlipDPTERowBandwidth[i][j] = TotalMaxPrefetchFlipDPTERowBandwidth[i][j] |
3003 | + NoOfDPP[i][j][k] * DPTEBytesPerRow[i][j][k] |
3004 | / (15.75 * HTotal[k] / PixelClock[k]); |
3005 | } |
3006 | |
3007 | for (k = 0; k <= NumberOfActiveSurfaces - 1; ++k) |
3008 | NoOfDPPState[k] = NoOfDPP[i][j][k]; |
3009 | |
3010 | DPTEBandwidth = TotalMaxPrefetchFlipDPTERowBandwidth[i][j]; |
3011 | DCFCLKRequiredForAverageBandwidth = dml_max(a: ProjectedDCFClkDeepSleep[i][j], b: DPTEBandwidth / NormalEfficiency / ReturnBusWidth); |
3012 | |
3013 | ExtraLatencyBytes = dml32_CalculateExtraLatencyBytes(ReorderingBytes, |
3014 | TotalNumberOfActiveDPP: TotalNumberOfActiveDPP[i][j], PixelChunkSizeInKByte, |
3015 | TotalNumberOfDCCActiveDPP: TotalNumberOfDCCActiveDPP[i][j], MetaChunkSize, GPUVMEnable, HostVMEnable, |
3016 | NumberOfActiveSurfaces, NumberOfDPP: NoOfDPPState, dpte_group_bytes, HostVMInefficiencyFactor: 1, HostVMMinPageSize, |
3017 | HostVMMaxNonCachedPageTableLevels); |
3018 | ExtraLatencyCycles = RoundTripPingLatencyCycles + __DML_ARB_TO_RET_DELAY__ |
3019 | + ExtraLatencyBytes / NormalEfficiency / ReturnBusWidth; |
3020 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
3021 | double DCFCLKCyclesRequiredInPrefetch; |
3022 | double PrefetchTime; |
3023 | |
3024 | PixelDCFCLKCyclesRequiredInPrefetch[k] = (PrefetchLinesY[i][j][k] |
3025 | * swath_width_luma_ub_all_states[i][j][k] * BytePerPixelY[k] |
3026 | + PrefetchLinesC[i][j][k] * swath_width_chroma_ub_all_states[i][j][k] |
3027 | * BytePerPixelC[k]) / NormalEfficiency |
3028 | / ReturnBusWidth; |
3029 | DCFCLKCyclesRequiredInPrefetch = 2 * ExtraLatencyCycles / NoOfDPPState[k] |
3030 | + PDEAndMetaPTEBytesPerFrame[i][j][k] / NormalEfficiency |
3031 | / NormalEfficiency / ReturnBusWidth |
3032 | * (GPUVMMaxPageTableLevels > 2 ? 1 : 0) |
3033 | + 2 * DPTEBytesPerRow[i][j][k] / NormalEfficiency / NormalEfficiency |
3034 | / ReturnBusWidth |
3035 | + 2 * MetaRowBytes[i][j][k] / NormalEfficiency / ReturnBusWidth |
3036 | + PixelDCFCLKCyclesRequiredInPrefetch[k]; |
3037 | PrefetchPixelLinesTime[k] = dml_max(a: PrefetchLinesY[i][j][k], b: PrefetchLinesC[i][j][k]) |
3038 | * HTotal[k] / PixelClock[k]; |
3039 | DynamicMetadataVMExtraLatency[k] = (GPUVMEnable == true && |
3040 | DynamicMetadataEnable[k] == true && DynamicMetadataVMEnabled == true) ? |
3041 | UrgLatency[i] * GPUVMMaxPageTableLevels * |
3042 | (HostVMEnable == true ? HostVMMaxNonCachedPageTableLevels + 1 : 1) : 0; |
3043 | |
3044 | MinimumTWait = dml32_CalculateTWait(PrefetchMode: MaxPrefetchMode, |
3045 | UseMALLForPStateChange: UseMALLForPStateChange[k], |
3046 | SynchronizeDRRDisplaysForUCLKPStateChangeFinal, |
3047 | DRRDisplay: DRRDisplay[k], |
3048 | DRAMClockChangeLatency: DRAMClockChangeLatencyFinal, |
3049 | FCLKChangeLatency, |
3050 | UrgentLatency: UrgLatency[i], |
3051 | SREnterPlusExitTime); |
3052 | |
3053 | PrefetchTime = (MaximumVStartup[i][j][k] - 1) * HTotal[k] / PixelClock[k] - |
3054 | MinimumTWait - UrgLatency[i] * |
3055 | ((GPUVMMaxPageTableLevels <= 2 ? GPUVMMaxPageTableLevels : |
3056 | GPUVMMaxPageTableLevels - 2) * (HostVMEnable == true ? |
3057 | HostVMMaxNonCachedPageTableLevels + 1 : 1) - 1) - |
3058 | DynamicMetadataVMExtraLatency[k]; |
3059 | |
3060 | if (PrefetchTime > 0) { |
3061 | double ExpectedVRatioPrefetch; |
3062 | |
3063 | ExpectedVRatioPrefetch = PrefetchPixelLinesTime[k] / (PrefetchTime * |
3064 | PixelDCFCLKCyclesRequiredInPrefetch[k] / |
3065 | DCFCLKCyclesRequiredInPrefetch); |
3066 | DCFCLKRequiredForPeakBandwidthPerSurface[k] = NoOfDPPState[k] * |
3067 | PixelDCFCLKCyclesRequiredInPrefetch[k] / |
3068 | PrefetchPixelLinesTime[k] * |
3069 | dml_max(a: 1.0, b: ExpectedVRatioPrefetch) * |
3070 | dml_max(a: 1.0, b: ExpectedVRatioPrefetch / 4); |
3071 | if (HostVMEnable == true || ImmediateFlipRequirement == true) { |
3072 | DCFCLKRequiredForPeakBandwidthPerSurface[k] = |
3073 | DCFCLKRequiredForPeakBandwidthPerSurface[k] + |
3074 | NoOfDPPState[k] * DPTEBandwidth / NormalEfficiency / |
3075 | NormalEfficiency / ReturnBusWidth; |
3076 | } |
3077 | } else { |
3078 | DCFCLKRequiredForPeakBandwidthPerSurface[k] = DCFCLKPerState[i]; |
3079 | } |
3080 | if (DynamicMetadataEnable[k] == true) { |
3081 | double TSetupPipe; |
3082 | double TdmbfPipe; |
3083 | double TdmsksPipe; |
3084 | double TdmecPipe; |
3085 | double ; |
3086 | |
3087 | dml32_CalculateVUpdateAndDynamicMetadataParameters( |
3088 | MaxInterDCNTileRepeaters, |
3089 | Dppclk: RequiredDPPCLKPerSurface[i][j][k], |
3090 | Dispclk: RequiredDISPCLK[i][j], |
3091 | DCFClkDeepSleep: ProjectedDCFClkDeepSleep[i][j], |
3092 | PixelClock: PixelClock[k], |
3093 | HTotal: HTotal[k], |
3094 | VBlank: VTotal[k] - VActive[k], |
3095 | DynamicMetadataTransmittedBytes: DynamicMetadataTransmittedBytes[k], |
3096 | DynamicMetadataLinesBeforeActiveRequired: DynamicMetadataLinesBeforeActiveRequired[k], |
3097 | InterlaceEnable: Interlace[k], |
3098 | ProgressiveToInterlaceUnitInOPP, |
3099 | |
3100 | /* output */ |
3101 | TSetup: &TSetupPipe, |
3102 | Tdmbf: &TdmbfPipe, |
3103 | Tdmec: &TdmecPipe, |
3104 | Tdmsks: &TdmsksPipe, |
3105 | VUpdateOffsetPix: &dummy1, |
3106 | VUpdateWidthPix: &dummy2, |
3107 | VReadyOffsetPix: &dummy3); |
3108 | AllowedTimeForUrgentExtraLatency = MaximumVStartup[i][j][k] * HTotal[k] / |
3109 | PixelClock[k] - MinimumTWait - TSetupPipe - TdmbfPipe - |
3110 | TdmecPipe - TdmsksPipe - DynamicMetadataVMExtraLatency[k]; |
3111 | if (AllowedTimeForUrgentExtraLatency > 0) |
3112 | DCFCLKRequiredForPeakBandwidthPerSurface[k] = |
3113 | dml_max(a: DCFCLKRequiredForPeakBandwidthPerSurface[k], |
3114 | b: ExtraLatencyCycles / AllowedTimeForUrgentExtraLatency); |
3115 | else |
3116 | DCFCLKRequiredForPeakBandwidthPerSurface[k] = DCFCLKPerState[i]; |
3117 | } |
3118 | } |
3119 | DCFCLKRequiredForPeakBandwidth = 0; |
3120 | for (k = 0; k <= NumberOfActiveSurfaces - 1; ++k) { |
3121 | DCFCLKRequiredForPeakBandwidth = DCFCLKRequiredForPeakBandwidth + |
3122 | DCFCLKRequiredForPeakBandwidthPerSurface[k]; |
3123 | } |
3124 | MinimumTvmPlus2Tr0 = UrgLatency[i] * (GPUVMEnable == true ? |
3125 | (HostVMEnable == true ? (GPUVMMaxPageTableLevels + 2) * |
3126 | (HostVMMaxNonCachedPageTableLevels + 1) - 1 : GPUVMMaxPageTableLevels + 1) : 0); |
3127 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
3128 | double MaximumTvmPlus2Tr0PlusTsw; |
3129 | |
3130 | MaximumTvmPlus2Tr0PlusTsw = (MaximumVStartup[i][j][k] - 2) * HTotal[k] / |
3131 | PixelClock[k] - MinimumTWait - DynamicMetadataVMExtraLatency[k]; |
3132 | if (MaximumTvmPlus2Tr0PlusTsw <= MinimumTvmPlus2Tr0 + PrefetchPixelLinesTime[k] / 4) { |
3133 | DCFCLKRequiredForPeakBandwidth = DCFCLKPerState[i]; |
3134 | } else { |
3135 | DCFCLKRequiredForPeakBandwidth = dml_max3(a: DCFCLKRequiredForPeakBandwidth, |
3136 | b: 2 * ExtraLatencyCycles / (MaximumTvmPlus2Tr0PlusTsw - |
3137 | MinimumTvmPlus2Tr0 - |
3138 | PrefetchPixelLinesTime[k] / 4), |
3139 | c: (2 * ExtraLatencyCycles + |
3140 | PixelDCFCLKCyclesRequiredInPrefetch[k]) / |
3141 | (MaximumTvmPlus2Tr0PlusTsw - MinimumTvmPlus2Tr0)); |
3142 | } |
3143 | } |
3144 | DCFCLKState[i][j] = dml_min(a: DCFCLKPerState[i], b: 1.05 * |
3145 | dml_max(a: DCFCLKRequiredForAverageBandwidth, b: DCFCLKRequiredForPeakBandwidth)); |
3146 | } |
3147 | } |
3148 | } |
3149 | |
3150 | unsigned int (unsigned int ReorderingBytes, |
3151 | unsigned int TotalNumberOfActiveDPP, |
3152 | unsigned int PixelChunkSizeInKByte, |
3153 | unsigned int TotalNumberOfDCCActiveDPP, |
3154 | unsigned int MetaChunkSize, |
3155 | bool GPUVMEnable, |
3156 | bool HostVMEnable, |
3157 | unsigned int NumberOfActiveSurfaces, |
3158 | unsigned int NumberOfDPP[], |
3159 | unsigned int dpte_group_bytes[], |
3160 | double HostVMInefficiencyFactor, |
3161 | double HostVMMinPageSize, |
3162 | unsigned int HostVMMaxNonCachedPageTableLevels) |
3163 | { |
3164 | unsigned int k; |
3165 | double ret; |
3166 | unsigned int HostVMDynamicLevels; |
3167 | |
3168 | if (GPUVMEnable == true && HostVMEnable == true) { |
3169 | if (HostVMMinPageSize < 2048) |
3170 | HostVMDynamicLevels = HostVMMaxNonCachedPageTableLevels; |
3171 | else if (HostVMMinPageSize >= 2048 && HostVMMinPageSize < 1048576) |
3172 | HostVMDynamicLevels = dml_max(a: 0, b: (int) HostVMMaxNonCachedPageTableLevels - 1); |
3173 | else |
3174 | HostVMDynamicLevels = dml_max(a: 0, b: (int) HostVMMaxNonCachedPageTableLevels - 2); |
3175 | } else { |
3176 | HostVMDynamicLevels = 0; |
3177 | } |
3178 | |
3179 | ret = ReorderingBytes + (TotalNumberOfActiveDPP * PixelChunkSizeInKByte + |
3180 | TotalNumberOfDCCActiveDPP * MetaChunkSize) * 1024.0; |
3181 | |
3182 | if (GPUVMEnable == true) { |
3183 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
3184 | ret = ret + NumberOfDPP[k] * dpte_group_bytes[k] * |
3185 | (1 + 8 * HostVMDynamicLevels) * HostVMInefficiencyFactor; |
3186 | } |
3187 | } |
3188 | return ret; |
3189 | } |
3190 | |
3191 | void dml32_CalculateVUpdateAndDynamicMetadataParameters( |
3192 | unsigned int MaxInterDCNTileRepeaters, |
3193 | double Dppclk, |
3194 | double Dispclk, |
3195 | double DCFClkDeepSleep, |
3196 | double PixelClock, |
3197 | unsigned int HTotal, |
3198 | unsigned int VBlank, |
3199 | unsigned int DynamicMetadataTransmittedBytes, |
3200 | unsigned int DynamicMetadataLinesBeforeActiveRequired, |
3201 | unsigned int InterlaceEnable, |
3202 | bool ProgressiveToInterlaceUnitInOPP, |
3203 | |
3204 | /* output */ |
3205 | double *TSetup, |
3206 | double *Tdmbf, |
3207 | double *Tdmec, |
3208 | double *Tdmsks, |
3209 | unsigned int *VUpdateOffsetPix, |
3210 | double *VUpdateWidthPix, |
3211 | double *VReadyOffsetPix) |
3212 | { |
3213 | double TotalRepeaterDelayTime; |
3214 | |
3215 | TotalRepeaterDelayTime = MaxInterDCNTileRepeaters * (2 / Dppclk + 3 / Dispclk); |
3216 | *VUpdateWidthPix = |
3217 | dml_ceil(a: (14.0 / DCFClkDeepSleep + 12.0 / Dppclk + TotalRepeaterDelayTime) * PixelClock, granularity: 1.0); |
3218 | *VReadyOffsetPix = dml_ceil(a: dml_max(a: 150.0 / Dppclk, |
3219 | b: TotalRepeaterDelayTime + 20.0 / DCFClkDeepSleep + 10.0 / Dppclk) * PixelClock, granularity: 1.0); |
3220 | *VUpdateOffsetPix = dml_ceil(a: HTotal / 4.0, granularity: 1.0); |
3221 | *TSetup = (*VUpdateOffsetPix + *VUpdateWidthPix + *VReadyOffsetPix) / PixelClock; |
3222 | *Tdmbf = DynamicMetadataTransmittedBytes / 4.0 / Dispclk; |
3223 | *Tdmec = HTotal / PixelClock; |
3224 | |
3225 | if (DynamicMetadataLinesBeforeActiveRequired == 0) |
3226 | *Tdmsks = VBlank * HTotal / PixelClock / 2.0; |
3227 | else |
3228 | *Tdmsks = DynamicMetadataLinesBeforeActiveRequired * HTotal / PixelClock; |
3229 | |
3230 | if (InterlaceEnable == 1 && ProgressiveToInterlaceUnitInOPP == false) |
3231 | *Tdmsks = *Tdmsks / 2; |
3232 | #ifdef __DML_VBA_DEBUG__ |
3233 | dml_print("DML::%s: VUpdateWidthPix = %d\n" , __func__, *VUpdateWidthPix); |
3234 | dml_print("DML::%s: VReadyOffsetPix = %d\n" , __func__, *VReadyOffsetPix); |
3235 | dml_print("DML::%s: VUpdateOffsetPix = %d\n" , __func__, *VUpdateOffsetPix); |
3236 | |
3237 | dml_print("DML::%s: DynamicMetadataLinesBeforeActiveRequired = %d\n" , |
3238 | __func__, DynamicMetadataLinesBeforeActiveRequired); |
3239 | dml_print("DML::%s: VBlank = %d\n" , __func__, VBlank); |
3240 | dml_print("DML::%s: HTotal = %d\n" , __func__, HTotal); |
3241 | dml_print("DML::%s: PixelClock = %f\n" , __func__, PixelClock); |
3242 | dml_print("DML::%s: Tdmsks = %f\n" , __func__, *Tdmsks); |
3243 | #endif |
3244 | } |
3245 | |
3246 | double dml32_CalculateTWait( |
3247 | unsigned int PrefetchMode, |
3248 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange, |
3249 | bool SynchronizeDRRDisplaysForUCLKPStateChangeFinal, |
3250 | bool DRRDisplay, |
3251 | double DRAMClockChangeLatency, |
3252 | double FCLKChangeLatency, |
3253 | double UrgentLatency, |
3254 | double SREnterPlusExitTime) |
3255 | { |
3256 | double TWait = 0.0; |
3257 | |
3258 | if (PrefetchMode == 0 && |
3259 | !(UseMALLForPStateChange == dm_use_mall_pstate_change_full_frame) && |
3260 | !(UseMALLForPStateChange == dm_use_mall_pstate_change_sub_viewport) && |
3261 | !(UseMALLForPStateChange == dm_use_mall_pstate_change_phantom_pipe) && |
3262 | !(SynchronizeDRRDisplaysForUCLKPStateChangeFinal && DRRDisplay)) { |
3263 | TWait = dml_max3(a: DRAMClockChangeLatency + UrgentLatency, b: SREnterPlusExitTime, c: UrgentLatency); |
3264 | } else if (PrefetchMode <= 1 && !(UseMALLForPStateChange == dm_use_mall_pstate_change_phantom_pipe)) { |
3265 | TWait = dml_max3(a: FCLKChangeLatency + UrgentLatency, b: SREnterPlusExitTime, c: UrgentLatency); |
3266 | } else if (PrefetchMode <= 2 && !(UseMALLForPStateChange == dm_use_mall_pstate_change_phantom_pipe)) { |
3267 | TWait = dml_max(a: SREnterPlusExitTime, b: UrgentLatency); |
3268 | } else { |
3269 | TWait = UrgentLatency; |
3270 | } |
3271 | |
3272 | #ifdef __DML_VBA_DEBUG__ |
3273 | dml_print("DML::%s: PrefetchMode = %d\n" , __func__, PrefetchMode); |
3274 | dml_print("DML::%s: TWait = %f\n" , __func__, TWait); |
3275 | #endif |
3276 | return TWait; |
3277 | } // CalculateTWait |
3278 | |
3279 | // Function: get_return_bw_mbps |
3280 | // Megabyte per second |
3281 | double dml32_get_return_bw_mbps(const soc_bounding_box_st *soc, |
3282 | const int VoltageLevel, |
3283 | const bool HostVMEnable, |
3284 | const double DCFCLK, |
3285 | const double FabricClock, |
3286 | const double DRAMSpeed) |
3287 | { |
3288 | double ReturnBW = 0.; |
3289 | double IdealSDPPortBandwidth = soc->return_bus_width_bytes /*mode_lib->vba.ReturnBusWidth*/ * DCFCLK; |
3290 | double IdealFabricBandwidth = FabricClock * soc->fabric_datapath_to_dcn_data_return_bytes; |
3291 | double IdealDRAMBandwidth = DRAMSpeed * soc->num_chans * soc->dram_channel_width_bytes; |
3292 | double PixelDataOnlyReturnBW = dml_min3(a: IdealSDPPortBandwidth * soc->pct_ideal_sdp_bw_after_urgent / 100, |
3293 | b: IdealFabricBandwidth * soc->pct_ideal_fabric_bw_after_urgent / 100, |
3294 | c: IdealDRAMBandwidth * (VoltageLevel < 2 ? soc->pct_ideal_dram_bw_after_urgent_strobe : |
3295 | soc->pct_ideal_dram_sdp_bw_after_urgent_pixel_only) / 100); |
3296 | double PixelMixedWithVMDataReturnBW = dml_min3(a: IdealSDPPortBandwidth * soc->pct_ideal_sdp_bw_after_urgent / 100, |
3297 | b: IdealFabricBandwidth * soc->pct_ideal_fabric_bw_after_urgent / 100, |
3298 | c: IdealDRAMBandwidth * (VoltageLevel < 2 ? soc->pct_ideal_dram_bw_after_urgent_strobe : |
3299 | soc->pct_ideal_dram_sdp_bw_after_urgent_pixel_only) / 100); |
3300 | |
3301 | if (HostVMEnable != true) |
3302 | ReturnBW = PixelDataOnlyReturnBW; |
3303 | else |
3304 | ReturnBW = PixelMixedWithVMDataReturnBW; |
3305 | |
3306 | #ifdef __DML_VBA_DEBUG__ |
3307 | dml_print("DML::%s: VoltageLevel = %d\n" , __func__, VoltageLevel); |
3308 | dml_print("DML::%s: HostVMEnable = %d\n" , __func__, HostVMEnable); |
3309 | dml_print("DML::%s: DCFCLK = %f\n" , __func__, DCFCLK); |
3310 | dml_print("DML::%s: FabricClock = %f\n" , __func__, FabricClock); |
3311 | dml_print("DML::%s: DRAMSpeed = %f\n" , __func__, DRAMSpeed); |
3312 | dml_print("DML::%s: IdealSDPPortBandwidth = %f\n" , __func__, IdealSDPPortBandwidth); |
3313 | dml_print("DML::%s: IdealFabricBandwidth = %f\n" , __func__, IdealFabricBandwidth); |
3314 | dml_print("DML::%s: IdealDRAMBandwidth = %f\n" , __func__, IdealDRAMBandwidth); |
3315 | dml_print("DML::%s: PixelDataOnlyReturnBW = %f\n" , __func__, PixelDataOnlyReturnBW); |
3316 | dml_print("DML::%s: PixelMixedWithVMDataReturnBW = %f\n" , __func__, PixelMixedWithVMDataReturnBW); |
3317 | dml_print("DML::%s: ReturnBW = %f MBps\n" , __func__, ReturnBW); |
3318 | #endif |
3319 | return ReturnBW; |
3320 | } |
3321 | |
3322 | // Function: get_return_bw_mbps_vm_only |
3323 | // Megabyte per second |
3324 | double dml32_get_return_bw_mbps_vm_only(const soc_bounding_box_st *soc, |
3325 | const int VoltageLevel, |
3326 | const double DCFCLK, |
3327 | const double FabricClock, |
3328 | const double DRAMSpeed) |
3329 | { |
3330 | double VMDataOnlyReturnBW = dml_min3( |
3331 | a: soc->return_bus_width_bytes * DCFCLK * soc->pct_ideal_sdp_bw_after_urgent / 100.0, |
3332 | b: FabricClock * soc->fabric_datapath_to_dcn_data_return_bytes |
3333 | * soc->pct_ideal_sdp_bw_after_urgent / 100.0, |
3334 | c: DRAMSpeed * soc->num_chans * soc->dram_channel_width_bytes |
3335 | * (VoltageLevel < 2 ? |
3336 | soc->pct_ideal_dram_bw_after_urgent_strobe : |
3337 | soc->pct_ideal_dram_sdp_bw_after_urgent_vm_only) / 100.0); |
3338 | #ifdef __DML_VBA_DEBUG__ |
3339 | dml_print("DML::%s: VoltageLevel = %d\n" , __func__, VoltageLevel); |
3340 | dml_print("DML::%s: DCFCLK = %f\n" , __func__, DCFCLK); |
3341 | dml_print("DML::%s: FabricClock = %f\n" , __func__, FabricClock); |
3342 | dml_print("DML::%s: DRAMSpeed = %f\n" , __func__, DRAMSpeed); |
3343 | dml_print("DML::%s: VMDataOnlyReturnBW = %f\n" , __func__, VMDataOnlyReturnBW); |
3344 | #endif |
3345 | return VMDataOnlyReturnBW; |
3346 | } |
3347 | |
3348 | double ( |
3349 | unsigned int RoundTripPingLatencyCycles, |
3350 | unsigned int ReorderingBytes, |
3351 | double DCFCLK, |
3352 | unsigned int TotalNumberOfActiveDPP, |
3353 | unsigned int PixelChunkSizeInKByte, |
3354 | unsigned int TotalNumberOfDCCActiveDPP, |
3355 | unsigned int MetaChunkSize, |
3356 | double ReturnBW, |
3357 | bool GPUVMEnable, |
3358 | bool HostVMEnable, |
3359 | unsigned int NumberOfActiveSurfaces, |
3360 | unsigned int NumberOfDPP[], |
3361 | unsigned int dpte_group_bytes[], |
3362 | double HostVMInefficiencyFactor, |
3363 | double HostVMMinPageSize, |
3364 | unsigned int HostVMMaxNonCachedPageTableLevels) |
3365 | { |
3366 | double ; |
3367 | double ; |
3368 | |
3369 | ExtraLatencyBytes = dml32_CalculateExtraLatencyBytes( |
3370 | ReorderingBytes, |
3371 | TotalNumberOfActiveDPP, |
3372 | PixelChunkSizeInKByte, |
3373 | TotalNumberOfDCCActiveDPP, |
3374 | MetaChunkSize, |
3375 | GPUVMEnable, |
3376 | HostVMEnable, |
3377 | NumberOfActiveSurfaces, |
3378 | NumberOfDPP, |
3379 | dpte_group_bytes, |
3380 | HostVMInefficiencyFactor, |
3381 | HostVMMinPageSize, |
3382 | HostVMMaxNonCachedPageTableLevels); |
3383 | |
3384 | ExtraLatency = (RoundTripPingLatencyCycles + __DML_ARB_TO_RET_DELAY__) / DCFCLK + ExtraLatencyBytes / ReturnBW; |
3385 | |
3386 | #ifdef __DML_VBA_DEBUG__ |
3387 | dml_print("DML::%s: RoundTripPingLatencyCycles=%d\n" , __func__, RoundTripPingLatencyCycles); |
3388 | dml_print("DML::%s: DCFCLK=%f\n" , __func__, DCFCLK); |
3389 | dml_print("DML::%s: ExtraLatencyBytes=%f\n" , __func__, ExtraLatencyBytes); |
3390 | dml_print("DML::%s: ReturnBW=%f\n" , __func__, ReturnBW); |
3391 | dml_print("DML::%s: ExtraLatency=%f\n" , __func__, ExtraLatency); |
3392 | #endif |
3393 | |
3394 | return ExtraLatency; |
3395 | } // CalculateExtraLatency |
3396 | |
3397 | bool dml32_CalculatePrefetchSchedule( |
3398 | struct vba_vars_st *v, |
3399 | unsigned int k, |
3400 | double HostVMInefficiencyFactor, |
3401 | DmlPipe *myPipe, |
3402 | unsigned int DSCDelay, |
3403 | unsigned int DPP_RECOUT_WIDTH, |
3404 | unsigned int VStartup, |
3405 | unsigned int MaxVStartup, |
3406 | double UrgentLatency, |
3407 | double , |
3408 | double TCalc, |
3409 | unsigned int PDEAndMetaPTEBytesFrame, |
3410 | unsigned int MetaRowByte, |
3411 | unsigned int PixelPTEBytesPerRow, |
3412 | double PrefetchSourceLinesY, |
3413 | unsigned int SwathWidthY, |
3414 | unsigned int VInitPreFillY, |
3415 | unsigned int MaxNumSwathY, |
3416 | double PrefetchSourceLinesC, |
3417 | unsigned int SwathWidthC, |
3418 | unsigned int VInitPreFillC, |
3419 | unsigned int MaxNumSwathC, |
3420 | unsigned int swath_width_luma_ub, |
3421 | unsigned int swath_width_chroma_ub, |
3422 | unsigned int SwathHeightY, |
3423 | unsigned int SwathHeightC, |
3424 | double TWait, |
3425 | double TPreReq, |
3426 | bool ExtendPrefetchIfPossible, |
3427 | /* Output */ |
3428 | double *DSTXAfterScaler, |
3429 | double *DSTYAfterScaler, |
3430 | double *DestinationLinesForPrefetch, |
3431 | double *PrefetchBandwidth, |
3432 | double *DestinationLinesToRequestVMInVBlank, |
3433 | double *DestinationLinesToRequestRowInVBlank, |
3434 | double *VRatioPrefetchY, |
3435 | double *VRatioPrefetchC, |
3436 | double *RequiredPrefetchPixDataBWLuma, |
3437 | double *RequiredPrefetchPixDataBWChroma, |
3438 | bool *NotEnoughTimeForDynamicMetadata, |
3439 | double *Tno_bw, |
3440 | double *prefetch_vmrow_bw, |
3441 | double *Tdmdl_vm, |
3442 | double *Tdmdl, |
3443 | double *TSetup, |
3444 | unsigned int *VUpdateOffsetPix, |
3445 | double *VUpdateWidthPix, |
3446 | double *VReadyOffsetPix) |
3447 | { |
3448 | double DPPCLKDelaySubtotalPlusCNVCFormater = v->DPPCLKDelaySubtotal + v->DPPCLKDelayCNVCFormater; |
3449 | bool MyError = false; |
3450 | unsigned int DPPCycles, DISPCLKCycles; |
3451 | double DSTTotalPixelsAfterScaler; |
3452 | double LineTime; |
3453 | double dst_y_prefetch_equ; |
3454 | double prefetch_bw_oto; |
3455 | double Tvm_oto; |
3456 | double Tr0_oto; |
3457 | double Tvm_oto_lines; |
3458 | double Tr0_oto_lines; |
3459 | double dst_y_prefetch_oto; |
3460 | double TimeForFetchingMetaPTE = 0; |
3461 | double TimeForFetchingRowInVBlank = 0; |
3462 | double LinesToRequestPrefetchPixelData = 0; |
3463 | double LinesForPrefetchBandwidth = 0; |
3464 | unsigned int HostVMDynamicLevelsTrips; |
3465 | double trip_to_mem; |
3466 | double Tvm_trips; |
3467 | double Tr0_trips; |
3468 | double Tvm_trips_rounded; |
3469 | double Tr0_trips_rounded; |
3470 | double Lsw_oto; |
3471 | double Tpre_rounded; |
3472 | double prefetch_bw_equ; |
3473 | double Tvm_equ; |
3474 | double Tr0_equ; |
3475 | double Tdmbf; |
3476 | double Tdmec; |
3477 | double Tdmsks; |
3478 | double prefetch_sw_bytes; |
3479 | double bytes_pp; |
3480 | double dep_bytes; |
3481 | unsigned int max_vratio_pre = v->MaxVRatioPre; |
3482 | double min_Lsw; |
3483 | double Tsw_est1 = 0; |
3484 | double Tsw_est3 = 0; |
3485 | |
3486 | if (v->GPUVMEnable == true && v->HostVMEnable == true) |
3487 | HostVMDynamicLevelsTrips = v->HostVMMaxNonCachedPageTableLevels; |
3488 | else |
3489 | HostVMDynamicLevelsTrips = 0; |
3490 | #ifdef __DML_VBA_DEBUG__ |
3491 | dml_print("DML::%s: v->GPUVMEnable = %d\n" , __func__, v->GPUVMEnable); |
3492 | dml_print("DML::%s: v->GPUVMMaxPageTableLevels = %d\n" , __func__, v->GPUVMMaxPageTableLevels); |
3493 | dml_print("DML::%s: DCCEnable = %d\n" , __func__, myPipe->DCCEnable); |
3494 | dml_print("DML::%s: v->HostVMEnable=%d HostVMInefficiencyFactor=%f\n" , |
3495 | __func__, v->HostVMEnable, HostVMInefficiencyFactor); |
3496 | #endif |
3497 | dml32_CalculateVUpdateAndDynamicMetadataParameters( |
3498 | MaxInterDCNTileRepeaters: v->MaxInterDCNTileRepeaters, |
3499 | Dppclk: myPipe->Dppclk, |
3500 | Dispclk: myPipe->Dispclk, |
3501 | DCFClkDeepSleep: myPipe->DCFClkDeepSleep, |
3502 | PixelClock: myPipe->PixelClock, |
3503 | HTotal: myPipe->HTotal, |
3504 | VBlank: myPipe->VBlank, |
3505 | DynamicMetadataTransmittedBytes: v->DynamicMetadataTransmittedBytes[k], |
3506 | DynamicMetadataLinesBeforeActiveRequired: v->DynamicMetadataLinesBeforeActiveRequired[k], |
3507 | InterlaceEnable: myPipe->InterlaceEnable, |
3508 | ProgressiveToInterlaceUnitInOPP: myPipe->ProgressiveToInterlaceUnitInOPP, |
3509 | TSetup, |
3510 | |
3511 | /* output */ |
3512 | Tdmbf: &Tdmbf, |
3513 | Tdmec: &Tdmec, |
3514 | Tdmsks: &Tdmsks, |
3515 | VUpdateOffsetPix, |
3516 | VUpdateWidthPix, |
3517 | VReadyOffsetPix); |
3518 | |
3519 | LineTime = myPipe->HTotal / myPipe->PixelClock; |
3520 | trip_to_mem = UrgentLatency; |
3521 | Tvm_trips = UrgentExtraLatency + trip_to_mem * (v->GPUVMMaxPageTableLevels * (HostVMDynamicLevelsTrips + 1) - 1); |
3522 | |
3523 | if (v->DynamicMetadataVMEnabled == true) |
3524 | *Tdmdl = TWait + Tvm_trips + trip_to_mem; |
3525 | else |
3526 | *Tdmdl = TWait + UrgentExtraLatency; |
3527 | |
3528 | #ifdef __DML_VBA_ALLOW_DELTA__ |
3529 | if (v->DynamicMetadataEnable[k] == false) |
3530 | *Tdmdl = 0.0; |
3531 | #endif |
3532 | |
3533 | if (v->DynamicMetadataEnable[k] == true) { |
3534 | if (VStartup * LineTime < *TSetup + *Tdmdl + Tdmbf + Tdmec + Tdmsks) { |
3535 | *NotEnoughTimeForDynamicMetadata = true; |
3536 | #ifdef __DML_VBA_DEBUG__ |
3537 | dml_print("DML::%s: Not Enough Time for Dynamic Meta!\n" , __func__); |
3538 | dml_print("DML::%s: Tdmbf: %fus - time for dmd transfer from dchub to dio output buffer\n" , |
3539 | __func__, Tdmbf); |
3540 | dml_print("DML::%s: Tdmec: %fus - time dio takes to transfer dmd\n" , __func__, Tdmec); |
3541 | dml_print("DML::%s: Tdmsks: %fus - time before active dmd must complete transmission at dio\n" , |
3542 | __func__, Tdmsks); |
3543 | dml_print("DML::%s: Tdmdl: %fus - time for fabric to become ready and fetch dmd\n" , |
3544 | __func__, *Tdmdl); |
3545 | #endif |
3546 | } else { |
3547 | *NotEnoughTimeForDynamicMetadata = false; |
3548 | } |
3549 | } else { |
3550 | *NotEnoughTimeForDynamicMetadata = false; |
3551 | } |
3552 | |
3553 | *Tdmdl_vm = (v->DynamicMetadataEnable[k] == true && v->DynamicMetadataVMEnabled == true && |
3554 | v->GPUVMEnable == true ? TWait + Tvm_trips : 0); |
3555 | |
3556 | if (myPipe->ScalerEnabled) |
3557 | DPPCycles = DPPCLKDelaySubtotalPlusCNVCFormater + v->DPPCLKDelaySCL; |
3558 | else |
3559 | DPPCycles = DPPCLKDelaySubtotalPlusCNVCFormater + v->DPPCLKDelaySCLLBOnly; |
3560 | |
3561 | DPPCycles = DPPCycles + myPipe->NumberOfCursors * v->DPPCLKDelayCNVCCursor; |
3562 | |
3563 | DISPCLKCycles = v->DISPCLKDelaySubtotal; |
3564 | |
3565 | if (myPipe->Dppclk == 0.0 || myPipe->Dispclk == 0.0) |
3566 | return true; |
3567 | |
3568 | *DSTXAfterScaler = DPPCycles * myPipe->PixelClock / myPipe->Dppclk + DISPCLKCycles * |
3569 | myPipe->PixelClock / myPipe->Dispclk + DSCDelay; |
3570 | |
3571 | *DSTXAfterScaler = *DSTXAfterScaler + (myPipe->ODMMode != dm_odm_combine_mode_disabled ? 18 : 0) |
3572 | + (myPipe->DPPPerSurface - 1) * DPP_RECOUT_WIDTH |
3573 | + ((myPipe->ODMMode == dm_odm_split_mode_1to2 || myPipe->ODMMode == dm_odm_mode_mso_1to2) ? |
3574 | myPipe->HActive / 2 : 0) |
3575 | + ((myPipe->ODMMode == dm_odm_mode_mso_1to4) ? myPipe->HActive * 3 / 4 : 0); |
3576 | |
3577 | #ifdef __DML_VBA_DEBUG__ |
3578 | dml_print("DML::%s: DPPCycles: %d\n" , __func__, DPPCycles); |
3579 | dml_print("DML::%s: PixelClock: %f\n" , __func__, myPipe->PixelClock); |
3580 | dml_print("DML::%s: Dppclk: %f\n" , __func__, myPipe->Dppclk); |
3581 | dml_print("DML::%s: DISPCLKCycles: %d\n" , __func__, DISPCLKCycles); |
3582 | dml_print("DML::%s: DISPCLK: %f\n" , __func__, myPipe->Dispclk); |
3583 | dml_print("DML::%s: DSCDelay: %d\n" , __func__, DSCDelay); |
3584 | dml_print("DML::%s: ODMMode: %d\n" , __func__, myPipe->ODMMode); |
3585 | dml_print("DML::%s: DPP_RECOUT_WIDTH: %d\n" , __func__, DPP_RECOUT_WIDTH); |
3586 | dml_print("DML::%s: DSTXAfterScaler: %d\n" , __func__, *DSTXAfterScaler); |
3587 | #endif |
3588 | |
3589 | if (v->OutputFormat[k] == dm_420 || (myPipe->InterlaceEnable && myPipe->ProgressiveToInterlaceUnitInOPP)) |
3590 | *DSTYAfterScaler = 1; |
3591 | else |
3592 | *DSTYAfterScaler = 0; |
3593 | |
3594 | DSTTotalPixelsAfterScaler = *DSTYAfterScaler * myPipe->HTotal + *DSTXAfterScaler; |
3595 | *DSTYAfterScaler = dml_floor(a: DSTTotalPixelsAfterScaler / myPipe->HTotal, granularity: 1); |
3596 | *DSTXAfterScaler = DSTTotalPixelsAfterScaler - ((double) (*DSTYAfterScaler * myPipe->HTotal)); |
3597 | #ifdef __DML_VBA_DEBUG__ |
3598 | dml_print("DML::%s: DSTXAfterScaler: %d (final)\n" , __func__, *DSTXAfterScaler); |
3599 | dml_print("DML::%s: DSTYAfterScaler: %d (final)\n" , __func__, *DSTYAfterScaler); |
3600 | #endif |
3601 | |
3602 | MyError = false; |
3603 | |
3604 | Tr0_trips = trip_to_mem * (HostVMDynamicLevelsTrips + 1); |
3605 | |
3606 | if (v->GPUVMEnable == true) { |
3607 | Tvm_trips_rounded = dml_ceil(a: 4.0 * Tvm_trips / LineTime, granularity: 1.0) / 4.0 * LineTime; |
3608 | Tr0_trips_rounded = dml_ceil(a: 4.0 * Tr0_trips / LineTime, granularity: 1.0) / 4.0 * LineTime; |
3609 | if (v->GPUVMMaxPageTableLevels >= 3) { |
3610 | *Tno_bw = UrgentExtraLatency + trip_to_mem * |
3611 | (double) ((v->GPUVMMaxPageTableLevels - 2) * (HostVMDynamicLevelsTrips + 1) - 1); |
3612 | } else if (v->GPUVMMaxPageTableLevels == 1 && myPipe->DCCEnable != true) { |
3613 | Tr0_trips_rounded = dml_ceil(a: 4.0 * UrgentExtraLatency / LineTime, granularity: 1.0) / |
3614 | 4.0 * LineTime; // VBA_ERROR |
3615 | *Tno_bw = UrgentExtraLatency; |
3616 | } else { |
3617 | *Tno_bw = 0; |
3618 | } |
3619 | } else if (myPipe->DCCEnable == true) { |
3620 | Tvm_trips_rounded = LineTime / 4.0; |
3621 | Tr0_trips_rounded = dml_ceil(a: 4.0 * Tr0_trips / LineTime, granularity: 1.0) / 4.0 * LineTime; |
3622 | *Tno_bw = 0; |
3623 | } else { |
3624 | Tvm_trips_rounded = LineTime / 4.0; |
3625 | Tr0_trips_rounded = LineTime / 2.0; |
3626 | *Tno_bw = 0; |
3627 | } |
3628 | Tvm_trips_rounded = dml_max(a: Tvm_trips_rounded, b: LineTime / 4.0); |
3629 | Tr0_trips_rounded = dml_max(a: Tr0_trips_rounded, b: LineTime / 4.0); |
3630 | |
3631 | if (myPipe->SourcePixelFormat == dm_420_8 || myPipe->SourcePixelFormat == dm_420_10 |
3632 | || myPipe->SourcePixelFormat == dm_420_12) { |
3633 | bytes_pp = myPipe->BytePerPixelY + myPipe->BytePerPixelC / 4; |
3634 | } else { |
3635 | bytes_pp = myPipe->BytePerPixelY + myPipe->BytePerPixelC; |
3636 | } |
3637 | |
3638 | prefetch_sw_bytes = PrefetchSourceLinesY * swath_width_luma_ub * myPipe->BytePerPixelY |
3639 | + PrefetchSourceLinesC * swath_width_chroma_ub * myPipe->BytePerPixelC; |
3640 | prefetch_bw_oto = dml_max(a: bytes_pp * myPipe->PixelClock / myPipe->DPPPerSurface, |
3641 | b: prefetch_sw_bytes / (dml_max(a: PrefetchSourceLinesY, b: PrefetchSourceLinesC) * LineTime)); |
3642 | |
3643 | min_Lsw = dml_max(a: PrefetchSourceLinesY, b: PrefetchSourceLinesC) / max_vratio_pre; |
3644 | min_Lsw = dml_max(a: min_Lsw, b: 1.0); |
3645 | Lsw_oto = dml_ceil(a: 4.0 * dml_max(a: prefetch_sw_bytes / prefetch_bw_oto / LineTime, b: min_Lsw), granularity: 1.0) / 4.0; |
3646 | |
3647 | if (v->GPUVMEnable == true) { |
3648 | Tvm_oto = dml_max3( |
3649 | a: Tvm_trips, |
3650 | b: *Tno_bw + PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor / prefetch_bw_oto, |
3651 | c: LineTime / 4.0); |
3652 | } else |
3653 | Tvm_oto = LineTime / 4.0; |
3654 | |
3655 | if ((v->GPUVMEnable == true || myPipe->DCCEnable == true)) { |
3656 | Tr0_oto = dml_max4( |
3657 | a: Tr0_trips, |
3658 | b: (MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor) / prefetch_bw_oto, |
3659 | c: (LineTime - Tvm_oto)/2.0, |
3660 | d: LineTime / 4.0); |
3661 | #ifdef __DML_VBA_DEBUG__ |
3662 | dml_print("DML::%s: Tr0_oto max0 = %f\n" , __func__, |
3663 | (MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor) / prefetch_bw_oto); |
3664 | dml_print("DML::%s: Tr0_oto max1 = %f\n" , __func__, Tr0_trips); |
3665 | dml_print("DML::%s: Tr0_oto max2 = %f\n" , __func__, LineTime - Tvm_oto); |
3666 | dml_print("DML::%s: Tr0_oto max3 = %f\n" , __func__, LineTime / 4); |
3667 | #endif |
3668 | } else |
3669 | Tr0_oto = (LineTime - Tvm_oto) / 2.0; |
3670 | |
3671 | Tvm_oto_lines = dml_ceil(a: 4.0 * Tvm_oto / LineTime, granularity: 1) / 4.0; |
3672 | Tr0_oto_lines = dml_ceil(a: 4.0 * Tr0_oto / LineTime, granularity: 1) / 4.0; |
3673 | dst_y_prefetch_oto = Tvm_oto_lines + 2 * Tr0_oto_lines + Lsw_oto; |
3674 | |
3675 | dst_y_prefetch_equ = VStartup - (*TSetup + dml_max(a: TWait + TCalc, b: *Tdmdl)) / LineTime - |
3676 | (*DSTYAfterScaler + (double) *DSTXAfterScaler / (double) myPipe->HTotal); |
3677 | |
3678 | dst_y_prefetch_equ = dml_min(a: dst_y_prefetch_equ, __DML_VBA_MAX_DST_Y_PRE__); |
3679 | #ifdef __DML_VBA_DEBUG__ |
3680 | dml_print("DML::%s: HTotal = %d\n" , __func__, myPipe->HTotal); |
3681 | dml_print("DML::%s: min_Lsw = %f\n" , __func__, min_Lsw); |
3682 | dml_print("DML::%s: *Tno_bw = %f\n" , __func__, *Tno_bw); |
3683 | dml_print("DML::%s: UrgentExtraLatency = %f\n" , __func__, UrgentExtraLatency); |
3684 | dml_print("DML::%s: trip_to_mem = %f\n" , __func__, trip_to_mem); |
3685 | dml_print("DML::%s: BytePerPixelY = %d\n" , __func__, myPipe->BytePerPixelY); |
3686 | dml_print("DML::%s: PrefetchSourceLinesY = %f\n" , __func__, PrefetchSourceLinesY); |
3687 | dml_print("DML::%s: swath_width_luma_ub = %d\n" , __func__, swath_width_luma_ub); |
3688 | dml_print("DML::%s: BytePerPixelC = %d\n" , __func__, myPipe->BytePerPixelC); |
3689 | dml_print("DML::%s: PrefetchSourceLinesC = %f\n" , __func__, PrefetchSourceLinesC); |
3690 | dml_print("DML::%s: swath_width_chroma_ub = %d\n" , __func__, swath_width_chroma_ub); |
3691 | dml_print("DML::%s: prefetch_sw_bytes = %f\n" , __func__, prefetch_sw_bytes); |
3692 | dml_print("DML::%s: bytes_pp = %f\n" , __func__, bytes_pp); |
3693 | dml_print("DML::%s: PDEAndMetaPTEBytesFrame = %d\n" , __func__, PDEAndMetaPTEBytesFrame); |
3694 | dml_print("DML::%s: MetaRowByte = %d\n" , __func__, MetaRowByte); |
3695 | dml_print("DML::%s: PixelPTEBytesPerRow = %d\n" , __func__, PixelPTEBytesPerRow); |
3696 | dml_print("DML::%s: HostVMInefficiencyFactor = %f\n" , __func__, HostVMInefficiencyFactor); |
3697 | dml_print("DML::%s: Tvm_trips = %f\n" , __func__, Tvm_trips); |
3698 | dml_print("DML::%s: Tr0_trips = %f\n" , __func__, Tr0_trips); |
3699 | dml_print("DML::%s: prefetch_bw_oto = %f\n" , __func__, prefetch_bw_oto); |
3700 | dml_print("DML::%s: Tr0_oto = %f\n" , __func__, Tr0_oto); |
3701 | dml_print("DML::%s: Tvm_oto = %f\n" , __func__, Tvm_oto); |
3702 | dml_print("DML::%s: Tvm_oto_lines = %f\n" , __func__, Tvm_oto_lines); |
3703 | dml_print("DML::%s: Tr0_oto_lines = %f\n" , __func__, Tr0_oto_lines); |
3704 | dml_print("DML::%s: Lsw_oto = %f\n" , __func__, Lsw_oto); |
3705 | dml_print("DML::%s: dst_y_prefetch_oto = %f\n" , __func__, dst_y_prefetch_oto); |
3706 | dml_print("DML::%s: dst_y_prefetch_equ = %f\n" , __func__, dst_y_prefetch_equ); |
3707 | #endif |
3708 | |
3709 | dst_y_prefetch_equ = dml_floor(a: 4.0 * (dst_y_prefetch_equ + 0.125), granularity: 1) / 4.0; |
3710 | Tpre_rounded = dst_y_prefetch_equ * LineTime; |
3711 | #ifdef __DML_VBA_DEBUG__ |
3712 | dml_print("DML::%s: dst_y_prefetch_equ: %f (after round)\n" , __func__, dst_y_prefetch_equ); |
3713 | dml_print("DML::%s: LineTime: %f\n" , __func__, LineTime); |
3714 | dml_print("DML::%s: VStartup: %d\n" , __func__, VStartup); |
3715 | dml_print("DML::%s: Tvstartup: %fus - time between vstartup and first pixel of active\n" , |
3716 | __func__, VStartup * LineTime); |
3717 | dml_print("DML::%s: TSetup: %fus - time from vstartup to vready\n" , __func__, *TSetup); |
3718 | dml_print("DML::%s: TCalc: %fus - time for calculations in dchub starting at vready\n" , __func__, TCalc); |
3719 | dml_print("DML::%s: Tdmbf: %fus - time for dmd transfer from dchub to dio output buffer\n" , __func__, Tdmbf); |
3720 | dml_print("DML::%s: Tdmec: %fus - time dio takes to transfer dmd\n" , __func__, Tdmec); |
3721 | dml_print("DML::%s: Tdmdl_vm: %fus - time for vm stages of dmd\n" , __func__, *Tdmdl_vm); |
3722 | dml_print("DML::%s: Tdmdl: %fus - time for fabric to become ready and fetch dmd\n" , __func__, *Tdmdl); |
3723 | dml_print("DML::%s: DSTYAfterScaler: %d lines - number of lines of pipeline and buffer delay after scaler\n" , |
3724 | __func__, *DSTYAfterScaler); |
3725 | #endif |
3726 | dep_bytes = dml_max(a: PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor, |
3727 | b: MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor); |
3728 | |
3729 | if (prefetch_sw_bytes < dep_bytes) |
3730 | prefetch_sw_bytes = 2 * dep_bytes; |
3731 | |
3732 | *PrefetchBandwidth = 0; |
3733 | *DestinationLinesToRequestVMInVBlank = 0; |
3734 | *DestinationLinesToRequestRowInVBlank = 0; |
3735 | *VRatioPrefetchY = 0; |
3736 | *VRatioPrefetchC = 0; |
3737 | *RequiredPrefetchPixDataBWLuma = 0; |
3738 | if (dst_y_prefetch_equ > 1 && |
3739 | (Tpre_rounded >= TPreReq || dst_y_prefetch_equ == __DML_VBA_MAX_DST_Y_PRE__)) { |
3740 | double PrefetchBandwidth1; |
3741 | double PrefetchBandwidth2; |
3742 | double PrefetchBandwidth3; |
3743 | double PrefetchBandwidth4; |
3744 | |
3745 | if (Tpre_rounded - *Tno_bw > 0) { |
3746 | PrefetchBandwidth1 = (PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor + 2 * MetaRowByte |
3747 | + 2 * PixelPTEBytesPerRow * HostVMInefficiencyFactor |
3748 | + prefetch_sw_bytes) / (Tpre_rounded - *Tno_bw); |
3749 | Tsw_est1 = prefetch_sw_bytes / PrefetchBandwidth1; |
3750 | } else |
3751 | PrefetchBandwidth1 = 0; |
3752 | |
3753 | if (VStartup == MaxVStartup && (Tsw_est1 / LineTime < min_Lsw) |
3754 | && Tpre_rounded - min_Lsw * LineTime - 0.75 * LineTime - *Tno_bw > 0) { |
3755 | PrefetchBandwidth1 = (PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor + 2 * MetaRowByte |
3756 | + 2 * PixelPTEBytesPerRow * HostVMInefficiencyFactor) |
3757 | / (Tpre_rounded - min_Lsw * LineTime - 0.75 * LineTime - *Tno_bw); |
3758 | } |
3759 | |
3760 | if (Tpre_rounded - *Tno_bw - 2 * Tr0_trips_rounded > 0) |
3761 | PrefetchBandwidth2 = (PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor + prefetch_sw_bytes) / |
3762 | (Tpre_rounded - *Tno_bw - 2 * Tr0_trips_rounded); |
3763 | else |
3764 | PrefetchBandwidth2 = 0; |
3765 | |
3766 | if (Tpre_rounded - Tvm_trips_rounded > 0) { |
3767 | PrefetchBandwidth3 = (2 * MetaRowByte + 2 * PixelPTEBytesPerRow * HostVMInefficiencyFactor |
3768 | + prefetch_sw_bytes) / (Tpre_rounded - Tvm_trips_rounded); |
3769 | Tsw_est3 = prefetch_sw_bytes / PrefetchBandwidth3; |
3770 | } else |
3771 | PrefetchBandwidth3 = 0; |
3772 | |
3773 | |
3774 | if (VStartup == MaxVStartup && |
3775 | (Tsw_est3 / LineTime < min_Lsw) && Tpre_rounded - min_Lsw * LineTime - 0.75 * |
3776 | LineTime - Tvm_trips_rounded > 0) { |
3777 | PrefetchBandwidth3 = (2 * MetaRowByte + 2 * PixelPTEBytesPerRow * HostVMInefficiencyFactor) |
3778 | / (Tpre_rounded - min_Lsw * LineTime - 0.75 * LineTime - Tvm_trips_rounded); |
3779 | } |
3780 | |
3781 | if (Tpre_rounded - Tvm_trips_rounded - 2 * Tr0_trips_rounded > 0) { |
3782 | PrefetchBandwidth4 = prefetch_sw_bytes / |
3783 | (Tpre_rounded - Tvm_trips_rounded - 2 * Tr0_trips_rounded); |
3784 | } else { |
3785 | PrefetchBandwidth4 = 0; |
3786 | } |
3787 | |
3788 | #ifdef __DML_VBA_DEBUG__ |
3789 | dml_print("DML::%s: Tpre_rounded: %f\n" , __func__, Tpre_rounded); |
3790 | dml_print("DML::%s: Tno_bw: %f\n" , __func__, *Tno_bw); |
3791 | dml_print("DML::%s: Tvm_trips_rounded: %f\n" , __func__, Tvm_trips_rounded); |
3792 | dml_print("DML::%s: Tsw_est1: %f\n" , __func__, Tsw_est1); |
3793 | dml_print("DML::%s: Tsw_est3: %f\n" , __func__, Tsw_est3); |
3794 | dml_print("DML::%s: PrefetchBandwidth1: %f\n" , __func__, PrefetchBandwidth1); |
3795 | dml_print("DML::%s: PrefetchBandwidth2: %f\n" , __func__, PrefetchBandwidth2); |
3796 | dml_print("DML::%s: PrefetchBandwidth3: %f\n" , __func__, PrefetchBandwidth3); |
3797 | dml_print("DML::%s: PrefetchBandwidth4: %f\n" , __func__, PrefetchBandwidth4); |
3798 | #endif |
3799 | { |
3800 | bool Case1OK; |
3801 | bool Case2OK; |
3802 | bool Case3OK; |
3803 | |
3804 | if (PrefetchBandwidth1 > 0) { |
3805 | if (*Tno_bw + PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor / PrefetchBandwidth1 |
3806 | >= Tvm_trips_rounded |
3807 | && (MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor) |
3808 | / PrefetchBandwidth1 >= Tr0_trips_rounded) { |
3809 | Case1OK = true; |
3810 | } else { |
3811 | Case1OK = false; |
3812 | } |
3813 | } else { |
3814 | Case1OK = false; |
3815 | } |
3816 | |
3817 | if (PrefetchBandwidth2 > 0) { |
3818 | if (*Tno_bw + PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor / PrefetchBandwidth2 |
3819 | >= Tvm_trips_rounded |
3820 | && (MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor) |
3821 | / PrefetchBandwidth2 < Tr0_trips_rounded) { |
3822 | Case2OK = true; |
3823 | } else { |
3824 | Case2OK = false; |
3825 | } |
3826 | } else { |
3827 | Case2OK = false; |
3828 | } |
3829 | |
3830 | if (PrefetchBandwidth3 > 0) { |
3831 | if (*Tno_bw + PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor / PrefetchBandwidth3 < |
3832 | Tvm_trips_rounded && (MetaRowByte + PixelPTEBytesPerRow * |
3833 | HostVMInefficiencyFactor) / PrefetchBandwidth3 >= |
3834 | Tr0_trips_rounded) { |
3835 | Case3OK = true; |
3836 | } else { |
3837 | Case3OK = false; |
3838 | } |
3839 | } else { |
3840 | Case3OK = false; |
3841 | } |
3842 | |
3843 | if (Case1OK) |
3844 | prefetch_bw_equ = PrefetchBandwidth1; |
3845 | else if (Case2OK) |
3846 | prefetch_bw_equ = PrefetchBandwidth2; |
3847 | else if (Case3OK) |
3848 | prefetch_bw_equ = PrefetchBandwidth3; |
3849 | else |
3850 | prefetch_bw_equ = PrefetchBandwidth4; |
3851 | |
3852 | #ifdef __DML_VBA_DEBUG__ |
3853 | dml_print("DML::%s: Case1OK: %d\n" , __func__, Case1OK); |
3854 | dml_print("DML::%s: Case2OK: %d\n" , __func__, Case2OK); |
3855 | dml_print("DML::%s: Case3OK: %d\n" , __func__, Case3OK); |
3856 | dml_print("DML::%s: prefetch_bw_equ: %f\n" , __func__, prefetch_bw_equ); |
3857 | #endif |
3858 | |
3859 | if (prefetch_bw_equ > 0) { |
3860 | if (v->GPUVMEnable == true) { |
3861 | Tvm_equ = dml_max3(a: *Tno_bw + PDEAndMetaPTEBytesFrame * |
3862 | HostVMInefficiencyFactor / prefetch_bw_equ, |
3863 | b: Tvm_trips, c: LineTime / 4); |
3864 | } else { |
3865 | Tvm_equ = LineTime / 4; |
3866 | } |
3867 | |
3868 | if ((v->GPUVMEnable == true || myPipe->DCCEnable == true)) { |
3869 | Tr0_equ = dml_max4(a: (MetaRowByte + PixelPTEBytesPerRow * |
3870 | HostVMInefficiencyFactor) / prefetch_bw_equ, b: Tr0_trips, |
3871 | c: (LineTime - Tvm_equ) / 2, d: LineTime / 4); |
3872 | } else { |
3873 | Tr0_equ = (LineTime - Tvm_equ) / 2; |
3874 | } |
3875 | } else { |
3876 | Tvm_equ = 0; |
3877 | Tr0_equ = 0; |
3878 | #ifdef __DML_VBA_DEBUG__ |
3879 | dml_print("DML: prefetch_bw_equ equals 0! %s:%d\n" , __FILE__, __LINE__); |
3880 | #endif |
3881 | } |
3882 | } |
3883 | |
3884 | if (dst_y_prefetch_oto < dst_y_prefetch_equ) { |
3885 | if (dst_y_prefetch_oto * LineTime < TPreReq) { |
3886 | *DestinationLinesForPrefetch = dst_y_prefetch_equ; |
3887 | } else { |
3888 | *DestinationLinesForPrefetch = dst_y_prefetch_oto; |
3889 | } |
3890 | TimeForFetchingMetaPTE = Tvm_oto; |
3891 | TimeForFetchingRowInVBlank = Tr0_oto; |
3892 | *PrefetchBandwidth = prefetch_bw_oto; |
3893 | /* Clamp to oto for bandwidth calculation */ |
3894 | LinesForPrefetchBandwidth = dst_y_prefetch_oto; |
3895 | } else { |
3896 | /* For mode programming we want to extend the prefetch as much as possible |
3897 | * (up to oto, or as long as we can for equ) if we're not already applying |
3898 | * the 60us prefetch requirement. This is to avoid intermittent underflow |
3899 | * issues during prefetch. |
3900 | * |
3901 | * The prefetch extension is applied under the following scenarios: |
3902 | * 1. We're in prefetch mode > 0 (i.e. we don't support MCLK switch in blank) |
3903 | * 2. We're using subvp or drr methods of p-state switch, in which case we |
3904 | * we don't care if prefetch takes up more of the blanking time |
3905 | * |
3906 | * Mode programming typically chooses the smallest prefetch time possible |
3907 | * (i.e. highest bandwidth during prefetch) presumably to create margin between |
3908 | * p-states / c-states that happen in vblank and prefetch. Therefore we only |
3909 | * apply this prefetch extension when p-state in vblank is not required (UCLK |
3910 | * p-states take up the most vblank time). |
3911 | */ |
3912 | if (ExtendPrefetchIfPossible && TPreReq == 0 && VStartup < MaxVStartup) { |
3913 | MyError = true; |
3914 | } else { |
3915 | *DestinationLinesForPrefetch = dst_y_prefetch_equ; |
3916 | TimeForFetchingMetaPTE = Tvm_equ; |
3917 | TimeForFetchingRowInVBlank = Tr0_equ; |
3918 | *PrefetchBandwidth = prefetch_bw_equ; |
3919 | /* Clamp to equ for bandwidth calculation */ |
3920 | LinesForPrefetchBandwidth = dst_y_prefetch_equ; |
3921 | } |
3922 | } |
3923 | |
3924 | *DestinationLinesToRequestVMInVBlank = dml_ceil(a: 4.0 * TimeForFetchingMetaPTE / LineTime, granularity: 1.0) / 4.0; |
3925 | |
3926 | *DestinationLinesToRequestRowInVBlank = |
3927 | dml_ceil(a: 4.0 * TimeForFetchingRowInVBlank / LineTime, granularity: 1.0) / 4.0; |
3928 | |
3929 | LinesToRequestPrefetchPixelData = LinesForPrefetchBandwidth - |
3930 | *DestinationLinesToRequestVMInVBlank - 2 * *DestinationLinesToRequestRowInVBlank; |
3931 | |
3932 | #ifdef __DML_VBA_DEBUG__ |
3933 | dml_print("DML::%s: DestinationLinesForPrefetch = %f\n" , __func__, *DestinationLinesForPrefetch); |
3934 | dml_print("DML::%s: DestinationLinesToRequestVMInVBlank = %f\n" , |
3935 | __func__, *DestinationLinesToRequestVMInVBlank); |
3936 | dml_print("DML::%s: TimeForFetchingRowInVBlank = %f\n" , __func__, TimeForFetchingRowInVBlank); |
3937 | dml_print("DML::%s: LineTime = %f\n" , __func__, LineTime); |
3938 | dml_print("DML::%s: DestinationLinesToRequestRowInVBlank = %f\n" , |
3939 | __func__, *DestinationLinesToRequestRowInVBlank); |
3940 | dml_print("DML::%s: PrefetchSourceLinesY = %f\n" , __func__, PrefetchSourceLinesY); |
3941 | dml_print("DML::%s: LinesToRequestPrefetchPixelData = %f\n" , __func__, LinesToRequestPrefetchPixelData); |
3942 | #endif |
3943 | |
3944 | if (LinesToRequestPrefetchPixelData >= 1 && prefetch_bw_equ > 0) { |
3945 | *VRatioPrefetchY = (double) PrefetchSourceLinesY / LinesToRequestPrefetchPixelData; |
3946 | *VRatioPrefetchY = dml_max(a: *VRatioPrefetchY, b: 1.0); |
3947 | #ifdef __DML_VBA_DEBUG__ |
3948 | dml_print("DML::%s: VRatioPrefetchY = %f\n" , __func__, *VRatioPrefetchY); |
3949 | dml_print("DML::%s: SwathHeightY = %d\n" , __func__, SwathHeightY); |
3950 | dml_print("DML::%s: VInitPreFillY = %d\n" , __func__, VInitPreFillY); |
3951 | #endif |
3952 | if ((SwathHeightY > 4) && (VInitPreFillY > 3)) { |
3953 | if (LinesToRequestPrefetchPixelData > (VInitPreFillY - 3.0) / 2.0) { |
3954 | *VRatioPrefetchY = |
3955 | dml_max(a: (double) PrefetchSourceLinesY / |
3956 | LinesToRequestPrefetchPixelData, |
3957 | b: (double) MaxNumSwathY * SwathHeightY / |
3958 | (LinesToRequestPrefetchPixelData - |
3959 | (VInitPreFillY - 3.0) / 2.0)); |
3960 | *VRatioPrefetchY = dml_max(a: *VRatioPrefetchY, b: 1.0); |
3961 | } else { |
3962 | MyError = true; |
3963 | *VRatioPrefetchY = 0; |
3964 | } |
3965 | #ifdef __DML_VBA_DEBUG__ |
3966 | dml_print("DML::%s: VRatioPrefetchY = %f\n" , __func__, *VRatioPrefetchY); |
3967 | dml_print("DML::%s: PrefetchSourceLinesY = %f\n" , __func__, PrefetchSourceLinesY); |
3968 | dml_print("DML::%s: MaxNumSwathY = %d\n" , __func__, MaxNumSwathY); |
3969 | #endif |
3970 | } |
3971 | |
3972 | *VRatioPrefetchC = (double) PrefetchSourceLinesC / LinesToRequestPrefetchPixelData; |
3973 | *VRatioPrefetchC = dml_max(a: *VRatioPrefetchC, b: 1.0); |
3974 | |
3975 | #ifdef __DML_VBA_DEBUG__ |
3976 | dml_print("DML::%s: VRatioPrefetchC = %f\n" , __func__, *VRatioPrefetchC); |
3977 | dml_print("DML::%s: SwathHeightC = %d\n" , __func__, SwathHeightC); |
3978 | dml_print("DML::%s: VInitPreFillC = %d\n" , __func__, VInitPreFillC); |
3979 | #endif |
3980 | if ((SwathHeightC > 4)) { |
3981 | if (LinesToRequestPrefetchPixelData > (VInitPreFillC - 3.0) / 2.0) { |
3982 | *VRatioPrefetchC = |
3983 | dml_max(a: *VRatioPrefetchC, |
3984 | b: (double) MaxNumSwathC * SwathHeightC / |
3985 | (LinesToRequestPrefetchPixelData - |
3986 | (VInitPreFillC - 3.0) / 2.0)); |
3987 | *VRatioPrefetchC = dml_max(a: *VRatioPrefetchC, b: 1.0); |
3988 | } else { |
3989 | MyError = true; |
3990 | *VRatioPrefetchC = 0; |
3991 | } |
3992 | #ifdef __DML_VBA_DEBUG__ |
3993 | dml_print("DML::%s: VRatioPrefetchC = %f\n" , __func__, *VRatioPrefetchC); |
3994 | dml_print("DML::%s: PrefetchSourceLinesC = %f\n" , __func__, PrefetchSourceLinesC); |
3995 | dml_print("DML::%s: MaxNumSwathC = %d\n" , __func__, MaxNumSwathC); |
3996 | #endif |
3997 | } |
3998 | |
3999 | *RequiredPrefetchPixDataBWLuma = (double) PrefetchSourceLinesY |
4000 | / LinesToRequestPrefetchPixelData * myPipe->BytePerPixelY * swath_width_luma_ub |
4001 | / LineTime; |
4002 | |
4003 | #ifdef __DML_VBA_DEBUG__ |
4004 | dml_print("DML::%s: BytePerPixelY = %d\n" , __func__, myPipe->BytePerPixelY); |
4005 | dml_print("DML::%s: swath_width_luma_ub = %d\n" , __func__, swath_width_luma_ub); |
4006 | dml_print("DML::%s: LineTime = %f\n" , __func__, LineTime); |
4007 | dml_print("DML::%s: RequiredPrefetchPixDataBWLuma = %f\n" , |
4008 | __func__, *RequiredPrefetchPixDataBWLuma); |
4009 | #endif |
4010 | *RequiredPrefetchPixDataBWChroma = (double) PrefetchSourceLinesC / |
4011 | LinesToRequestPrefetchPixelData |
4012 | * myPipe->BytePerPixelC |
4013 | * swath_width_chroma_ub / LineTime; |
4014 | } else { |
4015 | MyError = true; |
4016 | #ifdef __DML_VBA_DEBUG__ |
4017 | dml_print("DML:%s: MyErr set. LinesToRequestPrefetchPixelData: %f, should be > 0\n" , |
4018 | __func__, LinesToRequestPrefetchPixelData); |
4019 | #endif |
4020 | *VRatioPrefetchY = 0; |
4021 | *VRatioPrefetchC = 0; |
4022 | *RequiredPrefetchPixDataBWLuma = 0; |
4023 | *RequiredPrefetchPixDataBWChroma = 0; |
4024 | } |
4025 | #ifdef __DML_VBA_DEBUG__ |
4026 | dml_print("DML: Tpre: %fus - sum of time to request meta pte, 2 x data pte + meta data, swaths\n" , |
4027 | (double)LinesToRequestPrefetchPixelData * LineTime + |
4028 | 2.0*TimeForFetchingRowInVBlank + TimeForFetchingMetaPTE); |
4029 | dml_print("DML: Tvm: %fus - time to fetch page tables for meta surface\n" , TimeForFetchingMetaPTE); |
4030 | dml_print("DML: To: %fus - time for propagation from scaler to optc\n" , |
4031 | (*DSTYAfterScaler + ((double) (*DSTXAfterScaler) / (double) myPipe->HTotal)) * LineTime); |
4032 | dml_print("DML: Tvstartup - TSetup - Tcalc - Twait - Tpre - To > 0\n" ); |
4033 | dml_print("DML: Tslack(pre): %fus - time left over in schedule\n" , VStartup * LineTime - |
4034 | TimeForFetchingMetaPTE - 2*TimeForFetchingRowInVBlank - (*DSTYAfterScaler + |
4035 | ((double) (*DSTXAfterScaler) / (double) myPipe->HTotal)) * LineTime - TWait - TCalc - *TSetup); |
4036 | dml_print("DML: row_bytes = dpte_row_bytes (per_pipe) = PixelPTEBytesPerRow = : %d\n" , |
4037 | PixelPTEBytesPerRow); |
4038 | #endif |
4039 | } else { |
4040 | MyError = true; |
4041 | #ifdef __DML_VBA_DEBUG__ |
4042 | dml_print("DML::%s: MyErr set, dst_y_prefetch_equ = %f (should be > 1)\n" , |
4043 | __func__, dst_y_prefetch_equ); |
4044 | #endif |
4045 | } |
4046 | |
4047 | { |
4048 | double prefetch_vm_bw; |
4049 | double prefetch_row_bw; |
4050 | |
4051 | if (PDEAndMetaPTEBytesFrame == 0) { |
4052 | prefetch_vm_bw = 0; |
4053 | } else if (*DestinationLinesToRequestVMInVBlank > 0) { |
4054 | #ifdef __DML_VBA_DEBUG__ |
4055 | dml_print("DML::%s: PDEAndMetaPTEBytesFrame = %d\n" , __func__, PDEAndMetaPTEBytesFrame); |
4056 | dml_print("DML::%s: HostVMInefficiencyFactor = %f\n" , __func__, HostVMInefficiencyFactor); |
4057 | dml_print("DML::%s: DestinationLinesToRequestVMInVBlank = %f\n" , |
4058 | __func__, *DestinationLinesToRequestVMInVBlank); |
4059 | dml_print("DML::%s: LineTime = %f\n" , __func__, LineTime); |
4060 | #endif |
4061 | prefetch_vm_bw = PDEAndMetaPTEBytesFrame * HostVMInefficiencyFactor / |
4062 | (*DestinationLinesToRequestVMInVBlank * LineTime); |
4063 | #ifdef __DML_VBA_DEBUG__ |
4064 | dml_print("DML::%s: prefetch_vm_bw = %f\n" , __func__, prefetch_vm_bw); |
4065 | #endif |
4066 | } else { |
4067 | prefetch_vm_bw = 0; |
4068 | MyError = true; |
4069 | #ifdef __DML_VBA_DEBUG__ |
4070 | dml_print("DML::%s: MyErr set. DestinationLinesToRequestVMInVBlank=%f (should be > 0)\n" , |
4071 | __func__, *DestinationLinesToRequestVMInVBlank); |
4072 | #endif |
4073 | } |
4074 | |
4075 | if (MetaRowByte + PixelPTEBytesPerRow == 0) { |
4076 | prefetch_row_bw = 0; |
4077 | } else if (*DestinationLinesToRequestRowInVBlank > 0) { |
4078 | prefetch_row_bw = (MetaRowByte + PixelPTEBytesPerRow * HostVMInefficiencyFactor) / |
4079 | (*DestinationLinesToRequestRowInVBlank * LineTime); |
4080 | |
4081 | #ifdef __DML_VBA_DEBUG__ |
4082 | dml_print("DML::%s: MetaRowByte = %d\n" , __func__, MetaRowByte); |
4083 | dml_print("DML::%s: PixelPTEBytesPerRow = %d\n" , __func__, PixelPTEBytesPerRow); |
4084 | dml_print("DML::%s: DestinationLinesToRequestRowInVBlank = %f\n" , |
4085 | __func__, *DestinationLinesToRequestRowInVBlank); |
4086 | dml_print("DML::%s: prefetch_row_bw = %f\n" , __func__, prefetch_row_bw); |
4087 | #endif |
4088 | } else { |
4089 | prefetch_row_bw = 0; |
4090 | MyError = true; |
4091 | #ifdef __DML_VBA_DEBUG__ |
4092 | dml_print("DML::%s: MyErr set. DestinationLinesToRequestRowInVBlank=%f (should be > 0)\n" , |
4093 | __func__, *DestinationLinesToRequestRowInVBlank); |
4094 | #endif |
4095 | } |
4096 | |
4097 | *prefetch_vmrow_bw = dml_max(a: prefetch_vm_bw, b: prefetch_row_bw); |
4098 | } |
4099 | |
4100 | if (MyError) { |
4101 | *PrefetchBandwidth = 0; |
4102 | TimeForFetchingMetaPTE = 0; |
4103 | TimeForFetchingRowInVBlank = 0; |
4104 | *DestinationLinesToRequestVMInVBlank = 0; |
4105 | *DestinationLinesToRequestRowInVBlank = 0; |
4106 | *DestinationLinesForPrefetch = 0; |
4107 | LinesToRequestPrefetchPixelData = 0; |
4108 | *VRatioPrefetchY = 0; |
4109 | *VRatioPrefetchC = 0; |
4110 | *RequiredPrefetchPixDataBWLuma = 0; |
4111 | *RequiredPrefetchPixDataBWChroma = 0; |
4112 | } |
4113 | |
4114 | return MyError; |
4115 | } // CalculatePrefetchSchedule |
4116 | |
4117 | void dml32_CalculateFlipSchedule( |
4118 | double HostVMInefficiencyFactor, |
4119 | double , |
4120 | double UrgentLatency, |
4121 | unsigned int GPUVMMaxPageTableLevels, |
4122 | bool HostVMEnable, |
4123 | unsigned int HostVMMaxNonCachedPageTableLevels, |
4124 | bool GPUVMEnable, |
4125 | double HostVMMinPageSize, |
4126 | double PDEAndMetaPTEBytesPerFrame, |
4127 | double MetaRowBytes, |
4128 | double DPTEBytesPerRow, |
4129 | double BandwidthAvailableForImmediateFlip, |
4130 | unsigned int TotImmediateFlipBytes, |
4131 | enum source_format_class SourcePixelFormat, |
4132 | double LineTime, |
4133 | double VRatio, |
4134 | double VRatioChroma, |
4135 | double Tno_bw, |
4136 | bool DCCEnable, |
4137 | unsigned int dpte_row_height, |
4138 | unsigned int meta_row_height, |
4139 | unsigned int dpte_row_height_chroma, |
4140 | unsigned int meta_row_height_chroma, |
4141 | bool use_one_row_for_frame_flip, |
4142 | |
4143 | /* Output */ |
4144 | double *DestinationLinesToRequestVMInImmediateFlip, |
4145 | double *DestinationLinesToRequestRowInImmediateFlip, |
4146 | double *final_flip_bw, |
4147 | bool *ImmediateFlipSupportedForPipe) |
4148 | { |
4149 | double min_row_time = 0.0; |
4150 | unsigned int HostVMDynamicLevelsTrips; |
4151 | double TimeForFetchingMetaPTEImmediateFlip; |
4152 | double TimeForFetchingRowInVBlankImmediateFlip; |
4153 | double ImmediateFlipBW = 1.0; |
4154 | |
4155 | if (GPUVMEnable == true && HostVMEnable == true) |
4156 | HostVMDynamicLevelsTrips = HostVMMaxNonCachedPageTableLevels; |
4157 | else |
4158 | HostVMDynamicLevelsTrips = 0; |
4159 | |
4160 | #ifdef __DML_VBA_DEBUG__ |
4161 | dml_print("DML::%s: TotImmediateFlipBytes = %d\n" , __func__, TotImmediateFlipBytes); |
4162 | dml_print("DML::%s: BandwidthAvailableForImmediateFlip = %f\n" , __func__, BandwidthAvailableForImmediateFlip); |
4163 | #endif |
4164 | |
4165 | if (TotImmediateFlipBytes > 0) { |
4166 | if (use_one_row_for_frame_flip) { |
4167 | ImmediateFlipBW = (PDEAndMetaPTEBytesPerFrame + MetaRowBytes + 2 * DPTEBytesPerRow) * |
4168 | BandwidthAvailableForImmediateFlip / TotImmediateFlipBytes; |
4169 | } else { |
4170 | ImmediateFlipBW = (PDEAndMetaPTEBytesPerFrame + MetaRowBytes + DPTEBytesPerRow) * |
4171 | BandwidthAvailableForImmediateFlip / TotImmediateFlipBytes; |
4172 | } |
4173 | if (GPUVMEnable == true) { |
4174 | TimeForFetchingMetaPTEImmediateFlip = dml_max3(a: Tno_bw + PDEAndMetaPTEBytesPerFrame * |
4175 | HostVMInefficiencyFactor / ImmediateFlipBW, |
4176 | b: UrgentExtraLatency + UrgentLatency * |
4177 | (GPUVMMaxPageTableLevels * (HostVMDynamicLevelsTrips + 1) - 1), |
4178 | c: LineTime / 4.0); |
4179 | } else { |
4180 | TimeForFetchingMetaPTEImmediateFlip = 0; |
4181 | } |
4182 | if ((GPUVMEnable == true || DCCEnable == true)) { |
4183 | TimeForFetchingRowInVBlankImmediateFlip = dml_max3( |
4184 | a: (MetaRowBytes + DPTEBytesPerRow * HostVMInefficiencyFactor) / ImmediateFlipBW, |
4185 | b: UrgentLatency * (HostVMDynamicLevelsTrips + 1), c: LineTime / 4.0); |
4186 | } else { |
4187 | TimeForFetchingRowInVBlankImmediateFlip = 0; |
4188 | } |
4189 | |
4190 | *DestinationLinesToRequestVMInImmediateFlip = |
4191 | dml_ceil(a: 4.0 * (TimeForFetchingMetaPTEImmediateFlip / LineTime), granularity: 1.0) / 4.0; |
4192 | *DestinationLinesToRequestRowInImmediateFlip = |
4193 | dml_ceil(a: 4.0 * (TimeForFetchingRowInVBlankImmediateFlip / LineTime), granularity: 1.0) / 4.0; |
4194 | |
4195 | if (GPUVMEnable == true) { |
4196 | *final_flip_bw = dml_max(a: PDEAndMetaPTEBytesPerFrame * HostVMInefficiencyFactor / |
4197 | (*DestinationLinesToRequestVMInImmediateFlip * LineTime), |
4198 | b: (MetaRowBytes + DPTEBytesPerRow * HostVMInefficiencyFactor) / |
4199 | (*DestinationLinesToRequestRowInImmediateFlip * LineTime)); |
4200 | } else if ((GPUVMEnable == true || DCCEnable == true)) { |
4201 | *final_flip_bw = (MetaRowBytes + DPTEBytesPerRow * HostVMInefficiencyFactor) / |
4202 | (*DestinationLinesToRequestRowInImmediateFlip * LineTime); |
4203 | } else { |
4204 | *final_flip_bw = 0; |
4205 | } |
4206 | } else { |
4207 | TimeForFetchingMetaPTEImmediateFlip = 0; |
4208 | TimeForFetchingRowInVBlankImmediateFlip = 0; |
4209 | *DestinationLinesToRequestVMInImmediateFlip = 0; |
4210 | *DestinationLinesToRequestRowInImmediateFlip = 0; |
4211 | *final_flip_bw = 0; |
4212 | } |
4213 | |
4214 | if (SourcePixelFormat == dm_420_8 || SourcePixelFormat == dm_420_10 || SourcePixelFormat == dm_rgbe_alpha) { |
4215 | if (GPUVMEnable == true && DCCEnable != true) { |
4216 | min_row_time = dml_min(a: dpte_row_height * |
4217 | LineTime / VRatio, b: dpte_row_height_chroma * LineTime / VRatioChroma); |
4218 | } else if (GPUVMEnable != true && DCCEnable == true) { |
4219 | min_row_time = dml_min(a: meta_row_height * |
4220 | LineTime / VRatio, b: meta_row_height_chroma * LineTime / VRatioChroma); |
4221 | } else { |
4222 | min_row_time = dml_min4(a: dpte_row_height * LineTime / VRatio, b: meta_row_height * |
4223 | LineTime / VRatio, c: dpte_row_height_chroma * LineTime / |
4224 | VRatioChroma, d: meta_row_height_chroma * LineTime / VRatioChroma); |
4225 | } |
4226 | } else { |
4227 | if (GPUVMEnable == true && DCCEnable != true) { |
4228 | min_row_time = dpte_row_height * LineTime / VRatio; |
4229 | } else if (GPUVMEnable != true && DCCEnable == true) { |
4230 | min_row_time = meta_row_height * LineTime / VRatio; |
4231 | } else { |
4232 | min_row_time = |
4233 | dml_min(a: dpte_row_height * LineTime / VRatio, b: meta_row_height * LineTime / VRatio); |
4234 | } |
4235 | } |
4236 | |
4237 | if (*DestinationLinesToRequestVMInImmediateFlip >= 32 || *DestinationLinesToRequestRowInImmediateFlip >= 16 |
4238 | || TimeForFetchingMetaPTEImmediateFlip + 2 * TimeForFetchingRowInVBlankImmediateFlip |
4239 | > min_row_time) { |
4240 | *ImmediateFlipSupportedForPipe = false; |
4241 | } else { |
4242 | *ImmediateFlipSupportedForPipe = true; |
4243 | } |
4244 | |
4245 | #ifdef __DML_VBA_DEBUG__ |
4246 | dml_print("DML::%s: GPUVMEnable = %d\n" , __func__, GPUVMEnable); |
4247 | dml_print("DML::%s: DCCEnable = %d\n" , __func__, DCCEnable); |
4248 | dml_print("DML::%s: DestinationLinesToRequestVMInImmediateFlip = %f\n" , |
4249 | __func__, *DestinationLinesToRequestVMInImmediateFlip); |
4250 | dml_print("DML::%s: DestinationLinesToRequestRowInImmediateFlip = %f\n" , |
4251 | __func__, *DestinationLinesToRequestRowInImmediateFlip); |
4252 | dml_print("DML::%s: TimeForFetchingMetaPTEImmediateFlip = %f\n" , __func__, TimeForFetchingMetaPTEImmediateFlip); |
4253 | dml_print("DML::%s: TimeForFetchingRowInVBlankImmediateFlip = %f\n" , |
4254 | __func__, TimeForFetchingRowInVBlankImmediateFlip); |
4255 | dml_print("DML::%s: min_row_time = %f\n" , __func__, min_row_time); |
4256 | dml_print("DML::%s: ImmediateFlipSupportedForPipe = %d\n" , __func__, *ImmediateFlipSupportedForPipe); |
4257 | #endif |
4258 | } // CalculateFlipSchedule |
4259 | |
4260 | void dml32_CalculateWatermarksMALLUseAndDRAMSpeedChangeSupport( |
4261 | struct vba_vars_st *v, |
4262 | unsigned int PrefetchMode, |
4263 | double DCFCLK, |
4264 | double ReturnBW, |
4265 | SOCParametersList mmSOCParameters, |
4266 | double SOCCLK, |
4267 | double DCFClkDeepSleep, |
4268 | unsigned int DETBufferSizeY[], |
4269 | unsigned int DETBufferSizeC[], |
4270 | unsigned int SwathHeightY[], |
4271 | unsigned int SwathHeightC[], |
4272 | double SwathWidthY[], |
4273 | double SwathWidthC[], |
4274 | unsigned int DPPPerSurface[], |
4275 | double BytePerPixelDETY[], |
4276 | double BytePerPixelDETC[], |
4277 | double DSTXAfterScaler[], |
4278 | double DSTYAfterScaler[], |
4279 | bool UnboundedRequestEnabled, |
4280 | unsigned int CompressedBufferSizeInkByte, |
4281 | |
4282 | /* Output */ |
4283 | enum clock_change_support *DRAMClockChangeSupport, |
4284 | double MaxActiveDRAMClockChangeLatencySupported[], |
4285 | unsigned int SubViewportLinesNeededInMALL[], |
4286 | enum dm_fclock_change_support *FCLKChangeSupport, |
4287 | double *MinActiveFCLKChangeLatencySupported, |
4288 | bool *USRRetrainingSupport, |
4289 | double ActiveDRAMClockChangeLatencyMargin[]) |
4290 | { |
4291 | unsigned int i, j, k; |
4292 | unsigned int SurfaceWithMinActiveFCLKChangeMargin = 0; |
4293 | unsigned int DRAMClockChangeSupportNumber = 0; |
4294 | unsigned int LastSurfaceWithoutMargin; |
4295 | unsigned int DRAMClockChangeMethod = 0; |
4296 | bool FoundFirstSurfaceWithMinActiveFCLKChangeMargin = false; |
4297 | double MinActiveFCLKChangeMargin = 0.; |
4298 | double SecondMinActiveFCLKChangeMarginOneDisplayInVBLank = 0.; |
4299 | double ActiveClockChangeLatencyHidingY; |
4300 | double ActiveClockChangeLatencyHidingC; |
4301 | double ActiveClockChangeLatencyHiding; |
4302 | double EffectiveDETBufferSizeY; |
4303 | double ActiveFCLKChangeLatencyMargin[DC__NUM_DPP__MAX]; |
4304 | double USRRetrainingLatencyMargin[DC__NUM_DPP__MAX]; |
4305 | double TotalPixelBW = 0.0; |
4306 | bool SynchronizedSurfaces[DC__NUM_DPP__MAX][DC__NUM_DPP__MAX]; |
4307 | double EffectiveLBLatencyHidingY; |
4308 | double EffectiveLBLatencyHidingC; |
4309 | double LinesInDETY[DC__NUM_DPP__MAX]; |
4310 | double LinesInDETC[DC__NUM_DPP__MAX]; |
4311 | unsigned int LinesInDETYRoundedDownToSwath[DC__NUM_DPP__MAX]; |
4312 | unsigned int LinesInDETCRoundedDownToSwath[DC__NUM_DPP__MAX]; |
4313 | double FullDETBufferingTimeY; |
4314 | double FullDETBufferingTimeC; |
4315 | double WritebackDRAMClockChangeLatencyMargin; |
4316 | double WritebackFCLKChangeLatencyMargin; |
4317 | double WritebackLatencyHiding; |
4318 | bool SameTimingForFCLKChange; |
4319 | |
4320 | unsigned int TotalActiveWriteback = 0; |
4321 | unsigned int LBLatencyHidingSourceLinesY[DC__NUM_DPP__MAX]; |
4322 | unsigned int LBLatencyHidingSourceLinesC[DC__NUM_DPP__MAX]; |
4323 | |
4324 | v->Watermark.UrgentWatermark = mmSOCParameters.UrgentLatency + mmSOCParameters.ExtraLatency; |
4325 | v->Watermark.USRRetrainingWatermark = mmSOCParameters.UrgentLatency + mmSOCParameters.ExtraLatency |
4326 | + mmSOCParameters.USRRetrainingLatency + mmSOCParameters.SMNLatency; |
4327 | v->Watermark.DRAMClockChangeWatermark = mmSOCParameters.DRAMClockChangeLatency + v->Watermark.UrgentWatermark; |
4328 | v->Watermark.FCLKChangeWatermark = mmSOCParameters.FCLKChangeLatency + v->Watermark.UrgentWatermark; |
4329 | v->Watermark.StutterExitWatermark = mmSOCParameters.SRExitTime + mmSOCParameters.ExtraLatency |
4330 | + 10 / DCFClkDeepSleep; |
4331 | v->Watermark.StutterEnterPlusExitWatermark = mmSOCParameters.SREnterPlusExitTime + mmSOCParameters.ExtraLatency |
4332 | + 10 / DCFClkDeepSleep; |
4333 | v->Watermark.Z8StutterExitWatermark = mmSOCParameters.SRExitZ8Time + mmSOCParameters.ExtraLatency |
4334 | + 10 / DCFClkDeepSleep; |
4335 | v->Watermark.Z8StutterEnterPlusExitWatermark = mmSOCParameters.SREnterPlusExitZ8Time |
4336 | + mmSOCParameters.ExtraLatency + 10 / DCFClkDeepSleep; |
4337 | |
4338 | #ifdef __DML_VBA_DEBUG__ |
4339 | dml_print("DML::%s: UrgentLatency = %f\n" , __func__, mmSOCParameters.UrgentLatency); |
4340 | dml_print("DML::%s: ExtraLatency = %f\n" , __func__, mmSOCParameters.ExtraLatency); |
4341 | dml_print("DML::%s: DRAMClockChangeLatency = %f\n" , __func__, mmSOCParameters.DRAMClockChangeLatency); |
4342 | dml_print("DML::%s: UrgentWatermark = %f\n" , __func__, v->Watermark.UrgentWatermark); |
4343 | dml_print("DML::%s: USRRetrainingWatermark = %f\n" , __func__, v->Watermark.USRRetrainingWatermark); |
4344 | dml_print("DML::%s: DRAMClockChangeWatermark = %f\n" , __func__, v->Watermark.DRAMClockChangeWatermark); |
4345 | dml_print("DML::%s: FCLKChangeWatermark = %f\n" , __func__, v->Watermark.FCLKChangeWatermark); |
4346 | dml_print("DML::%s: StutterExitWatermark = %f\n" , __func__, v->Watermark.StutterExitWatermark); |
4347 | dml_print("DML::%s: StutterEnterPlusExitWatermark = %f\n" , __func__, v->Watermark.StutterEnterPlusExitWatermark); |
4348 | dml_print("DML::%s: Z8StutterExitWatermark = %f\n" , __func__, v->Watermark.Z8StutterExitWatermark); |
4349 | dml_print("DML::%s: Z8StutterEnterPlusExitWatermark = %f\n" , |
4350 | __func__, v->Watermark.Z8StutterEnterPlusExitWatermark); |
4351 | #endif |
4352 | |
4353 | |
4354 | TotalActiveWriteback = 0; |
4355 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4356 | if (v->WritebackEnable[k] == true) |
4357 | TotalActiveWriteback = TotalActiveWriteback + 1; |
4358 | } |
4359 | |
4360 | if (TotalActiveWriteback <= 1) { |
4361 | v->Watermark.WritebackUrgentWatermark = mmSOCParameters.WritebackLatency; |
4362 | } else { |
4363 | v->Watermark.WritebackUrgentWatermark = mmSOCParameters.WritebackLatency |
4364 | + v->WritebackChunkSize * 1024.0 / 32.0 / SOCCLK; |
4365 | } |
4366 | if (v->USRRetrainingRequiredFinal) |
4367 | v->Watermark.WritebackDRAMClockChangeWatermark = v->Watermark.WritebackDRAMClockChangeWatermark |
4368 | + mmSOCParameters.USRRetrainingLatency; |
4369 | |
4370 | if (TotalActiveWriteback <= 1) { |
4371 | v->Watermark.WritebackDRAMClockChangeWatermark = mmSOCParameters.DRAMClockChangeLatency |
4372 | + mmSOCParameters.WritebackLatency; |
4373 | v->Watermark.WritebackFCLKChangeWatermark = mmSOCParameters.FCLKChangeLatency |
4374 | + mmSOCParameters.WritebackLatency; |
4375 | } else { |
4376 | v->Watermark.WritebackDRAMClockChangeWatermark = mmSOCParameters.DRAMClockChangeLatency |
4377 | + mmSOCParameters.WritebackLatency + v->WritebackChunkSize * 1024.0 / 32.0 / SOCCLK; |
4378 | v->Watermark.WritebackFCLKChangeWatermark = mmSOCParameters.FCLKChangeLatency |
4379 | + mmSOCParameters.WritebackLatency + v->WritebackChunkSize * 1024 / 32 / SOCCLK; |
4380 | } |
4381 | |
4382 | if (v->USRRetrainingRequiredFinal) |
4383 | v->Watermark.WritebackDRAMClockChangeWatermark = v->Watermark.WritebackDRAMClockChangeWatermark |
4384 | + mmSOCParameters.USRRetrainingLatency; |
4385 | |
4386 | if (v->USRRetrainingRequiredFinal) |
4387 | v->Watermark.WritebackFCLKChangeWatermark = v->Watermark.WritebackFCLKChangeWatermark |
4388 | + mmSOCParameters.USRRetrainingLatency; |
4389 | |
4390 | #ifdef __DML_VBA_DEBUG__ |
4391 | dml_print("DML::%s: WritebackDRAMClockChangeWatermark = %f\n" , |
4392 | __func__, v->Watermark.WritebackDRAMClockChangeWatermark); |
4393 | dml_print("DML::%s: WritebackFCLKChangeWatermark = %f\n" , __func__, v->Watermark.WritebackFCLKChangeWatermark); |
4394 | dml_print("DML::%s: WritebackUrgentWatermark = %f\n" , __func__, v->Watermark.WritebackUrgentWatermark); |
4395 | dml_print("DML::%s: v->USRRetrainingRequiredFinal = %d\n" , __func__, v->USRRetrainingRequiredFinal); |
4396 | dml_print("DML::%s: USRRetrainingLatency = %f\n" , __func__, mmSOCParameters.USRRetrainingLatency); |
4397 | #endif |
4398 | |
4399 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4400 | TotalPixelBW = TotalPixelBW + DPPPerSurface[k] * (SwathWidthY[k] * BytePerPixelDETY[k] * v->VRatio[k] + |
4401 | SwathWidthC[k] * BytePerPixelDETC[k] * v->VRatioChroma[k]) / (v->HTotal[k] / v->PixelClock[k]); |
4402 | } |
4403 | |
4404 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4405 | |
4406 | LBLatencyHidingSourceLinesY[k] = dml_min(a: (double) v->MaxLineBufferLines, b: dml_floor(a: v->LineBufferSizeFinal / v->LBBitPerPixel[k] / (SwathWidthY[k] / dml_max(a: v->HRatio[k], b: 1.0)), granularity: 1)) - (v->vtaps[k] - 1); |
4407 | LBLatencyHidingSourceLinesC[k] = dml_min(a: (double) v->MaxLineBufferLines, b: dml_floor(a: v->LineBufferSizeFinal / v->LBBitPerPixel[k] / (SwathWidthC[k] / dml_max(a: v->HRatioChroma[k], b: 1.0)), granularity: 1)) - (v->VTAPsChroma[k] - 1); |
4408 | |
4409 | |
4410 | #ifdef __DML_VBA_DEBUG__ |
4411 | dml_print("DML::%s: k=%d, v->MaxLineBufferLines = %d\n" , __func__, k, v->MaxLineBufferLines); |
4412 | dml_print("DML::%s: k=%d, v->LineBufferSizeFinal = %d\n" , __func__, k, v->LineBufferSizeFinal); |
4413 | dml_print("DML::%s: k=%d, v->LBBitPerPixel = %d\n" , __func__, k, v->LBBitPerPixel[k]); |
4414 | dml_print("DML::%s: k=%d, v->HRatio = %f\n" , __func__, k, v->HRatio[k]); |
4415 | dml_print("DML::%s: k=%d, v->vtaps = %d\n" , __func__, k, v->vtaps[k]); |
4416 | #endif |
4417 | |
4418 | EffectiveLBLatencyHidingY = LBLatencyHidingSourceLinesY[k] / v->VRatio[k] * (v->HTotal[k] / v->PixelClock[k]); |
4419 | EffectiveLBLatencyHidingC = LBLatencyHidingSourceLinesC[k] / v->VRatioChroma[k] * (v->HTotal[k] / v->PixelClock[k]); |
4420 | EffectiveDETBufferSizeY = DETBufferSizeY[k]; |
4421 | |
4422 | if (UnboundedRequestEnabled) { |
4423 | EffectiveDETBufferSizeY = EffectiveDETBufferSizeY |
4424 | + CompressedBufferSizeInkByte * 1024 |
4425 | * (SwathWidthY[k] * BytePerPixelDETY[k] * v->VRatio[k]) |
4426 | / (v->HTotal[k] / v->PixelClock[k]) / TotalPixelBW; |
4427 | } |
4428 | |
4429 | LinesInDETY[k] = (double) EffectiveDETBufferSizeY / BytePerPixelDETY[k] / SwathWidthY[k]; |
4430 | LinesInDETYRoundedDownToSwath[k] = dml_floor(a: LinesInDETY[k], granularity: SwathHeightY[k]); |
4431 | FullDETBufferingTimeY = LinesInDETYRoundedDownToSwath[k] * (v->HTotal[k] / v->PixelClock[k]) / v->VRatio[k]; |
4432 | |
4433 | ActiveClockChangeLatencyHidingY = EffectiveLBLatencyHidingY + FullDETBufferingTimeY |
4434 | - (DSTXAfterScaler[k] / v->HTotal[k] + DSTYAfterScaler[k]) * v->HTotal[k] / v->PixelClock[k]; |
4435 | |
4436 | if (v->NumberOfActiveSurfaces > 1) { |
4437 | ActiveClockChangeLatencyHidingY = ActiveClockChangeLatencyHidingY |
4438 | - (1.0 - 1.0 / v->NumberOfActiveSurfaces) * SwathHeightY[k] * v->HTotal[k] |
4439 | / v->PixelClock[k] / v->VRatio[k]; |
4440 | } |
4441 | |
4442 | if (BytePerPixelDETC[k] > 0) { |
4443 | LinesInDETC[k] = DETBufferSizeC[k] / BytePerPixelDETC[k] / SwathWidthC[k]; |
4444 | LinesInDETCRoundedDownToSwath[k] = dml_floor(a: LinesInDETC[k], granularity: SwathHeightC[k]); |
4445 | FullDETBufferingTimeC = LinesInDETCRoundedDownToSwath[k] * (v->HTotal[k] / v->PixelClock[k]) |
4446 | / v->VRatioChroma[k]; |
4447 | ActiveClockChangeLatencyHidingC = EffectiveLBLatencyHidingC + FullDETBufferingTimeC |
4448 | - (DSTXAfterScaler[k] / v->HTotal[k] + DSTYAfterScaler[k]) * v->HTotal[k] |
4449 | / v->PixelClock[k]; |
4450 | if (v->NumberOfActiveSurfaces > 1) { |
4451 | ActiveClockChangeLatencyHidingC = ActiveClockChangeLatencyHidingC |
4452 | - (1 - 1 / v->NumberOfActiveSurfaces) * SwathHeightC[k] * v->HTotal[k] |
4453 | / v->PixelClock[k] / v->VRatioChroma[k]; |
4454 | } |
4455 | ActiveClockChangeLatencyHiding = dml_min(a: ActiveClockChangeLatencyHidingY, |
4456 | b: ActiveClockChangeLatencyHidingC); |
4457 | } else { |
4458 | ActiveClockChangeLatencyHiding = ActiveClockChangeLatencyHidingY; |
4459 | } |
4460 | |
4461 | ActiveDRAMClockChangeLatencyMargin[k] = ActiveClockChangeLatencyHiding - v->Watermark.UrgentWatermark |
4462 | - v->Watermark.DRAMClockChangeWatermark; |
4463 | ActiveFCLKChangeLatencyMargin[k] = ActiveClockChangeLatencyHiding - v->Watermark.UrgentWatermark |
4464 | - v->Watermark.FCLKChangeWatermark; |
4465 | USRRetrainingLatencyMargin[k] = ActiveClockChangeLatencyHiding - v->Watermark.USRRetrainingWatermark; |
4466 | |
4467 | if (v->WritebackEnable[k]) { |
4468 | WritebackLatencyHiding = v->WritebackInterfaceBufferSize * 1024 |
4469 | / (v->WritebackDestinationWidth[k] * v->WritebackDestinationHeight[k] |
4470 | / (v->WritebackSourceHeight[k] * v->HTotal[k] / v->PixelClock[k]) * 4); |
4471 | if (v->WritebackPixelFormat[k] == dm_444_64) |
4472 | WritebackLatencyHiding = WritebackLatencyHiding / 2; |
4473 | |
4474 | WritebackDRAMClockChangeLatencyMargin = WritebackLatencyHiding |
4475 | - v->Watermark.WritebackDRAMClockChangeWatermark; |
4476 | |
4477 | WritebackFCLKChangeLatencyMargin = WritebackLatencyHiding |
4478 | - v->Watermark.WritebackFCLKChangeWatermark; |
4479 | |
4480 | ActiveDRAMClockChangeLatencyMargin[k] = dml_min(a: ActiveDRAMClockChangeLatencyMargin[k], |
4481 | b: WritebackFCLKChangeLatencyMargin); |
4482 | ActiveFCLKChangeLatencyMargin[k] = dml_min(a: ActiveFCLKChangeLatencyMargin[k], |
4483 | b: WritebackDRAMClockChangeLatencyMargin); |
4484 | } |
4485 | MaxActiveDRAMClockChangeLatencySupported[k] = |
4486 | (v->UsesMALLForPStateChange[k] == dm_use_mall_pstate_change_phantom_pipe) ? |
4487 | 0 : |
4488 | (ActiveDRAMClockChangeLatencyMargin[k] |
4489 | + mmSOCParameters.DRAMClockChangeLatency); |
4490 | } |
4491 | |
4492 | for (i = 0; i < v->NumberOfActiveSurfaces; ++i) { |
4493 | for (j = 0; j < v->NumberOfActiveSurfaces; ++j) { |
4494 | if (i == j || |
4495 | (v->BlendingAndTiming[i] == i && v->BlendingAndTiming[j] == i) || |
4496 | (v->BlendingAndTiming[j] == j && v->BlendingAndTiming[i] == j) || |
4497 | (v->BlendingAndTiming[i] == v->BlendingAndTiming[j] && v->BlendingAndTiming[i] != i) || |
4498 | (v->SynchronizeTimingsFinal && v->PixelClock[i] == v->PixelClock[j] && |
4499 | v->HTotal[i] == v->HTotal[j] && v->VTotal[i] == v->VTotal[j] && |
4500 | v->VActive[i] == v->VActive[j]) || (v->SynchronizeDRRDisplaysForUCLKPStateChangeFinal && |
4501 | (v->DRRDisplay[i] || v->DRRDisplay[j]))) { |
4502 | SynchronizedSurfaces[i][j] = true; |
4503 | } else { |
4504 | SynchronizedSurfaces[i][j] = false; |
4505 | } |
4506 | } |
4507 | } |
4508 | |
4509 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4510 | if ((v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) && |
4511 | (!FoundFirstSurfaceWithMinActiveFCLKChangeMargin || |
4512 | ActiveFCLKChangeLatencyMargin[k] < MinActiveFCLKChangeMargin)) { |
4513 | FoundFirstSurfaceWithMinActiveFCLKChangeMargin = true; |
4514 | MinActiveFCLKChangeMargin = ActiveFCLKChangeLatencyMargin[k]; |
4515 | SurfaceWithMinActiveFCLKChangeMargin = k; |
4516 | } |
4517 | } |
4518 | |
4519 | *MinActiveFCLKChangeLatencySupported = MinActiveFCLKChangeMargin + mmSOCParameters.FCLKChangeLatency; |
4520 | |
4521 | SameTimingForFCLKChange = true; |
4522 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4523 | if (!SynchronizedSurfaces[k][SurfaceWithMinActiveFCLKChangeMargin]) { |
4524 | if ((v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) && |
4525 | (SameTimingForFCLKChange || |
4526 | ActiveFCLKChangeLatencyMargin[k] < |
4527 | SecondMinActiveFCLKChangeMarginOneDisplayInVBLank)) { |
4528 | SecondMinActiveFCLKChangeMarginOneDisplayInVBLank = ActiveFCLKChangeLatencyMargin[k]; |
4529 | } |
4530 | SameTimingForFCLKChange = false; |
4531 | } |
4532 | } |
4533 | |
4534 | if (MinActiveFCLKChangeMargin > 0) { |
4535 | *FCLKChangeSupport = dm_fclock_change_vactive; |
4536 | } else if ((SameTimingForFCLKChange || SecondMinActiveFCLKChangeMarginOneDisplayInVBLank > 0) && |
4537 | (PrefetchMode <= 1)) { |
4538 | *FCLKChangeSupport = dm_fclock_change_vblank; |
4539 | } else { |
4540 | *FCLKChangeSupport = dm_fclock_change_unsupported; |
4541 | } |
4542 | |
4543 | *USRRetrainingSupport = true; |
4544 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4545 | if ((v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) && |
4546 | (USRRetrainingLatencyMargin[k] < 0)) { |
4547 | *USRRetrainingSupport = false; |
4548 | } |
4549 | } |
4550 | |
4551 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4552 | if (v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_full_frame && |
4553 | v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_sub_viewport && |
4554 | v->UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe && |
4555 | ActiveDRAMClockChangeLatencyMargin[k] < 0) { |
4556 | if (PrefetchMode > 0) { |
4557 | DRAMClockChangeSupportNumber = 2; |
4558 | } else if (DRAMClockChangeSupportNumber == 0) { |
4559 | DRAMClockChangeSupportNumber = 1; |
4560 | LastSurfaceWithoutMargin = k; |
4561 | } else if (DRAMClockChangeSupportNumber == 1 && |
4562 | !SynchronizedSurfaces[LastSurfaceWithoutMargin][k]) { |
4563 | DRAMClockChangeSupportNumber = 2; |
4564 | } |
4565 | } |
4566 | } |
4567 | |
4568 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4569 | if (v->UsesMALLForPStateChange[k] == dm_use_mall_pstate_change_full_frame) |
4570 | DRAMClockChangeMethod = 1; |
4571 | else if (v->UsesMALLForPStateChange[k] == dm_use_mall_pstate_change_sub_viewport) |
4572 | DRAMClockChangeMethod = 2; |
4573 | } |
4574 | |
4575 | if (DRAMClockChangeMethod == 0) { |
4576 | if (DRAMClockChangeSupportNumber == 0) |
4577 | *DRAMClockChangeSupport = dm_dram_clock_change_vactive; |
4578 | else if (DRAMClockChangeSupportNumber == 1) |
4579 | *DRAMClockChangeSupport = dm_dram_clock_change_vblank; |
4580 | else |
4581 | *DRAMClockChangeSupport = dm_dram_clock_change_unsupported; |
4582 | } else if (DRAMClockChangeMethod == 1) { |
4583 | if (DRAMClockChangeSupportNumber == 0) |
4584 | *DRAMClockChangeSupport = dm_dram_clock_change_vactive_w_mall_full_frame; |
4585 | else if (DRAMClockChangeSupportNumber == 1) |
4586 | *DRAMClockChangeSupport = dm_dram_clock_change_vblank_w_mall_full_frame; |
4587 | else |
4588 | *DRAMClockChangeSupport = dm_dram_clock_change_unsupported; |
4589 | } else { |
4590 | if (DRAMClockChangeSupportNumber == 0) |
4591 | *DRAMClockChangeSupport = dm_dram_clock_change_vactive_w_mall_sub_vp; |
4592 | else if (DRAMClockChangeSupportNumber == 1) |
4593 | *DRAMClockChangeSupport = dm_dram_clock_change_vblank_w_mall_sub_vp; |
4594 | else |
4595 | *DRAMClockChangeSupport = dm_dram_clock_change_unsupported; |
4596 | } |
4597 | |
4598 | for (k = 0; k < v->NumberOfActiveSurfaces; ++k) { |
4599 | unsigned int dst_y_pstate; |
4600 | unsigned int src_y_pstate_l; |
4601 | unsigned int src_y_pstate_c; |
4602 | unsigned int src_y_ahead_l, src_y_ahead_c, sub_vp_lines_l, sub_vp_lines_c; |
4603 | |
4604 | dst_y_pstate = dml_ceil(a: (mmSOCParameters.DRAMClockChangeLatency + mmSOCParameters.UrgentLatency) / (v->HTotal[k] / v->PixelClock[k]), granularity: 1); |
4605 | src_y_pstate_l = dml_ceil(a: dst_y_pstate * v->VRatio[k], granularity: SwathHeightY[k]); |
4606 | src_y_ahead_l = dml_floor(a: DETBufferSizeY[k] / BytePerPixelDETY[k] / SwathWidthY[k], granularity: SwathHeightY[k]) + LBLatencyHidingSourceLinesY[k]; |
4607 | sub_vp_lines_l = src_y_pstate_l + src_y_ahead_l + v->meta_row_height[k]; |
4608 | |
4609 | #ifdef __DML_VBA_DEBUG__ |
4610 | dml_print("DML::%s: k=%d, DETBufferSizeY = %d\n" , __func__, k, DETBufferSizeY[k]); |
4611 | dml_print("DML::%s: k=%d, BytePerPixelDETY = %f\n" , __func__, k, BytePerPixelDETY[k]); |
4612 | dml_print("DML::%s: k=%d, SwathWidthY = %d\n" , __func__, k, SwathWidthY[k]); |
4613 | dml_print("DML::%s: k=%d, SwathHeightY = %d\n" , __func__, k, SwathHeightY[k]); |
4614 | dml_print("DML::%s: k=%d, LBLatencyHidingSourceLinesY = %d\n" , __func__, k, LBLatencyHidingSourceLinesY[k]); |
4615 | dml_print("DML::%s: k=%d, dst_y_pstate = %d\n" , __func__, k, dst_y_pstate); |
4616 | dml_print("DML::%s: k=%d, src_y_pstate_l = %d\n" , __func__, k, src_y_pstate_l); |
4617 | dml_print("DML::%s: k=%d, src_y_ahead_l = %d\n" , __func__, k, src_y_ahead_l); |
4618 | dml_print("DML::%s: k=%d, v->meta_row_height = %d\n" , __func__, k, v->meta_row_height[k]); |
4619 | dml_print("DML::%s: k=%d, sub_vp_lines_l = %d\n" , __func__, k, sub_vp_lines_l); |
4620 | #endif |
4621 | SubViewportLinesNeededInMALL[k] = sub_vp_lines_l; |
4622 | |
4623 | if (BytePerPixelDETC[k] > 0) { |
4624 | src_y_pstate_c = dml_ceil(a: dst_y_pstate * v->VRatioChroma[k], granularity: SwathHeightC[k]); |
4625 | src_y_ahead_c = dml_floor(a: DETBufferSizeC[k] / BytePerPixelDETC[k] / SwathWidthC[k], granularity: SwathHeightC[k]) + LBLatencyHidingSourceLinesC[k]; |
4626 | sub_vp_lines_c = src_y_pstate_c + src_y_ahead_c + v->meta_row_height_chroma[k]; |
4627 | SubViewportLinesNeededInMALL[k] = dml_max(a: sub_vp_lines_l, b: sub_vp_lines_c); |
4628 | |
4629 | #ifdef __DML_VBA_DEBUG__ |
4630 | dml_print("DML::%s: k=%d, src_y_pstate_c = %d\n" , __func__, k, src_y_pstate_c); |
4631 | dml_print("DML::%s: k=%d, src_y_ahead_c = %d\n" , __func__, k, src_y_ahead_c); |
4632 | dml_print("DML::%s: k=%d, v->meta_row_height_chroma = %d\n" , __func__, k, v->meta_row_height_chroma[k]); |
4633 | dml_print("DML::%s: k=%d, sub_vp_lines_c = %d\n" , __func__, k, sub_vp_lines_c); |
4634 | #endif |
4635 | } |
4636 | } |
4637 | #ifdef __DML_VBA_DEBUG__ |
4638 | dml_print("DML::%s: DRAMClockChangeSupport = %d\n" , __func__, *DRAMClockChangeSupport); |
4639 | dml_print("DML::%s: FCLKChangeSupport = %d\n" , __func__, *FCLKChangeSupport); |
4640 | dml_print("DML::%s: MinActiveFCLKChangeLatencySupported = %f\n" , |
4641 | __func__, *MinActiveFCLKChangeLatencySupported); |
4642 | dml_print("DML::%s: USRRetrainingSupport = %d\n" , __func__, *USRRetrainingSupport); |
4643 | #endif |
4644 | } // CalculateWatermarksMALLUseAndDRAMSpeedChangeSupport |
4645 | |
4646 | double dml32_CalculateWriteBackDISPCLK( |
4647 | enum source_format_class WritebackPixelFormat, |
4648 | double PixelClock, |
4649 | double WritebackHRatio, |
4650 | double WritebackVRatio, |
4651 | unsigned int WritebackHTaps, |
4652 | unsigned int WritebackVTaps, |
4653 | unsigned int WritebackSourceWidth, |
4654 | unsigned int WritebackDestinationWidth, |
4655 | unsigned int HTotal, |
4656 | unsigned int WritebackLineBufferSize, |
4657 | double DISPCLKDPPCLKVCOSpeed) |
4658 | { |
4659 | double DISPCLK_H, DISPCLK_V, DISPCLK_HB; |
4660 | |
4661 | DISPCLK_H = PixelClock * dml_ceil(a: WritebackHTaps / 8.0, granularity: 1) / WritebackHRatio; |
4662 | DISPCLK_V = PixelClock * (WritebackVTaps * dml_ceil(a: WritebackDestinationWidth / 6.0, granularity: 1) + 8.0) / HTotal; |
4663 | DISPCLK_HB = PixelClock * WritebackVTaps * (WritebackDestinationWidth * |
4664 | WritebackVTaps - WritebackLineBufferSize / 57.0) / 6.0 / WritebackSourceWidth; |
4665 | return dml32_RoundToDFSGranularity(Clock: dml_max3(a: DISPCLK_H, b: DISPCLK_V, c: DISPCLK_HB), round_up: 1, VCOSpeed: DISPCLKDPPCLKVCOSpeed); |
4666 | } |
4667 | |
4668 | void dml32_CalculateMinAndMaxPrefetchMode( |
4669 | enum dm_prefetch_modes AllowForPStateChangeOrStutterInVBlankFinal, |
4670 | unsigned int *MinPrefetchMode, |
4671 | unsigned int *MaxPrefetchMode) |
4672 | { |
4673 | if (AllowForPStateChangeOrStutterInVBlankFinal == dm_prefetch_support_none) { |
4674 | *MinPrefetchMode = 3; |
4675 | *MaxPrefetchMode = 3; |
4676 | } else if (AllowForPStateChangeOrStutterInVBlankFinal == dm_prefetch_support_stutter) { |
4677 | *MinPrefetchMode = 2; |
4678 | *MaxPrefetchMode = 2; |
4679 | } else if (AllowForPStateChangeOrStutterInVBlankFinal == dm_prefetch_support_fclk_and_stutter) { |
4680 | *MinPrefetchMode = 1; |
4681 | *MaxPrefetchMode = 1; |
4682 | } else if (AllowForPStateChangeOrStutterInVBlankFinal == dm_prefetch_support_uclk_fclk_and_stutter) { |
4683 | *MinPrefetchMode = 0; |
4684 | *MaxPrefetchMode = 0; |
4685 | } else { |
4686 | *MinPrefetchMode = 0; |
4687 | *MaxPrefetchMode = 3; |
4688 | } |
4689 | } // CalculateMinAndMaxPrefetchMode |
4690 | |
4691 | void dml32_CalculatePixelDeliveryTimes( |
4692 | unsigned int NumberOfActiveSurfaces, |
4693 | double VRatio[], |
4694 | double VRatioChroma[], |
4695 | double VRatioPrefetchY[], |
4696 | double VRatioPrefetchC[], |
4697 | unsigned int swath_width_luma_ub[], |
4698 | unsigned int swath_width_chroma_ub[], |
4699 | unsigned int DPPPerSurface[], |
4700 | double HRatio[], |
4701 | double HRatioChroma[], |
4702 | double PixelClock[], |
4703 | double PSCL_THROUGHPUT[], |
4704 | double PSCL_THROUGHPUT_CHROMA[], |
4705 | double Dppclk[], |
4706 | unsigned int BytePerPixelC[], |
4707 | enum dm_rotation_angle SourceRotation[], |
4708 | unsigned int NumberOfCursors[], |
4709 | unsigned int CursorWidth[][DC__NUM_CURSOR__MAX], |
4710 | unsigned int CursorBPP[][DC__NUM_CURSOR__MAX], |
4711 | unsigned int BlockWidth256BytesY[], |
4712 | unsigned int BlockHeight256BytesY[], |
4713 | unsigned int BlockWidth256BytesC[], |
4714 | unsigned int BlockHeight256BytesC[], |
4715 | |
4716 | /* Output */ |
4717 | double DisplayPipeLineDeliveryTimeLuma[], |
4718 | double DisplayPipeLineDeliveryTimeChroma[], |
4719 | double DisplayPipeLineDeliveryTimeLumaPrefetch[], |
4720 | double DisplayPipeLineDeliveryTimeChromaPrefetch[], |
4721 | double DisplayPipeRequestDeliveryTimeLuma[], |
4722 | double DisplayPipeRequestDeliveryTimeChroma[], |
4723 | double DisplayPipeRequestDeliveryTimeLumaPrefetch[], |
4724 | double DisplayPipeRequestDeliveryTimeChromaPrefetch[], |
4725 | double CursorRequestDeliveryTime[], |
4726 | double CursorRequestDeliveryTimePrefetch[]) |
4727 | { |
4728 | double req_per_swath_ub; |
4729 | unsigned int k; |
4730 | |
4731 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
4732 | |
4733 | #ifdef __DML_VBA_DEBUG__ |
4734 | dml_print("DML::%s: k=%d : HRatio = %f\n" , __func__, k, HRatio[k]); |
4735 | dml_print("DML::%s: k=%d : VRatio = %f\n" , __func__, k, VRatio[k]); |
4736 | dml_print("DML::%s: k=%d : HRatioChroma = %f\n" , __func__, k, HRatioChroma[k]); |
4737 | dml_print("DML::%s: k=%d : VRatioChroma = %f\n" , __func__, k, VRatioChroma[k]); |
4738 | dml_print("DML::%s: k=%d : swath_width_luma_ub = %d\n" , __func__, k, swath_width_luma_ub[k]); |
4739 | dml_print("DML::%s: k=%d : swath_width_chroma_ub = %d\n" , __func__, k, swath_width_chroma_ub[k]); |
4740 | dml_print("DML::%s: k=%d : PSCL_THROUGHPUT = %f\n" , __func__, k, PSCL_THROUGHPUT[k]); |
4741 | dml_print("DML::%s: k=%d : PSCL_THROUGHPUT_CHROMA = %f\n" , __func__, k, PSCL_THROUGHPUT_CHROMA[k]); |
4742 | dml_print("DML::%s: k=%d : DPPPerSurface = %d\n" , __func__, k, DPPPerSurface[k]); |
4743 | dml_print("DML::%s: k=%d : PixelClock = %f\n" , __func__, k, PixelClock[k]); |
4744 | dml_print("DML::%s: k=%d : Dppclk = %f\n" , __func__, k, Dppclk[k]); |
4745 | #endif |
4746 | |
4747 | if (VRatio[k] <= 1) { |
4748 | DisplayPipeLineDeliveryTimeLuma[k] = |
4749 | swath_width_luma_ub[k] * DPPPerSurface[k] / HRatio[k] / PixelClock[k]; |
4750 | } else { |
4751 | DisplayPipeLineDeliveryTimeLuma[k] = swath_width_luma_ub[k] / PSCL_THROUGHPUT[k] / Dppclk[k]; |
4752 | } |
4753 | |
4754 | if (BytePerPixelC[k] == 0) { |
4755 | DisplayPipeLineDeliveryTimeChroma[k] = 0; |
4756 | } else { |
4757 | if (VRatioChroma[k] <= 1) { |
4758 | DisplayPipeLineDeliveryTimeChroma[k] = |
4759 | swath_width_chroma_ub[k] * DPPPerSurface[k] / HRatioChroma[k] / PixelClock[k]; |
4760 | } else { |
4761 | DisplayPipeLineDeliveryTimeChroma[k] = |
4762 | swath_width_chroma_ub[k] / PSCL_THROUGHPUT_CHROMA[k] / Dppclk[k]; |
4763 | } |
4764 | } |
4765 | |
4766 | if (VRatioPrefetchY[k] <= 1) { |
4767 | DisplayPipeLineDeliveryTimeLumaPrefetch[k] = |
4768 | swath_width_luma_ub[k] * DPPPerSurface[k] / HRatio[k] / PixelClock[k]; |
4769 | } else { |
4770 | DisplayPipeLineDeliveryTimeLumaPrefetch[k] = |
4771 | swath_width_luma_ub[k] / PSCL_THROUGHPUT[k] / Dppclk[k]; |
4772 | } |
4773 | |
4774 | if (BytePerPixelC[k] == 0) { |
4775 | DisplayPipeLineDeliveryTimeChromaPrefetch[k] = 0; |
4776 | } else { |
4777 | if (VRatioPrefetchC[k] <= 1) { |
4778 | DisplayPipeLineDeliveryTimeChromaPrefetch[k] = swath_width_chroma_ub[k] * |
4779 | DPPPerSurface[k] / HRatioChroma[k] / PixelClock[k]; |
4780 | } else { |
4781 | DisplayPipeLineDeliveryTimeChromaPrefetch[k] = |
4782 | swath_width_chroma_ub[k] / PSCL_THROUGHPUT_CHROMA[k] / Dppclk[k]; |
4783 | } |
4784 | } |
4785 | #ifdef __DML_VBA_DEBUG__ |
4786 | dml_print("DML::%s: k=%d : DisplayPipeLineDeliveryTimeLuma = %f\n" , |
4787 | __func__, k, DisplayPipeLineDeliveryTimeLuma[k]); |
4788 | dml_print("DML::%s: k=%d : DisplayPipeLineDeliveryTimeLumaPrefetch = %f\n" , |
4789 | __func__, k, DisplayPipeLineDeliveryTimeLumaPrefetch[k]); |
4790 | dml_print("DML::%s: k=%d : DisplayPipeLineDeliveryTimeChroma = %f\n" , |
4791 | __func__, k, DisplayPipeLineDeliveryTimeChroma[k]); |
4792 | dml_print("DML::%s: k=%d : DisplayPipeLineDeliveryTimeChromaPrefetch = %f\n" , |
4793 | __func__, k, DisplayPipeLineDeliveryTimeChromaPrefetch[k]); |
4794 | #endif |
4795 | } |
4796 | |
4797 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
4798 | if (!IsVertical(Scan: SourceRotation[k])) |
4799 | req_per_swath_ub = swath_width_luma_ub[k] / BlockWidth256BytesY[k]; |
4800 | else |
4801 | req_per_swath_ub = swath_width_luma_ub[k] / BlockHeight256BytesY[k]; |
4802 | #ifdef __DML_VBA_DEBUG__ |
4803 | dml_print("DML::%s: k=%d : req_per_swath_ub = %f (Luma)\n" , __func__, k, req_per_swath_ub); |
4804 | #endif |
4805 | |
4806 | DisplayPipeRequestDeliveryTimeLuma[k] = DisplayPipeLineDeliveryTimeLuma[k] / req_per_swath_ub; |
4807 | DisplayPipeRequestDeliveryTimeLumaPrefetch[k] = |
4808 | DisplayPipeLineDeliveryTimeLumaPrefetch[k] / req_per_swath_ub; |
4809 | if (BytePerPixelC[k] == 0) { |
4810 | DisplayPipeRequestDeliveryTimeChroma[k] = 0; |
4811 | DisplayPipeRequestDeliveryTimeChromaPrefetch[k] = 0; |
4812 | } else { |
4813 | if (!IsVertical(Scan: SourceRotation[k])) |
4814 | req_per_swath_ub = swath_width_chroma_ub[k] / BlockWidth256BytesC[k]; |
4815 | else |
4816 | req_per_swath_ub = swath_width_chroma_ub[k] / BlockHeight256BytesC[k]; |
4817 | #ifdef __DML_VBA_DEBUG__ |
4818 | dml_print("DML::%s: k=%d : req_per_swath_ub = %f (Chroma)\n" , __func__, k, req_per_swath_ub); |
4819 | #endif |
4820 | DisplayPipeRequestDeliveryTimeChroma[k] = |
4821 | DisplayPipeLineDeliveryTimeChroma[k] / req_per_swath_ub; |
4822 | DisplayPipeRequestDeliveryTimeChromaPrefetch[k] = |
4823 | DisplayPipeLineDeliveryTimeChromaPrefetch[k] / req_per_swath_ub; |
4824 | } |
4825 | #ifdef __DML_VBA_DEBUG__ |
4826 | dml_print("DML::%s: k=%d : DisplayPipeRequestDeliveryTimeLuma = %f\n" , |
4827 | __func__, k, DisplayPipeRequestDeliveryTimeLuma[k]); |
4828 | dml_print("DML::%s: k=%d : DisplayPipeRequestDeliveryTimeLumaPrefetch = %f\n" , |
4829 | __func__, k, DisplayPipeRequestDeliveryTimeLumaPrefetch[k]); |
4830 | dml_print("DML::%s: k=%d : DisplayPipeRequestDeliveryTimeChroma = %f\n" , |
4831 | __func__, k, DisplayPipeRequestDeliveryTimeChroma[k]); |
4832 | dml_print("DML::%s: k=%d : DisplayPipeRequestDeliveryTimeChromaPrefetch = %f\n" , |
4833 | __func__, k, DisplayPipeRequestDeliveryTimeChromaPrefetch[k]); |
4834 | #endif |
4835 | } |
4836 | |
4837 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
4838 | unsigned int cursor_req_per_width; |
4839 | |
4840 | cursor_req_per_width = dml_ceil(a: (double) CursorWidth[k][0] * (double) CursorBPP[k][0] / |
4841 | 256.0 / 8.0, granularity: 1.0); |
4842 | if (NumberOfCursors[k] > 0) { |
4843 | if (VRatio[k] <= 1) { |
4844 | CursorRequestDeliveryTime[k] = (double) CursorWidth[k][0] / |
4845 | HRatio[k] / PixelClock[k] / cursor_req_per_width; |
4846 | } else { |
4847 | CursorRequestDeliveryTime[k] = (double) CursorWidth[k][0] / |
4848 | PSCL_THROUGHPUT[k] / Dppclk[k] / cursor_req_per_width; |
4849 | } |
4850 | if (VRatioPrefetchY[k] <= 1) { |
4851 | CursorRequestDeliveryTimePrefetch[k] = (double) CursorWidth[k][0] / |
4852 | HRatio[k] / PixelClock[k] / cursor_req_per_width; |
4853 | } else { |
4854 | CursorRequestDeliveryTimePrefetch[k] = (double) CursorWidth[k][0] / |
4855 | PSCL_THROUGHPUT[k] / Dppclk[k] / cursor_req_per_width; |
4856 | } |
4857 | } else { |
4858 | CursorRequestDeliveryTime[k] = 0; |
4859 | CursorRequestDeliveryTimePrefetch[k] = 0; |
4860 | } |
4861 | #ifdef __DML_VBA_DEBUG__ |
4862 | dml_print("DML::%s: k=%d : NumberOfCursors = %d\n" , |
4863 | __func__, k, NumberOfCursors[k]); |
4864 | dml_print("DML::%s: k=%d : CursorRequestDeliveryTime = %f\n" , |
4865 | __func__, k, CursorRequestDeliveryTime[k]); |
4866 | dml_print("DML::%s: k=%d : CursorRequestDeliveryTimePrefetch = %f\n" , |
4867 | __func__, k, CursorRequestDeliveryTimePrefetch[k]); |
4868 | #endif |
4869 | } |
4870 | } // CalculatePixelDeliveryTimes |
4871 | |
4872 | void dml32_CalculateMetaAndPTETimes( |
4873 | bool use_one_row_for_frame[], |
4874 | unsigned int NumberOfActiveSurfaces, |
4875 | bool GPUVMEnable, |
4876 | unsigned int MetaChunkSize, |
4877 | unsigned int MinMetaChunkSizeBytes, |
4878 | unsigned int HTotal[], |
4879 | double VRatio[], |
4880 | double VRatioChroma[], |
4881 | double DestinationLinesToRequestRowInVBlank[], |
4882 | double DestinationLinesToRequestRowInImmediateFlip[], |
4883 | bool DCCEnable[], |
4884 | double PixelClock[], |
4885 | unsigned int BytePerPixelY[], |
4886 | unsigned int BytePerPixelC[], |
4887 | enum dm_rotation_angle SourceRotation[], |
4888 | unsigned int dpte_row_height[], |
4889 | unsigned int dpte_row_height_chroma[], |
4890 | unsigned int meta_row_width[], |
4891 | unsigned int meta_row_width_chroma[], |
4892 | unsigned int meta_row_height[], |
4893 | unsigned int meta_row_height_chroma[], |
4894 | unsigned int meta_req_width[], |
4895 | unsigned int meta_req_width_chroma[], |
4896 | unsigned int meta_req_height[], |
4897 | unsigned int meta_req_height_chroma[], |
4898 | unsigned int dpte_group_bytes[], |
4899 | unsigned int PTERequestSizeY[], |
4900 | unsigned int PTERequestSizeC[], |
4901 | unsigned int PixelPTEReqWidthY[], |
4902 | unsigned int PixelPTEReqHeightY[], |
4903 | unsigned int PixelPTEReqWidthC[], |
4904 | unsigned int PixelPTEReqHeightC[], |
4905 | unsigned int dpte_row_width_luma_ub[], |
4906 | unsigned int dpte_row_width_chroma_ub[], |
4907 | |
4908 | /* Output */ |
4909 | double DST_Y_PER_PTE_ROW_NOM_L[], |
4910 | double DST_Y_PER_PTE_ROW_NOM_C[], |
4911 | double DST_Y_PER_META_ROW_NOM_L[], |
4912 | double DST_Y_PER_META_ROW_NOM_C[], |
4913 | double TimePerMetaChunkNominal[], |
4914 | double TimePerChromaMetaChunkNominal[], |
4915 | double TimePerMetaChunkVBlank[], |
4916 | double TimePerChromaMetaChunkVBlank[], |
4917 | double TimePerMetaChunkFlip[], |
4918 | double TimePerChromaMetaChunkFlip[], |
4919 | double time_per_pte_group_nom_luma[], |
4920 | double time_per_pte_group_vblank_luma[], |
4921 | double time_per_pte_group_flip_luma[], |
4922 | double time_per_pte_group_nom_chroma[], |
4923 | double time_per_pte_group_vblank_chroma[], |
4924 | double time_per_pte_group_flip_chroma[]) |
4925 | { |
4926 | unsigned int meta_chunk_width; |
4927 | unsigned int min_meta_chunk_width; |
4928 | unsigned int meta_chunk_per_row_int; |
4929 | unsigned int meta_row_remainder; |
4930 | unsigned int meta_chunk_threshold; |
4931 | unsigned int meta_chunks_per_row_ub; |
4932 | unsigned int meta_chunk_width_chroma; |
4933 | unsigned int min_meta_chunk_width_chroma; |
4934 | unsigned int meta_chunk_per_row_int_chroma; |
4935 | unsigned int meta_row_remainder_chroma; |
4936 | unsigned int meta_chunk_threshold_chroma; |
4937 | unsigned int meta_chunks_per_row_ub_chroma; |
4938 | unsigned int dpte_group_width_luma; |
4939 | unsigned int dpte_groups_per_row_luma_ub; |
4940 | unsigned int dpte_group_width_chroma; |
4941 | unsigned int dpte_groups_per_row_chroma_ub; |
4942 | unsigned int k; |
4943 | |
4944 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
4945 | DST_Y_PER_PTE_ROW_NOM_L[k] = dpte_row_height[k] / VRatio[k]; |
4946 | if (BytePerPixelC[k] == 0) |
4947 | DST_Y_PER_PTE_ROW_NOM_C[k] = 0; |
4948 | else |
4949 | DST_Y_PER_PTE_ROW_NOM_C[k] = dpte_row_height_chroma[k] / VRatioChroma[k]; |
4950 | DST_Y_PER_META_ROW_NOM_L[k] = meta_row_height[k] / VRatio[k]; |
4951 | if (BytePerPixelC[k] == 0) |
4952 | DST_Y_PER_META_ROW_NOM_C[k] = 0; |
4953 | else |
4954 | DST_Y_PER_META_ROW_NOM_C[k] = meta_row_height_chroma[k] / VRatioChroma[k]; |
4955 | } |
4956 | |
4957 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
4958 | if (DCCEnable[k] == true) { |
4959 | meta_chunk_width = MetaChunkSize * 1024 * 256 / BytePerPixelY[k] / meta_row_height[k]; |
4960 | min_meta_chunk_width = MinMetaChunkSizeBytes * 256 / BytePerPixelY[k] / meta_row_height[k]; |
4961 | meta_chunk_per_row_int = meta_row_width[k] / meta_chunk_width; |
4962 | meta_row_remainder = meta_row_width[k] % meta_chunk_width; |
4963 | if (!IsVertical(Scan: SourceRotation[k])) |
4964 | meta_chunk_threshold = 2 * min_meta_chunk_width - meta_req_width[k]; |
4965 | else |
4966 | meta_chunk_threshold = 2 * min_meta_chunk_width - meta_req_height[k]; |
4967 | |
4968 | if (meta_row_remainder <= meta_chunk_threshold) |
4969 | meta_chunks_per_row_ub = meta_chunk_per_row_int + 1; |
4970 | else |
4971 | meta_chunks_per_row_ub = meta_chunk_per_row_int + 2; |
4972 | |
4973 | TimePerMetaChunkNominal[k] = meta_row_height[k] / VRatio[k] * |
4974 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub; |
4975 | TimePerMetaChunkVBlank[k] = DestinationLinesToRequestRowInVBlank[k] * |
4976 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub; |
4977 | TimePerMetaChunkFlip[k] = DestinationLinesToRequestRowInImmediateFlip[k] * |
4978 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub; |
4979 | if (BytePerPixelC[k] == 0) { |
4980 | TimePerChromaMetaChunkNominal[k] = 0; |
4981 | TimePerChromaMetaChunkVBlank[k] = 0; |
4982 | TimePerChromaMetaChunkFlip[k] = 0; |
4983 | } else { |
4984 | meta_chunk_width_chroma = MetaChunkSize * 1024 * 256 / BytePerPixelC[k] / |
4985 | meta_row_height_chroma[k]; |
4986 | min_meta_chunk_width_chroma = MinMetaChunkSizeBytes * 256 / BytePerPixelC[k] / |
4987 | meta_row_height_chroma[k]; |
4988 | meta_chunk_per_row_int_chroma = (double) meta_row_width_chroma[k] / |
4989 | meta_chunk_width_chroma; |
4990 | meta_row_remainder_chroma = meta_row_width_chroma[k] % meta_chunk_width_chroma; |
4991 | if (!IsVertical(Scan: SourceRotation[k])) { |
4992 | meta_chunk_threshold_chroma = 2 * min_meta_chunk_width_chroma - |
4993 | meta_req_width_chroma[k]; |
4994 | } else { |
4995 | meta_chunk_threshold_chroma = 2 * min_meta_chunk_width_chroma - |
4996 | meta_req_height_chroma[k]; |
4997 | } |
4998 | if (meta_row_remainder_chroma <= meta_chunk_threshold_chroma) |
4999 | meta_chunks_per_row_ub_chroma = meta_chunk_per_row_int_chroma + 1; |
5000 | else |
5001 | meta_chunks_per_row_ub_chroma = meta_chunk_per_row_int_chroma + 2; |
5002 | |
5003 | TimePerChromaMetaChunkNominal[k] = meta_row_height_chroma[k] / VRatioChroma[k] * |
5004 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub_chroma; |
5005 | TimePerChromaMetaChunkVBlank[k] = DestinationLinesToRequestRowInVBlank[k] * |
5006 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub_chroma; |
5007 | TimePerChromaMetaChunkFlip[k] = DestinationLinesToRequestRowInImmediateFlip[k] * |
5008 | HTotal[k] / PixelClock[k] / meta_chunks_per_row_ub_chroma; |
5009 | } |
5010 | } else { |
5011 | TimePerMetaChunkNominal[k] = 0; |
5012 | TimePerMetaChunkVBlank[k] = 0; |
5013 | TimePerMetaChunkFlip[k] = 0; |
5014 | TimePerChromaMetaChunkNominal[k] = 0; |
5015 | TimePerChromaMetaChunkVBlank[k] = 0; |
5016 | TimePerChromaMetaChunkFlip[k] = 0; |
5017 | } |
5018 | } |
5019 | |
5020 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5021 | if (GPUVMEnable == true) { |
5022 | if (!IsVertical(Scan: SourceRotation[k])) { |
5023 | dpte_group_width_luma = (double) dpte_group_bytes[k] / |
5024 | (double) PTERequestSizeY[k] * PixelPTEReqWidthY[k]; |
5025 | } else { |
5026 | dpte_group_width_luma = (double) dpte_group_bytes[k] / |
5027 | (double) PTERequestSizeY[k] * PixelPTEReqHeightY[k]; |
5028 | } |
5029 | |
5030 | if (use_one_row_for_frame[k]) { |
5031 | dpte_groups_per_row_luma_ub = dml_ceil(a: (double) dpte_row_width_luma_ub[k] / |
5032 | (double) dpte_group_width_luma / 2.0, granularity: 1.0); |
5033 | } else { |
5034 | dpte_groups_per_row_luma_ub = dml_ceil(a: (double) dpte_row_width_luma_ub[k] / |
5035 | (double) dpte_group_width_luma, granularity: 1.0); |
5036 | } |
5037 | #ifdef __DML_VBA_DEBUG__ |
5038 | dml_print("DML::%s: k=%0d, use_one_row_for_frame = %d\n" , |
5039 | __func__, k, use_one_row_for_frame[k]); |
5040 | dml_print("DML::%s: k=%0d, dpte_group_bytes = %d\n" , |
5041 | __func__, k, dpte_group_bytes[k]); |
5042 | dml_print("DML::%s: k=%0d, PTERequestSizeY = %d\n" , |
5043 | __func__, k, PTERequestSizeY[k]); |
5044 | dml_print("DML::%s: k=%0d, PixelPTEReqWidthY = %d\n" , |
5045 | __func__, k, PixelPTEReqWidthY[k]); |
5046 | dml_print("DML::%s: k=%0d, PixelPTEReqHeightY = %d\n" , |
5047 | __func__, k, PixelPTEReqHeightY[k]); |
5048 | dml_print("DML::%s: k=%0d, dpte_row_width_luma_ub = %d\n" , |
5049 | __func__, k, dpte_row_width_luma_ub[k]); |
5050 | dml_print("DML::%s: k=%0d, dpte_group_width_luma = %d\n" , |
5051 | __func__, k, dpte_group_width_luma); |
5052 | dml_print("DML::%s: k=%0d, dpte_groups_per_row_luma_ub = %d\n" , |
5053 | __func__, k, dpte_groups_per_row_luma_ub); |
5054 | #endif |
5055 | |
5056 | time_per_pte_group_nom_luma[k] = DST_Y_PER_PTE_ROW_NOM_L[k] * |
5057 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_luma_ub; |
5058 | time_per_pte_group_vblank_luma[k] = DestinationLinesToRequestRowInVBlank[k] * |
5059 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_luma_ub; |
5060 | time_per_pte_group_flip_luma[k] = DestinationLinesToRequestRowInImmediateFlip[k] * |
5061 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_luma_ub; |
5062 | if (BytePerPixelC[k] == 0) { |
5063 | time_per_pte_group_nom_chroma[k] = 0; |
5064 | time_per_pte_group_vblank_chroma[k] = 0; |
5065 | time_per_pte_group_flip_chroma[k] = 0; |
5066 | } else { |
5067 | if (!IsVertical(Scan: SourceRotation[k])) { |
5068 | dpte_group_width_chroma = (double) dpte_group_bytes[k] / |
5069 | (double) PTERequestSizeC[k] * PixelPTEReqWidthC[k]; |
5070 | } else { |
5071 | dpte_group_width_chroma = (double) dpte_group_bytes[k] / |
5072 | (double) PTERequestSizeC[k] * PixelPTEReqHeightC[k]; |
5073 | } |
5074 | |
5075 | if (use_one_row_for_frame[k]) { |
5076 | dpte_groups_per_row_chroma_ub = dml_ceil(a: (double) dpte_row_width_chroma_ub[k] / |
5077 | (double) dpte_group_width_chroma / 2.0, granularity: 1.0); |
5078 | } else { |
5079 | dpte_groups_per_row_chroma_ub = dml_ceil(a: (double) dpte_row_width_chroma_ub[k] / |
5080 | (double) dpte_group_width_chroma, granularity: 1.0); |
5081 | } |
5082 | #ifdef __DML_VBA_DEBUG__ |
5083 | dml_print("DML::%s: k=%0d, dpte_row_width_chroma_ub = %d\n" , |
5084 | __func__, k, dpte_row_width_chroma_ub[k]); |
5085 | dml_print("DML::%s: k=%0d, dpte_group_width_chroma = %d\n" , |
5086 | __func__, k, dpte_group_width_chroma); |
5087 | dml_print("DML::%s: k=%0d, dpte_groups_per_row_chroma_ub = %d\n" , |
5088 | __func__, k, dpte_groups_per_row_chroma_ub); |
5089 | #endif |
5090 | time_per_pte_group_nom_chroma[k] = DST_Y_PER_PTE_ROW_NOM_C[k] * |
5091 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_chroma_ub; |
5092 | time_per_pte_group_vblank_chroma[k] = DestinationLinesToRequestRowInVBlank[k] * |
5093 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_chroma_ub; |
5094 | time_per_pte_group_flip_chroma[k] = DestinationLinesToRequestRowInImmediateFlip[k] * |
5095 | HTotal[k] / PixelClock[k] / dpte_groups_per_row_chroma_ub; |
5096 | } |
5097 | } else { |
5098 | time_per_pte_group_nom_luma[k] = 0; |
5099 | time_per_pte_group_vblank_luma[k] = 0; |
5100 | time_per_pte_group_flip_luma[k] = 0; |
5101 | time_per_pte_group_nom_chroma[k] = 0; |
5102 | time_per_pte_group_vblank_chroma[k] = 0; |
5103 | time_per_pte_group_flip_chroma[k] = 0; |
5104 | } |
5105 | #ifdef __DML_VBA_DEBUG__ |
5106 | dml_print("DML::%s: k=%0d, DestinationLinesToRequestRowInVBlank = %f\n" , |
5107 | __func__, k, DestinationLinesToRequestRowInVBlank[k]); |
5108 | dml_print("DML::%s: k=%0d, DestinationLinesToRequestRowInImmediateFlip = %f\n" , |
5109 | __func__, k, DestinationLinesToRequestRowInImmediateFlip[k]); |
5110 | dml_print("DML::%s: k=%0d, DST_Y_PER_PTE_ROW_NOM_L = %f\n" , |
5111 | __func__, k, DST_Y_PER_PTE_ROW_NOM_L[k]); |
5112 | dml_print("DML::%s: k=%0d, DST_Y_PER_PTE_ROW_NOM_C = %f\n" , |
5113 | __func__, k, DST_Y_PER_PTE_ROW_NOM_C[k]); |
5114 | dml_print("DML::%s: k=%0d, DST_Y_PER_META_ROW_NOM_L = %f\n" , |
5115 | __func__, k, DST_Y_PER_META_ROW_NOM_L[k]); |
5116 | dml_print("DML::%s: k=%0d, DST_Y_PER_META_ROW_NOM_C = %f\n" , |
5117 | __func__, k, DST_Y_PER_META_ROW_NOM_C[k]); |
5118 | dml_print("DML::%s: k=%0d, TimePerMetaChunkNominal = %f\n" , |
5119 | __func__, k, TimePerMetaChunkNominal[k]); |
5120 | dml_print("DML::%s: k=%0d, TimePerMetaChunkVBlank = %f\n" , |
5121 | __func__, k, TimePerMetaChunkVBlank[k]); |
5122 | dml_print("DML::%s: k=%0d, TimePerMetaChunkFlip = %f\n" , |
5123 | __func__, k, TimePerMetaChunkFlip[k]); |
5124 | dml_print("DML::%s: k=%0d, TimePerChromaMetaChunkNominal = %f\n" , |
5125 | __func__, k, TimePerChromaMetaChunkNominal[k]); |
5126 | dml_print("DML::%s: k=%0d, TimePerChromaMetaChunkVBlank = %f\n" , |
5127 | __func__, k, TimePerChromaMetaChunkVBlank[k]); |
5128 | dml_print("DML::%s: k=%0d, TimePerChromaMetaChunkFlip = %f\n" , |
5129 | __func__, k, TimePerChromaMetaChunkFlip[k]); |
5130 | dml_print("DML::%s: k=%0d, time_per_pte_group_nom_luma = %f\n" , |
5131 | __func__, k, time_per_pte_group_nom_luma[k]); |
5132 | dml_print("DML::%s: k=%0d, time_per_pte_group_vblank_luma = %f\n" , |
5133 | __func__, k, time_per_pte_group_vblank_luma[k]); |
5134 | dml_print("DML::%s: k=%0d, time_per_pte_group_flip_luma = %f\n" , |
5135 | __func__, k, time_per_pte_group_flip_luma[k]); |
5136 | dml_print("DML::%s: k=%0d, time_per_pte_group_nom_chroma = %f\n" , |
5137 | __func__, k, time_per_pte_group_nom_chroma[k]); |
5138 | dml_print("DML::%s: k=%0d, time_per_pte_group_vblank_chroma = %f\n" , |
5139 | __func__, k, time_per_pte_group_vblank_chroma[k]); |
5140 | dml_print("DML::%s: k=%0d, time_per_pte_group_flip_chroma = %f\n" , |
5141 | __func__, k, time_per_pte_group_flip_chroma[k]); |
5142 | #endif |
5143 | } |
5144 | } // CalculateMetaAndPTETimes |
5145 | |
5146 | void dml32_CalculateVMGroupAndRequestTimes( |
5147 | unsigned int NumberOfActiveSurfaces, |
5148 | bool GPUVMEnable, |
5149 | unsigned int GPUVMMaxPageTableLevels, |
5150 | unsigned int HTotal[], |
5151 | unsigned int BytePerPixelC[], |
5152 | double DestinationLinesToRequestVMInVBlank[], |
5153 | double DestinationLinesToRequestVMInImmediateFlip[], |
5154 | bool DCCEnable[], |
5155 | double PixelClock[], |
5156 | unsigned int dpte_row_width_luma_ub[], |
5157 | unsigned int dpte_row_width_chroma_ub[], |
5158 | unsigned int vm_group_bytes[], |
5159 | unsigned int dpde0_bytes_per_frame_ub_l[], |
5160 | unsigned int dpde0_bytes_per_frame_ub_c[], |
5161 | unsigned int meta_pte_bytes_per_frame_ub_l[], |
5162 | unsigned int meta_pte_bytes_per_frame_ub_c[], |
5163 | |
5164 | /* Output */ |
5165 | double TimePerVMGroupVBlank[], |
5166 | double TimePerVMGroupFlip[], |
5167 | double TimePerVMRequestVBlank[], |
5168 | double TimePerVMRequestFlip[]) |
5169 | { |
5170 | unsigned int k; |
5171 | unsigned int num_group_per_lower_vm_stage; |
5172 | unsigned int num_req_per_lower_vm_stage; |
5173 | |
5174 | #ifdef __DML_VBA_DEBUG__ |
5175 | dml_print("DML::%s: NumberOfActiveSurfaces = %d\n" , __func__, NumberOfActiveSurfaces); |
5176 | dml_print("DML::%s: GPUVMEnable = %d\n" , __func__, GPUVMEnable); |
5177 | #endif |
5178 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5179 | |
5180 | #ifdef __DML_VBA_DEBUG__ |
5181 | dml_print("DML::%s: k=%0d, DCCEnable = %d\n" , __func__, k, DCCEnable[k]); |
5182 | dml_print("DML::%s: k=%0d, vm_group_bytes = %d\n" , __func__, k, vm_group_bytes[k]); |
5183 | dml_print("DML::%s: k=%0d, dpde0_bytes_per_frame_ub_l = %d\n" , |
5184 | __func__, k, dpde0_bytes_per_frame_ub_l[k]); |
5185 | dml_print("DML::%s: k=%0d, dpde0_bytes_per_frame_ub_c = %d\n" , |
5186 | __func__, k, dpde0_bytes_per_frame_ub_c[k]); |
5187 | dml_print("DML::%s: k=%0d, meta_pte_bytes_per_frame_ub_l = %d\n" , |
5188 | __func__, k, meta_pte_bytes_per_frame_ub_l[k]); |
5189 | dml_print("DML::%s: k=%0d, meta_pte_bytes_per_frame_ub_c = %d\n" , |
5190 | __func__, k, meta_pte_bytes_per_frame_ub_c[k]); |
5191 | #endif |
5192 | |
5193 | if (GPUVMEnable == true && (DCCEnable[k] == true || GPUVMMaxPageTableLevels > 1)) { |
5194 | if (DCCEnable[k] == false) { |
5195 | if (BytePerPixelC[k] > 0) { |
5196 | num_group_per_lower_vm_stage = dml_ceil( |
5197 | a: (double) (dpde0_bytes_per_frame_ub_l[k]) / |
5198 | (double) (vm_group_bytes[k]), granularity: 1.0) + |
5199 | dml_ceil(a: (double) (dpde0_bytes_per_frame_ub_c[k]) / |
5200 | (double) (vm_group_bytes[k]), granularity: 1.0); |
5201 | } else { |
5202 | num_group_per_lower_vm_stage = dml_ceil( |
5203 | a: (double) (dpde0_bytes_per_frame_ub_l[k]) / |
5204 | (double) (vm_group_bytes[k]), granularity: 1.0); |
5205 | } |
5206 | } else { |
5207 | if (GPUVMMaxPageTableLevels == 1) { |
5208 | if (BytePerPixelC[k] > 0) { |
5209 | num_group_per_lower_vm_stage = dml_ceil( |
5210 | a: (double) (meta_pte_bytes_per_frame_ub_l[k]) / |
5211 | (double) (vm_group_bytes[k]), granularity: 1.0) + |
5212 | dml_ceil(a: (double) (meta_pte_bytes_per_frame_ub_c[k]) / |
5213 | (double) (vm_group_bytes[k]), granularity: 1.0); |
5214 | } else { |
5215 | num_group_per_lower_vm_stage = dml_ceil( |
5216 | a: (double) (meta_pte_bytes_per_frame_ub_l[k]) / |
5217 | (double) (vm_group_bytes[k]), granularity: 1.0); |
5218 | } |
5219 | } else { |
5220 | if (BytePerPixelC[k] > 0) { |
5221 | num_group_per_lower_vm_stage = 2 + dml_ceil( |
5222 | a: (double) (dpde0_bytes_per_frame_ub_l[k]) / |
5223 | (double) (vm_group_bytes[k]), granularity: 1) + |
5224 | dml_ceil(a: (double) (dpde0_bytes_per_frame_ub_c[k]) / |
5225 | (double) (vm_group_bytes[k]), granularity: 1) + |
5226 | dml_ceil(a: (double) (meta_pte_bytes_per_frame_ub_l[k]) / |
5227 | (double) (vm_group_bytes[k]), granularity: 1) + |
5228 | dml_ceil(a: (double) (meta_pte_bytes_per_frame_ub_c[k]) / |
5229 | (double) (vm_group_bytes[k]), granularity: 1); |
5230 | } else { |
5231 | num_group_per_lower_vm_stage = 1 + dml_ceil( |
5232 | a: (double) (dpde0_bytes_per_frame_ub_l[k]) / |
5233 | (double) (vm_group_bytes[k]), granularity: 1) + dml_ceil( |
5234 | a: (double) (meta_pte_bytes_per_frame_ub_l[k]) / |
5235 | (double) (vm_group_bytes[k]), granularity: 1); |
5236 | } |
5237 | } |
5238 | } |
5239 | |
5240 | if (DCCEnable[k] == false) { |
5241 | if (BytePerPixelC[k] > 0) { |
5242 | num_req_per_lower_vm_stage = dpde0_bytes_per_frame_ub_l[k] / 64 + |
5243 | dpde0_bytes_per_frame_ub_c[k] / 64; |
5244 | } else { |
5245 | num_req_per_lower_vm_stage = dpde0_bytes_per_frame_ub_l[k] / 64; |
5246 | } |
5247 | } else { |
5248 | if (GPUVMMaxPageTableLevels == 1) { |
5249 | if (BytePerPixelC[k] > 0) { |
5250 | num_req_per_lower_vm_stage = meta_pte_bytes_per_frame_ub_l[k] / 64 + |
5251 | meta_pte_bytes_per_frame_ub_c[k] / 64; |
5252 | } else { |
5253 | num_req_per_lower_vm_stage = meta_pte_bytes_per_frame_ub_l[k] / 64; |
5254 | } |
5255 | } else { |
5256 | if (BytePerPixelC[k] > 0) { |
5257 | num_req_per_lower_vm_stage = dpde0_bytes_per_frame_ub_l[k] / |
5258 | 64 + dpde0_bytes_per_frame_ub_c[k] / 64 + |
5259 | meta_pte_bytes_per_frame_ub_l[k] / 64 + |
5260 | meta_pte_bytes_per_frame_ub_c[k] / 64; |
5261 | } else { |
5262 | num_req_per_lower_vm_stage = dpde0_bytes_per_frame_ub_l[k] / |
5263 | 64 + meta_pte_bytes_per_frame_ub_l[k] / 64; |
5264 | } |
5265 | } |
5266 | } |
5267 | |
5268 | TimePerVMGroupVBlank[k] = DestinationLinesToRequestVMInVBlank[k] * |
5269 | HTotal[k] / PixelClock[k] / num_group_per_lower_vm_stage; |
5270 | TimePerVMGroupFlip[k] = DestinationLinesToRequestVMInImmediateFlip[k] * |
5271 | HTotal[k] / PixelClock[k] / num_group_per_lower_vm_stage; |
5272 | TimePerVMRequestVBlank[k] = DestinationLinesToRequestVMInVBlank[k] * |
5273 | HTotal[k] / PixelClock[k] / num_req_per_lower_vm_stage; |
5274 | TimePerVMRequestFlip[k] = DestinationLinesToRequestVMInImmediateFlip[k] * |
5275 | HTotal[k] / PixelClock[k] / num_req_per_lower_vm_stage; |
5276 | |
5277 | if (GPUVMMaxPageTableLevels > 2) { |
5278 | TimePerVMGroupVBlank[k] = TimePerVMGroupVBlank[k] / 2; |
5279 | TimePerVMGroupFlip[k] = TimePerVMGroupFlip[k] / 2; |
5280 | TimePerVMRequestVBlank[k] = TimePerVMRequestVBlank[k] / 2; |
5281 | TimePerVMRequestFlip[k] = TimePerVMRequestFlip[k] / 2; |
5282 | } |
5283 | |
5284 | } else { |
5285 | TimePerVMGroupVBlank[k] = 0; |
5286 | TimePerVMGroupFlip[k] = 0; |
5287 | TimePerVMRequestVBlank[k] = 0; |
5288 | TimePerVMRequestFlip[k] = 0; |
5289 | } |
5290 | |
5291 | #ifdef __DML_VBA_DEBUG__ |
5292 | dml_print("DML::%s: k=%0d, TimePerVMGroupVBlank = %f\n" , __func__, k, TimePerVMGroupVBlank[k]); |
5293 | dml_print("DML::%s: k=%0d, TimePerVMGroupFlip = %f\n" , __func__, k, TimePerVMGroupFlip[k]); |
5294 | dml_print("DML::%s: k=%0d, TimePerVMRequestVBlank = %f\n" , __func__, k, TimePerVMRequestVBlank[k]); |
5295 | dml_print("DML::%s: k=%0d, TimePerVMRequestFlip = %f\n" , __func__, k, TimePerVMRequestFlip[k]); |
5296 | #endif |
5297 | } |
5298 | } // CalculateVMGroupAndRequestTimes |
5299 | |
5300 | void dml32_CalculateDCCConfiguration( |
5301 | bool DCCEnabled, |
5302 | bool DCCProgrammingAssumesScanDirectionUnknown, |
5303 | enum source_format_class SourcePixelFormat, |
5304 | unsigned int SurfaceWidthLuma, |
5305 | unsigned int SurfaceWidthChroma, |
5306 | unsigned int SurfaceHeightLuma, |
5307 | unsigned int SurfaceHeightChroma, |
5308 | unsigned int nomDETInKByte, |
5309 | unsigned int RequestHeight256ByteLuma, |
5310 | unsigned int RequestHeight256ByteChroma, |
5311 | enum dm_swizzle_mode TilingFormat, |
5312 | unsigned int BytePerPixelY, |
5313 | unsigned int BytePerPixelC, |
5314 | double BytePerPixelDETY, |
5315 | double BytePerPixelDETC, |
5316 | enum dm_rotation_angle SourceRotation, |
5317 | /* Output */ |
5318 | unsigned int *MaxUncompressedBlockLuma, |
5319 | unsigned int *MaxUncompressedBlockChroma, |
5320 | unsigned int *MaxCompressedBlockLuma, |
5321 | unsigned int *MaxCompressedBlockChroma, |
5322 | unsigned int *IndependentBlockLuma, |
5323 | unsigned int *IndependentBlockChroma) |
5324 | { |
5325 | typedef enum { |
5326 | REQ_256Bytes, |
5327 | REQ_128BytesNonContiguous, |
5328 | REQ_128BytesContiguous, |
5329 | REQ_NA |
5330 | } RequestType; |
5331 | |
5332 | RequestType RequestLuma; |
5333 | RequestType RequestChroma; |
5334 | |
5335 | unsigned int segment_order_horz_contiguous_luma; |
5336 | unsigned int segment_order_horz_contiguous_chroma; |
5337 | unsigned int segment_order_vert_contiguous_luma; |
5338 | unsigned int segment_order_vert_contiguous_chroma; |
5339 | unsigned int req128_horz_wc_l; |
5340 | unsigned int req128_horz_wc_c; |
5341 | unsigned int req128_vert_wc_l; |
5342 | unsigned int req128_vert_wc_c; |
5343 | unsigned int MAS_vp_horz_limit; |
5344 | unsigned int MAS_vp_vert_limit; |
5345 | unsigned int max_vp_horz_width; |
5346 | unsigned int max_vp_vert_height; |
5347 | unsigned int eff_surf_width_l; |
5348 | unsigned int eff_surf_width_c; |
5349 | unsigned int eff_surf_height_l; |
5350 | unsigned int eff_surf_height_c; |
5351 | unsigned int full_swath_bytes_horz_wc_l; |
5352 | unsigned int full_swath_bytes_horz_wc_c; |
5353 | unsigned int full_swath_bytes_vert_wc_l; |
5354 | unsigned int full_swath_bytes_vert_wc_c; |
5355 | unsigned int DETBufferSizeForDCC = nomDETInKByte * 1024; |
5356 | |
5357 | unsigned int yuv420; |
5358 | unsigned int horz_div_l; |
5359 | unsigned int horz_div_c; |
5360 | unsigned int vert_div_l; |
5361 | unsigned int vert_div_c; |
5362 | |
5363 | unsigned int swath_buf_size; |
5364 | double detile_buf_vp_horz_limit; |
5365 | double detile_buf_vp_vert_limit; |
5366 | |
5367 | yuv420 = ((SourcePixelFormat == dm_420_8 || SourcePixelFormat == dm_420_10 || |
5368 | SourcePixelFormat == dm_420_12) ? 1 : 0); |
5369 | horz_div_l = 1; |
5370 | horz_div_c = 1; |
5371 | vert_div_l = 1; |
5372 | vert_div_c = 1; |
5373 | |
5374 | if (BytePerPixelY == 1) |
5375 | vert_div_l = 0; |
5376 | if (BytePerPixelC == 1) |
5377 | vert_div_c = 0; |
5378 | |
5379 | if (BytePerPixelC == 0) { |
5380 | swath_buf_size = DETBufferSizeForDCC / 2 - 2 * 256; |
5381 | detile_buf_vp_horz_limit = (double) swath_buf_size / ((double) RequestHeight256ByteLuma * |
5382 | BytePerPixelY / (1 + horz_div_l)); |
5383 | detile_buf_vp_vert_limit = (double) swath_buf_size / (256.0 / RequestHeight256ByteLuma / |
5384 | (1 + vert_div_l)); |
5385 | } else { |
5386 | swath_buf_size = DETBufferSizeForDCC / 2 - 2 * 2 * 256; |
5387 | detile_buf_vp_horz_limit = (double) swath_buf_size / ((double) RequestHeight256ByteLuma * |
5388 | BytePerPixelY / (1 + horz_div_l) + (double) RequestHeight256ByteChroma * |
5389 | BytePerPixelC / (1 + horz_div_c) / (1 + yuv420)); |
5390 | detile_buf_vp_vert_limit = (double) swath_buf_size / (256.0 / RequestHeight256ByteLuma / |
5391 | (1 + vert_div_l) + 256.0 / RequestHeight256ByteChroma / |
5392 | (1 + vert_div_c) / (1 + yuv420)); |
5393 | } |
5394 | |
5395 | if (SourcePixelFormat == dm_420_10) { |
5396 | detile_buf_vp_horz_limit = 1.5 * detile_buf_vp_horz_limit; |
5397 | detile_buf_vp_vert_limit = 1.5 * detile_buf_vp_vert_limit; |
5398 | } |
5399 | |
5400 | detile_buf_vp_horz_limit = dml_floor(a: detile_buf_vp_horz_limit - 1, granularity: 16); |
5401 | detile_buf_vp_vert_limit = dml_floor(a: detile_buf_vp_vert_limit - 1, granularity: 16); |
5402 | |
5403 | MAS_vp_horz_limit = SourcePixelFormat == dm_rgbe_alpha ? 3840 : 6144; |
5404 | MAS_vp_vert_limit = SourcePixelFormat == dm_rgbe_alpha ? 3840 : (BytePerPixelY == 8 ? 3072 : 6144); |
5405 | max_vp_horz_width = dml_min(a: (double) MAS_vp_horz_limit, b: detile_buf_vp_horz_limit); |
5406 | max_vp_vert_height = dml_min(a: (double) MAS_vp_vert_limit, b: detile_buf_vp_vert_limit); |
5407 | eff_surf_width_l = (SurfaceWidthLuma > max_vp_horz_width ? max_vp_horz_width : SurfaceWidthLuma); |
5408 | eff_surf_width_c = eff_surf_width_l / (1 + yuv420); |
5409 | eff_surf_height_l = (SurfaceHeightLuma > max_vp_vert_height ? max_vp_vert_height : SurfaceHeightLuma); |
5410 | eff_surf_height_c = eff_surf_height_l / (1 + yuv420); |
5411 | |
5412 | full_swath_bytes_horz_wc_l = eff_surf_width_l * RequestHeight256ByteLuma * BytePerPixelY; |
5413 | full_swath_bytes_vert_wc_l = eff_surf_height_l * 256 / RequestHeight256ByteLuma; |
5414 | if (BytePerPixelC > 0) { |
5415 | full_swath_bytes_horz_wc_c = eff_surf_width_c * RequestHeight256ByteChroma * BytePerPixelC; |
5416 | full_swath_bytes_vert_wc_c = eff_surf_height_c * 256 / RequestHeight256ByteChroma; |
5417 | } else { |
5418 | full_swath_bytes_horz_wc_c = 0; |
5419 | full_swath_bytes_vert_wc_c = 0; |
5420 | } |
5421 | |
5422 | if (SourcePixelFormat == dm_420_10) { |
5423 | full_swath_bytes_horz_wc_l = dml_ceil(a: (double) full_swath_bytes_horz_wc_l * 2.0 / 3.0, granularity: 256.0); |
5424 | full_swath_bytes_horz_wc_c = dml_ceil(a: (double) full_swath_bytes_horz_wc_c * 2.0 / 3.0, granularity: 256.0); |
5425 | full_swath_bytes_vert_wc_l = dml_ceil(a: (double) full_swath_bytes_vert_wc_l * 2.0 / 3.0, granularity: 256.0); |
5426 | full_swath_bytes_vert_wc_c = dml_ceil(a: (double) full_swath_bytes_vert_wc_c * 2.0 / 3.0, granularity: 256.0); |
5427 | } |
5428 | |
5429 | if (2 * full_swath_bytes_horz_wc_l + 2 * full_swath_bytes_horz_wc_c <= DETBufferSizeForDCC) { |
5430 | req128_horz_wc_l = 0; |
5431 | req128_horz_wc_c = 0; |
5432 | } else if (full_swath_bytes_horz_wc_l < 1.5 * full_swath_bytes_horz_wc_c && 2 * full_swath_bytes_horz_wc_l + |
5433 | full_swath_bytes_horz_wc_c <= DETBufferSizeForDCC) { |
5434 | req128_horz_wc_l = 0; |
5435 | req128_horz_wc_c = 1; |
5436 | } else if (full_swath_bytes_horz_wc_l >= 1.5 * full_swath_bytes_horz_wc_c && full_swath_bytes_horz_wc_l + 2 * |
5437 | full_swath_bytes_horz_wc_c <= DETBufferSizeForDCC) { |
5438 | req128_horz_wc_l = 1; |
5439 | req128_horz_wc_c = 0; |
5440 | } else { |
5441 | req128_horz_wc_l = 1; |
5442 | req128_horz_wc_c = 1; |
5443 | } |
5444 | |
5445 | if (2 * full_swath_bytes_vert_wc_l + 2 * full_swath_bytes_vert_wc_c <= DETBufferSizeForDCC) { |
5446 | req128_vert_wc_l = 0; |
5447 | req128_vert_wc_c = 0; |
5448 | } else if (full_swath_bytes_vert_wc_l < 1.5 * full_swath_bytes_vert_wc_c && 2 * |
5449 | full_swath_bytes_vert_wc_l + full_swath_bytes_vert_wc_c <= DETBufferSizeForDCC) { |
5450 | req128_vert_wc_l = 0; |
5451 | req128_vert_wc_c = 1; |
5452 | } else if (full_swath_bytes_vert_wc_l >= 1.5 * full_swath_bytes_vert_wc_c && |
5453 | full_swath_bytes_vert_wc_l + 2 * full_swath_bytes_vert_wc_c <= DETBufferSizeForDCC) { |
5454 | req128_vert_wc_l = 1; |
5455 | req128_vert_wc_c = 0; |
5456 | } else { |
5457 | req128_vert_wc_l = 1; |
5458 | req128_vert_wc_c = 1; |
5459 | } |
5460 | |
5461 | if (BytePerPixelY == 2) { |
5462 | segment_order_horz_contiguous_luma = 0; |
5463 | segment_order_vert_contiguous_luma = 1; |
5464 | } else { |
5465 | segment_order_horz_contiguous_luma = 1; |
5466 | segment_order_vert_contiguous_luma = 0; |
5467 | } |
5468 | |
5469 | if (BytePerPixelC == 2) { |
5470 | segment_order_horz_contiguous_chroma = 0; |
5471 | segment_order_vert_contiguous_chroma = 1; |
5472 | } else { |
5473 | segment_order_horz_contiguous_chroma = 1; |
5474 | segment_order_vert_contiguous_chroma = 0; |
5475 | } |
5476 | #ifdef __DML_VBA_DEBUG__ |
5477 | dml_print("DML::%s: DCCEnabled = %d\n" , __func__, DCCEnabled); |
5478 | dml_print("DML::%s: nomDETInKByte = %d\n" , __func__, nomDETInKByte); |
5479 | dml_print("DML::%s: DETBufferSizeForDCC = %d\n" , __func__, DETBufferSizeForDCC); |
5480 | dml_print("DML::%s: req128_horz_wc_l = %d\n" , __func__, req128_horz_wc_l); |
5481 | dml_print("DML::%s: req128_horz_wc_c = %d\n" , __func__, req128_horz_wc_c); |
5482 | dml_print("DML::%s: full_swath_bytes_horz_wc_l = %d\n" , __func__, full_swath_bytes_horz_wc_l); |
5483 | dml_print("DML::%s: full_swath_bytes_vert_wc_c = %d\n" , __func__, full_swath_bytes_vert_wc_c); |
5484 | dml_print("DML::%s: segment_order_horz_contiguous_luma = %d\n" , __func__, segment_order_horz_contiguous_luma); |
5485 | dml_print("DML::%s: segment_order_horz_contiguous_chroma = %d\n" , |
5486 | __func__, segment_order_horz_contiguous_chroma); |
5487 | #endif |
5488 | |
5489 | if (DCCProgrammingAssumesScanDirectionUnknown == true) { |
5490 | if (req128_horz_wc_l == 0 && req128_vert_wc_l == 0) |
5491 | RequestLuma = REQ_256Bytes; |
5492 | else if ((req128_horz_wc_l == 1 && segment_order_horz_contiguous_luma == 0) || |
5493 | (req128_vert_wc_l == 1 && segment_order_vert_contiguous_luma == 0)) |
5494 | RequestLuma = REQ_128BytesNonContiguous; |
5495 | else |
5496 | RequestLuma = REQ_128BytesContiguous; |
5497 | |
5498 | if (req128_horz_wc_c == 0 && req128_vert_wc_c == 0) |
5499 | RequestChroma = REQ_256Bytes; |
5500 | else if ((req128_horz_wc_c == 1 && segment_order_horz_contiguous_chroma == 0) || |
5501 | (req128_vert_wc_c == 1 && segment_order_vert_contiguous_chroma == 0)) |
5502 | RequestChroma = REQ_128BytesNonContiguous; |
5503 | else |
5504 | RequestChroma = REQ_128BytesContiguous; |
5505 | |
5506 | } else if (!IsVertical(Scan: SourceRotation)) { |
5507 | if (req128_horz_wc_l == 0) |
5508 | RequestLuma = REQ_256Bytes; |
5509 | else if (segment_order_horz_contiguous_luma == 0) |
5510 | RequestLuma = REQ_128BytesNonContiguous; |
5511 | else |
5512 | RequestLuma = REQ_128BytesContiguous; |
5513 | |
5514 | if (req128_horz_wc_c == 0) |
5515 | RequestChroma = REQ_256Bytes; |
5516 | else if (segment_order_horz_contiguous_chroma == 0) |
5517 | RequestChroma = REQ_128BytesNonContiguous; |
5518 | else |
5519 | RequestChroma = REQ_128BytesContiguous; |
5520 | |
5521 | } else { |
5522 | if (req128_vert_wc_l == 0) |
5523 | RequestLuma = REQ_256Bytes; |
5524 | else if (segment_order_vert_contiguous_luma == 0) |
5525 | RequestLuma = REQ_128BytesNonContiguous; |
5526 | else |
5527 | RequestLuma = REQ_128BytesContiguous; |
5528 | |
5529 | if (req128_vert_wc_c == 0) |
5530 | RequestChroma = REQ_256Bytes; |
5531 | else if (segment_order_vert_contiguous_chroma == 0) |
5532 | RequestChroma = REQ_128BytesNonContiguous; |
5533 | else |
5534 | RequestChroma = REQ_128BytesContiguous; |
5535 | } |
5536 | |
5537 | if (RequestLuma == REQ_256Bytes) { |
5538 | *MaxUncompressedBlockLuma = 256; |
5539 | *MaxCompressedBlockLuma = 256; |
5540 | *IndependentBlockLuma = 0; |
5541 | } else if (RequestLuma == REQ_128BytesContiguous) { |
5542 | *MaxUncompressedBlockLuma = 256; |
5543 | *MaxCompressedBlockLuma = 128; |
5544 | *IndependentBlockLuma = 128; |
5545 | } else { |
5546 | *MaxUncompressedBlockLuma = 256; |
5547 | *MaxCompressedBlockLuma = 64; |
5548 | *IndependentBlockLuma = 64; |
5549 | } |
5550 | |
5551 | if (RequestChroma == REQ_256Bytes) { |
5552 | *MaxUncompressedBlockChroma = 256; |
5553 | *MaxCompressedBlockChroma = 256; |
5554 | *IndependentBlockChroma = 0; |
5555 | } else if (RequestChroma == REQ_128BytesContiguous) { |
5556 | *MaxUncompressedBlockChroma = 256; |
5557 | *MaxCompressedBlockChroma = 128; |
5558 | *IndependentBlockChroma = 128; |
5559 | } else { |
5560 | *MaxUncompressedBlockChroma = 256; |
5561 | *MaxCompressedBlockChroma = 64; |
5562 | *IndependentBlockChroma = 64; |
5563 | } |
5564 | |
5565 | if (DCCEnabled != true || BytePerPixelC == 0) { |
5566 | *MaxUncompressedBlockChroma = 0; |
5567 | *MaxCompressedBlockChroma = 0; |
5568 | *IndependentBlockChroma = 0; |
5569 | } |
5570 | |
5571 | if (DCCEnabled != true) { |
5572 | *MaxUncompressedBlockLuma = 0; |
5573 | *MaxCompressedBlockLuma = 0; |
5574 | *IndependentBlockLuma = 0; |
5575 | } |
5576 | |
5577 | #ifdef __DML_VBA_DEBUG__ |
5578 | dml_print("DML::%s: MaxUncompressedBlockLuma = %d\n" , __func__, *MaxUncompressedBlockLuma); |
5579 | dml_print("DML::%s: MaxCompressedBlockLuma = %d\n" , __func__, *MaxCompressedBlockLuma); |
5580 | dml_print("DML::%s: IndependentBlockLuma = %d\n" , __func__, *IndependentBlockLuma); |
5581 | dml_print("DML::%s: MaxUncompressedBlockChroma = %d\n" , __func__, *MaxUncompressedBlockChroma); |
5582 | dml_print("DML::%s: MaxCompressedBlockChroma = %d\n" , __func__, *MaxCompressedBlockChroma); |
5583 | dml_print("DML::%s: IndependentBlockChroma = %d\n" , __func__, *IndependentBlockChroma); |
5584 | #endif |
5585 | |
5586 | } // CalculateDCCConfiguration |
5587 | |
5588 | void dml32_CalculateStutterEfficiency( |
5589 | unsigned int CompressedBufferSizeInkByte, |
5590 | enum dm_use_mall_for_pstate_change_mode UseMALLForPStateChange[], |
5591 | bool UnboundedRequestEnabled, |
5592 | unsigned int MetaFIFOSizeInKEntries, |
5593 | unsigned int ZeroSizeBufferEntries, |
5594 | unsigned int PixelChunkSizeInKByte, |
5595 | unsigned int NumberOfActiveSurfaces, |
5596 | unsigned int ROBBufferSizeInKByte, |
5597 | double TotalDataReadBandwidth, |
5598 | double DCFCLK, |
5599 | double ReturnBW, |
5600 | unsigned int CompbufReservedSpace64B, |
5601 | unsigned int CompbufReservedSpaceZs, |
5602 | double SRExitTime, |
5603 | double SRExitZ8Time, |
5604 | bool SynchronizeTimingsFinal, |
5605 | unsigned int BlendingAndTiming[], |
5606 | double StutterEnterPlusExitWatermark, |
5607 | double Z8StutterEnterPlusExitWatermark, |
5608 | bool ProgressiveToInterlaceUnitInOPP, |
5609 | bool Interlace[], |
5610 | double MinTTUVBlank[], |
5611 | unsigned int DPPPerSurface[], |
5612 | unsigned int DETBufferSizeY[], |
5613 | unsigned int BytePerPixelY[], |
5614 | double BytePerPixelDETY[], |
5615 | double SwathWidthY[], |
5616 | unsigned int SwathHeightY[], |
5617 | unsigned int SwathHeightC[], |
5618 | double NetDCCRateLuma[], |
5619 | double NetDCCRateChroma[], |
5620 | double DCCFractionOfZeroSizeRequestsLuma[], |
5621 | double DCCFractionOfZeroSizeRequestsChroma[], |
5622 | unsigned int HTotal[], |
5623 | unsigned int VTotal[], |
5624 | double PixelClock[], |
5625 | double VRatio[], |
5626 | enum dm_rotation_angle SourceRotation[], |
5627 | unsigned int BlockHeight256BytesY[], |
5628 | unsigned int BlockWidth256BytesY[], |
5629 | unsigned int BlockHeight256BytesC[], |
5630 | unsigned int BlockWidth256BytesC[], |
5631 | unsigned int DCCYMaxUncompressedBlock[], |
5632 | unsigned int DCCCMaxUncompressedBlock[], |
5633 | unsigned int VActive[], |
5634 | bool DCCEnable[], |
5635 | bool WritebackEnable[], |
5636 | double ReadBandwidthSurfaceLuma[], |
5637 | double ReadBandwidthSurfaceChroma[], |
5638 | double meta_row_bw[], |
5639 | double dpte_row_bw[], |
5640 | |
5641 | /* Output */ |
5642 | double *StutterEfficiencyNotIncludingVBlank, |
5643 | double *StutterEfficiency, |
5644 | unsigned int *NumberOfStutterBurstsPerFrame, |
5645 | double *Z8StutterEfficiencyNotIncludingVBlank, |
5646 | double *Z8StutterEfficiency, |
5647 | unsigned int *Z8NumberOfStutterBurstsPerFrame, |
5648 | double *StutterPeriod, |
5649 | bool *DCHUBBUB_ARB_CSTATE_MAX_CAP_MODE) |
5650 | { |
5651 | |
5652 | bool FoundCriticalSurface = false; |
5653 | unsigned int SwathSizeCriticalSurface = 0; |
5654 | unsigned int LastChunkOfSwathSize; |
5655 | unsigned int MissingPartOfLastSwathOfDETSize; |
5656 | double LastZ8StutterPeriod = 0.0; |
5657 | double LastStutterPeriod = 0.0; |
5658 | unsigned int TotalNumberOfActiveOTG = 0; |
5659 | double doublePixelClock; |
5660 | unsigned int doubleHTotal; |
5661 | unsigned int doubleVTotal; |
5662 | bool SameTiming = true; |
5663 | double DETBufferingTimeY; |
5664 | double SwathWidthYCriticalSurface = 0.0; |
5665 | double SwathHeightYCriticalSurface = 0.0; |
5666 | double VActiveTimeCriticalSurface = 0.0; |
5667 | double FrameTimeCriticalSurface = 0.0; |
5668 | unsigned int BytePerPixelYCriticalSurface = 0; |
5669 | double LinesToFinishSwathTransferStutterCriticalSurface = 0.0; |
5670 | unsigned int DETBufferSizeYCriticalSurface = 0; |
5671 | double MinTTUVBlankCriticalSurface = 0.0; |
5672 | unsigned int BlockWidth256BytesYCriticalSurface = 0; |
5673 | bool doublePlaneCriticalSurface = 0; |
5674 | bool doublePipeCriticalSurface = 0; |
5675 | double TotalCompressedReadBandwidth; |
5676 | double TotalRowReadBandwidth; |
5677 | double AverageDCCCompressionRate; |
5678 | double EffectiveCompressedBufferSize; |
5679 | double PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer; |
5680 | double StutterBurstTime; |
5681 | unsigned int TotalActiveWriteback; |
5682 | double LinesInDETY; |
5683 | double LinesInDETYRoundedDownToSwath; |
5684 | double MaximumEffectiveCompressionLuma; |
5685 | double MaximumEffectiveCompressionChroma; |
5686 | double TotalZeroSizeRequestReadBandwidth; |
5687 | double TotalZeroSizeCompressedReadBandwidth; |
5688 | double AverageDCCZeroSizeFraction; |
5689 | double AverageZeroSizeCompressionRate; |
5690 | unsigned int k; |
5691 | |
5692 | TotalZeroSizeRequestReadBandwidth = 0; |
5693 | TotalZeroSizeCompressedReadBandwidth = 0; |
5694 | TotalRowReadBandwidth = 0; |
5695 | TotalCompressedReadBandwidth = 0; |
5696 | |
5697 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5698 | if (UseMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) { |
5699 | if (DCCEnable[k] == true) { |
5700 | if ((IsVertical(Scan: SourceRotation[k]) && BlockWidth256BytesY[k] > SwathHeightY[k]) |
5701 | || (!IsVertical(Scan: SourceRotation[k]) |
5702 | && BlockHeight256BytesY[k] > SwathHeightY[k]) |
5703 | || DCCYMaxUncompressedBlock[k] < 256) { |
5704 | MaximumEffectiveCompressionLuma = 2; |
5705 | } else { |
5706 | MaximumEffectiveCompressionLuma = 4; |
5707 | } |
5708 | TotalCompressedReadBandwidth = TotalCompressedReadBandwidth |
5709 | + ReadBandwidthSurfaceLuma[k] |
5710 | / dml_min(a: NetDCCRateLuma[k], |
5711 | b: MaximumEffectiveCompressionLuma); |
5712 | #ifdef __DML_VBA_DEBUG__ |
5713 | dml_print("DML::%s: k=%0d, ReadBandwidthSurfaceLuma = %f\n" , |
5714 | __func__, k, ReadBandwidthSurfaceLuma[k]); |
5715 | dml_print("DML::%s: k=%0d, NetDCCRateLuma = %f\n" , |
5716 | __func__, k, NetDCCRateLuma[k]); |
5717 | dml_print("DML::%s: k=%0d, MaximumEffectiveCompressionLuma = %f\n" , |
5718 | __func__, k, MaximumEffectiveCompressionLuma); |
5719 | #endif |
5720 | TotalZeroSizeRequestReadBandwidth = TotalZeroSizeRequestReadBandwidth |
5721 | + ReadBandwidthSurfaceLuma[k] * DCCFractionOfZeroSizeRequestsLuma[k]; |
5722 | TotalZeroSizeCompressedReadBandwidth = TotalZeroSizeCompressedReadBandwidth |
5723 | + ReadBandwidthSurfaceLuma[k] * DCCFractionOfZeroSizeRequestsLuma[k] |
5724 | / MaximumEffectiveCompressionLuma; |
5725 | |
5726 | if (ReadBandwidthSurfaceChroma[k] > 0) { |
5727 | if ((IsVertical(Scan: SourceRotation[k]) && BlockWidth256BytesC[k] > SwathHeightC[k]) |
5728 | || (!IsVertical(Scan: SourceRotation[k]) |
5729 | && BlockHeight256BytesC[k] > SwathHeightC[k]) |
5730 | || DCCCMaxUncompressedBlock[k] < 256) { |
5731 | MaximumEffectiveCompressionChroma = 2; |
5732 | } else { |
5733 | MaximumEffectiveCompressionChroma = 4; |
5734 | } |
5735 | TotalCompressedReadBandwidth = |
5736 | TotalCompressedReadBandwidth |
5737 | + ReadBandwidthSurfaceChroma[k] |
5738 | / dml_min(a: NetDCCRateChroma[k], |
5739 | b: MaximumEffectiveCompressionChroma); |
5740 | #ifdef __DML_VBA_DEBUG__ |
5741 | dml_print("DML::%s: k=%0d, ReadBandwidthSurfaceChroma = %f\n" , |
5742 | __func__, k, ReadBandwidthSurfaceChroma[k]); |
5743 | dml_print("DML::%s: k=%0d, NetDCCRateChroma = %f\n" , |
5744 | __func__, k, NetDCCRateChroma[k]); |
5745 | dml_print("DML::%s: k=%0d, MaximumEffectiveCompressionChroma = %f\n" , |
5746 | __func__, k, MaximumEffectiveCompressionChroma); |
5747 | #endif |
5748 | TotalZeroSizeRequestReadBandwidth = TotalZeroSizeRequestReadBandwidth |
5749 | + ReadBandwidthSurfaceChroma[k] |
5750 | * DCCFractionOfZeroSizeRequestsChroma[k]; |
5751 | TotalZeroSizeCompressedReadBandwidth = TotalZeroSizeCompressedReadBandwidth |
5752 | + ReadBandwidthSurfaceChroma[k] |
5753 | * DCCFractionOfZeroSizeRequestsChroma[k] |
5754 | / MaximumEffectiveCompressionChroma; |
5755 | } |
5756 | } else { |
5757 | TotalCompressedReadBandwidth = TotalCompressedReadBandwidth |
5758 | + ReadBandwidthSurfaceLuma[k] + ReadBandwidthSurfaceChroma[k]; |
5759 | } |
5760 | TotalRowReadBandwidth = TotalRowReadBandwidth |
5761 | + DPPPerSurface[k] * (meta_row_bw[k] + dpte_row_bw[k]); |
5762 | } |
5763 | } |
5764 | |
5765 | AverageDCCCompressionRate = TotalDataReadBandwidth / TotalCompressedReadBandwidth; |
5766 | AverageDCCZeroSizeFraction = TotalZeroSizeRequestReadBandwidth / TotalDataReadBandwidth; |
5767 | |
5768 | #ifdef __DML_VBA_DEBUG__ |
5769 | dml_print("DML::%s: UnboundedRequestEnabled = %d\n" , __func__, UnboundedRequestEnabled); |
5770 | dml_print("DML::%s: TotalCompressedReadBandwidth = %f\n" , __func__, TotalCompressedReadBandwidth); |
5771 | dml_print("DML::%s: TotalZeroSizeRequestReadBandwidth = %f\n" , __func__, TotalZeroSizeRequestReadBandwidth); |
5772 | dml_print("DML::%s: TotalZeroSizeCompressedReadBandwidth = %f\n" , |
5773 | __func__, TotalZeroSizeCompressedReadBandwidth); |
5774 | dml_print("DML::%s: MaximumEffectiveCompressionLuma = %f\n" , __func__, MaximumEffectiveCompressionLuma); |
5775 | dml_print("DML::%s: MaximumEffectiveCompressionChroma = %f\n" , __func__, MaximumEffectiveCompressionChroma); |
5776 | dml_print("DML::%s: AverageDCCCompressionRate = %f\n" , __func__, AverageDCCCompressionRate); |
5777 | dml_print("DML::%s: AverageDCCZeroSizeFraction = %f\n" , __func__, AverageDCCZeroSizeFraction); |
5778 | dml_print("DML::%s: CompbufReservedSpace64B = %d\n" , __func__, CompbufReservedSpace64B); |
5779 | dml_print("DML::%s: CompbufReservedSpaceZs = %d\n" , __func__, CompbufReservedSpaceZs); |
5780 | dml_print("DML::%s: CompressedBufferSizeInkByte = %d\n" , __func__, CompressedBufferSizeInkByte); |
5781 | #endif |
5782 | if (AverageDCCZeroSizeFraction == 1) { |
5783 | AverageZeroSizeCompressionRate = TotalZeroSizeRequestReadBandwidth |
5784 | / TotalZeroSizeCompressedReadBandwidth; |
5785 | EffectiveCompressedBufferSize = (double) MetaFIFOSizeInKEntries * 1024 * 64 |
5786 | * AverageZeroSizeCompressionRate |
5787 | + ((double) ZeroSizeBufferEntries - CompbufReservedSpaceZs) * 64 |
5788 | * AverageZeroSizeCompressionRate; |
5789 | } else if (AverageDCCZeroSizeFraction > 0) { |
5790 | AverageZeroSizeCompressionRate = TotalZeroSizeRequestReadBandwidth |
5791 | / TotalZeroSizeCompressedReadBandwidth; |
5792 | EffectiveCompressedBufferSize = dml_min( |
5793 | a: (double) CompressedBufferSizeInkByte * 1024 * AverageDCCCompressionRate, |
5794 | b: (double) MetaFIFOSizeInKEntries * 1024 * 64 |
5795 | / (AverageDCCZeroSizeFraction / AverageZeroSizeCompressionRate |
5796 | + 1 / AverageDCCCompressionRate)) |
5797 | + dml_min(a: ((double) ROBBufferSizeInKByte * 1024 - CompbufReservedSpace64B * 64) |
5798 | * AverageDCCCompressionRate, |
5799 | b: ((double) ZeroSizeBufferEntries - CompbufReservedSpaceZs) * 64 |
5800 | / (AverageDCCZeroSizeFraction / AverageZeroSizeCompressionRate)); |
5801 | |
5802 | #ifdef __DML_VBA_DEBUG__ |
5803 | dml_print("DML::%s: min 1 = %f\n" , __func__, |
5804 | CompressedBufferSizeInkByte * 1024 * AverageDCCCompressionRate); |
5805 | dml_print("DML::%s: min 2 = %f\n" , __func__, MetaFIFOSizeInKEntries * 1024 * 64 / |
5806 | (AverageDCCZeroSizeFraction / AverageZeroSizeCompressionRate + 1 / |
5807 | AverageDCCCompressionRate)); |
5808 | dml_print("DML::%s: min 3 = %f\n" , __func__, (ROBBufferSizeInKByte * 1024 - |
5809 | CompbufReservedSpace64B * 64) * AverageDCCCompressionRate); |
5810 | dml_print("DML::%s: min 4 = %f\n" , __func__, (ZeroSizeBufferEntries - CompbufReservedSpaceZs) * 64 / |
5811 | (AverageDCCZeroSizeFraction / AverageZeroSizeCompressionRate)); |
5812 | #endif |
5813 | } else { |
5814 | EffectiveCompressedBufferSize = dml_min( |
5815 | a: (double) CompressedBufferSizeInkByte * 1024 * AverageDCCCompressionRate, |
5816 | b: (double) MetaFIFOSizeInKEntries * 1024 * 64 * AverageDCCCompressionRate) |
5817 | + ((double) ROBBufferSizeInKByte * 1024 - CompbufReservedSpace64B * 64) |
5818 | * AverageDCCCompressionRate; |
5819 | |
5820 | #ifdef __DML_VBA_DEBUG__ |
5821 | dml_print("DML::%s: min 1 = %f\n" , __func__, |
5822 | CompressedBufferSizeInkByte * 1024 * AverageDCCCompressionRate); |
5823 | dml_print("DML::%s: min 2 = %f\n" , __func__, |
5824 | MetaFIFOSizeInKEntries * 1024 * 64 * AverageDCCCompressionRate); |
5825 | #endif |
5826 | } |
5827 | |
5828 | #ifdef __DML_VBA_DEBUG__ |
5829 | dml_print("DML::%s: MetaFIFOSizeInKEntries = %d\n" , __func__, MetaFIFOSizeInKEntries); |
5830 | dml_print("DML::%s: AverageZeroSizeCompressionRate = %f\n" , __func__, AverageZeroSizeCompressionRate); |
5831 | dml_print("DML::%s: EffectiveCompressedBufferSize = %f\n" , __func__, EffectiveCompressedBufferSize); |
5832 | #endif |
5833 | |
5834 | *StutterPeriod = 0; |
5835 | |
5836 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5837 | if (UseMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) { |
5838 | LinesInDETY = ((double) DETBufferSizeY[k] |
5839 | + (UnboundedRequestEnabled == true ? EffectiveCompressedBufferSize : 0) |
5840 | * ReadBandwidthSurfaceLuma[k] / TotalDataReadBandwidth) |
5841 | / BytePerPixelDETY[k] / SwathWidthY[k]; |
5842 | LinesInDETYRoundedDownToSwath = dml_floor(a: LinesInDETY, granularity: SwathHeightY[k]); |
5843 | DETBufferingTimeY = LinesInDETYRoundedDownToSwath * ((double) HTotal[k] / PixelClock[k]) |
5844 | / VRatio[k]; |
5845 | #ifdef __DML_VBA_DEBUG__ |
5846 | dml_print("DML::%s: k=%0d, DETBufferSizeY = %d\n" , __func__, k, DETBufferSizeY[k]); |
5847 | dml_print("DML::%s: k=%0d, BytePerPixelDETY = %f\n" , __func__, k, BytePerPixelDETY[k]); |
5848 | dml_print("DML::%s: k=%0d, SwathWidthY = %d\n" , __func__, k, SwathWidthY[k]); |
5849 | dml_print("DML::%s: k=%0d, ReadBandwidthSurfaceLuma = %f\n" , |
5850 | __func__, k, ReadBandwidthSurfaceLuma[k]); |
5851 | dml_print("DML::%s: k=%0d, TotalDataReadBandwidth = %f\n" , __func__, k, TotalDataReadBandwidth); |
5852 | dml_print("DML::%s: k=%0d, LinesInDETY = %f\n" , __func__, k, LinesInDETY); |
5853 | dml_print("DML::%s: k=%0d, LinesInDETYRoundedDownToSwath = %f\n" , |
5854 | __func__, k, LinesInDETYRoundedDownToSwath); |
5855 | dml_print("DML::%s: k=%0d, HTotal = %d\n" , __func__, k, HTotal[k]); |
5856 | dml_print("DML::%s: k=%0d, PixelClock = %f\n" , __func__, k, PixelClock[k]); |
5857 | dml_print("DML::%s: k=%0d, VRatio = %f\n" , __func__, k, VRatio[k]); |
5858 | dml_print("DML::%s: k=%0d, DETBufferingTimeY = %f\n" , __func__, k, DETBufferingTimeY); |
5859 | dml_print("DML::%s: k=%0d, PixelClock = %f\n" , __func__, k, PixelClock[k]); |
5860 | #endif |
5861 | |
5862 | if (!FoundCriticalSurface || DETBufferingTimeY < *StutterPeriod) { |
5863 | bool isInterlaceTiming = Interlace[k] && !ProgressiveToInterlaceUnitInOPP; |
5864 | |
5865 | FoundCriticalSurface = true; |
5866 | *StutterPeriod = DETBufferingTimeY; |
5867 | FrameTimeCriticalSurface = ( |
5868 | isInterlaceTiming ? |
5869 | dml_floor(a: (double) VTotal[k] / 2.0, granularity: 1.0) : VTotal[k]) |
5870 | * (double) HTotal[k] / PixelClock[k]; |
5871 | VActiveTimeCriticalSurface = ( |
5872 | isInterlaceTiming ? |
5873 | dml_floor(a: (double) VActive[k] / 2.0, granularity: 1.0) : VActive[k]) |
5874 | * (double) HTotal[k] / PixelClock[k]; |
5875 | BytePerPixelYCriticalSurface = BytePerPixelY[k]; |
5876 | SwathWidthYCriticalSurface = SwathWidthY[k]; |
5877 | SwathHeightYCriticalSurface = SwathHeightY[k]; |
5878 | BlockWidth256BytesYCriticalSurface = BlockWidth256BytesY[k]; |
5879 | LinesToFinishSwathTransferStutterCriticalSurface = SwathHeightY[k] |
5880 | - (LinesInDETY - LinesInDETYRoundedDownToSwath); |
5881 | DETBufferSizeYCriticalSurface = DETBufferSizeY[k]; |
5882 | MinTTUVBlankCriticalSurface = MinTTUVBlank[k]; |
5883 | doublePlaneCriticalSurface = (ReadBandwidthSurfaceChroma[k] == 0); |
5884 | doublePipeCriticalSurface = (DPPPerSurface[k] == 1); |
5885 | |
5886 | #ifdef __DML_VBA_DEBUG__ |
5887 | dml_print("DML::%s: k=%0d, FoundCriticalSurface = %d\n" , |
5888 | __func__, k, FoundCriticalSurface); |
5889 | dml_print("DML::%s: k=%0d, StutterPeriod = %f\n" , |
5890 | __func__, k, *StutterPeriod); |
5891 | dml_print("DML::%s: k=%0d, MinTTUVBlankCriticalSurface = %f\n" , |
5892 | __func__, k, MinTTUVBlankCriticalSurface); |
5893 | dml_print("DML::%s: k=%0d, FrameTimeCriticalSurface = %f\n" , |
5894 | __func__, k, FrameTimeCriticalSurface); |
5895 | dml_print("DML::%s: k=%0d, VActiveTimeCriticalSurface = %f\n" , |
5896 | __func__, k, VActiveTimeCriticalSurface); |
5897 | dml_print("DML::%s: k=%0d, BytePerPixelYCriticalSurface = %d\n" , |
5898 | __func__, k, BytePerPixelYCriticalSurface); |
5899 | dml_print("DML::%s: k=%0d, SwathWidthYCriticalSurface = %f\n" , |
5900 | __func__, k, SwathWidthYCriticalSurface); |
5901 | dml_print("DML::%s: k=%0d, SwathHeightYCriticalSurface = %f\n" , |
5902 | __func__, k, SwathHeightYCriticalSurface); |
5903 | dml_print("DML::%s: k=%0d, BlockWidth256BytesYCriticalSurface = %d\n" , |
5904 | __func__, k, BlockWidth256BytesYCriticalSurface); |
5905 | dml_print("DML::%s: k=%0d, doublePlaneCriticalSurface = %d\n" , |
5906 | __func__, k, doublePlaneCriticalSurface); |
5907 | dml_print("DML::%s: k=%0d, doublePipeCriticalSurface = %d\n" , |
5908 | __func__, k, doublePipeCriticalSurface); |
5909 | dml_print("DML::%s: k=%0d, LinesToFinishSwathTransferStutterCriticalSurface = %f\n" , |
5910 | __func__, k, LinesToFinishSwathTransferStutterCriticalSurface); |
5911 | #endif |
5912 | } |
5913 | } |
5914 | } |
5915 | |
5916 | PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer = dml_min(a: *StutterPeriod * TotalDataReadBandwidth, |
5917 | b: EffectiveCompressedBufferSize); |
5918 | #ifdef __DML_VBA_DEBUG__ |
5919 | dml_print("DML::%s: ROBBufferSizeInKByte = %d\n" , __func__, ROBBufferSizeInKByte); |
5920 | dml_print("DML::%s: AverageDCCCompressionRate = %f\n" , __func__, AverageDCCCompressionRate); |
5921 | dml_print("DML::%s: StutterPeriod * TotalDataReadBandwidth = %f\n" , |
5922 | __func__, *StutterPeriod * TotalDataReadBandwidth); |
5923 | dml_print("DML::%s: EffectiveCompressedBufferSize = %f\n" , __func__, EffectiveCompressedBufferSize); |
5924 | dml_print("DML::%s: PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer = %f\n" , __func__, |
5925 | PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer); |
5926 | dml_print("DML::%s: ReturnBW = %f\n" , __func__, ReturnBW); |
5927 | dml_print("DML::%s: TotalDataReadBandwidth = %f\n" , __func__, TotalDataReadBandwidth); |
5928 | dml_print("DML::%s: TotalRowReadBandwidth = %f\n" , __func__, TotalRowReadBandwidth); |
5929 | dml_print("DML::%s: DCFCLK = %f\n" , __func__, DCFCLK); |
5930 | #endif |
5931 | |
5932 | StutterBurstTime = PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer / AverageDCCCompressionRate |
5933 | / ReturnBW |
5934 | + (*StutterPeriod * TotalDataReadBandwidth |
5935 | - PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer) / (DCFCLK * 64) |
5936 | + *StutterPeriod * TotalRowReadBandwidth / ReturnBW; |
5937 | #ifdef __DML_VBA_DEBUG__ |
5938 | dml_print("DML::%s: Part 1 = %f\n" , __func__, PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer / |
5939 | AverageDCCCompressionRate / ReturnBW); |
5940 | dml_print("DML::%s: StutterPeriod * TotalDataReadBandwidth = %f\n" , |
5941 | __func__, (*StutterPeriod * TotalDataReadBandwidth)); |
5942 | dml_print("DML::%s: Part 2 = %f\n" , __func__, (*StutterPeriod * TotalDataReadBandwidth - |
5943 | PartOfUncompressedPixelBurstThatFitsInROBAndCompressedBuffer) / (DCFCLK * 64)); |
5944 | dml_print("DML::%s: Part 3 = %f\n" , __func__, *StutterPeriod * TotalRowReadBandwidth / ReturnBW); |
5945 | dml_print("DML::%s: StutterBurstTime = %f\n" , __func__, StutterBurstTime); |
5946 | #endif |
5947 | StutterBurstTime = dml_max(a: StutterBurstTime, |
5948 | b: LinesToFinishSwathTransferStutterCriticalSurface * BytePerPixelYCriticalSurface |
5949 | * SwathWidthYCriticalSurface / ReturnBW); |
5950 | |
5951 | #ifdef __DML_VBA_DEBUG__ |
5952 | dml_print("DML::%s: Time to finish residue swath=%f\n" , |
5953 | __func__, |
5954 | LinesToFinishSwathTransferStutterCriticalSurface * |
5955 | BytePerPixelYCriticalSurface * SwathWidthYCriticalSurface / ReturnBW); |
5956 | #endif |
5957 | |
5958 | TotalActiveWriteback = 0; |
5959 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5960 | if (WritebackEnable[k]) |
5961 | TotalActiveWriteback = TotalActiveWriteback + 1; |
5962 | } |
5963 | |
5964 | if (TotalActiveWriteback == 0) { |
5965 | #ifdef __DML_VBA_DEBUG__ |
5966 | dml_print("DML::%s: SRExitTime = %f\n" , __func__, SRExitTime); |
5967 | dml_print("DML::%s: SRExitZ8Time = %f\n" , __func__, SRExitZ8Time); |
5968 | dml_print("DML::%s: StutterBurstTime = %f (final)\n" , __func__, StutterBurstTime); |
5969 | dml_print("DML::%s: StutterPeriod = %f\n" , __func__, *StutterPeriod); |
5970 | #endif |
5971 | *StutterEfficiencyNotIncludingVBlank = dml_max(a: 0., |
5972 | b: 1 - (SRExitTime + StutterBurstTime) / *StutterPeriod) * 100; |
5973 | *Z8StutterEfficiencyNotIncludingVBlank = dml_max(a: 0., |
5974 | b: 1 - (SRExitZ8Time + StutterBurstTime) / *StutterPeriod) * 100; |
5975 | *NumberOfStutterBurstsPerFrame = ( |
5976 | *StutterEfficiencyNotIncludingVBlank > 0 ? |
5977 | dml_ceil(a: VActiveTimeCriticalSurface / *StutterPeriod, granularity: 1) : 0); |
5978 | *Z8NumberOfStutterBurstsPerFrame = ( |
5979 | *Z8StutterEfficiencyNotIncludingVBlank > 0 ? |
5980 | dml_ceil(a: VActiveTimeCriticalSurface / *StutterPeriod, granularity: 1) : 0); |
5981 | } else { |
5982 | *StutterEfficiencyNotIncludingVBlank = 0.; |
5983 | *Z8StutterEfficiencyNotIncludingVBlank = 0.; |
5984 | *NumberOfStutterBurstsPerFrame = 0; |
5985 | *Z8NumberOfStutterBurstsPerFrame = 0; |
5986 | } |
5987 | #ifdef __DML_VBA_DEBUG__ |
5988 | dml_print("DML::%s: VActiveTimeCriticalSurface = %f\n" , __func__, VActiveTimeCriticalSurface); |
5989 | dml_print("DML::%s: StutterEfficiencyNotIncludingVBlank = %f\n" , |
5990 | __func__, *StutterEfficiencyNotIncludingVBlank); |
5991 | dml_print("DML::%s: Z8StutterEfficiencyNotIncludingVBlank = %f\n" , |
5992 | __func__, *Z8StutterEfficiencyNotIncludingVBlank); |
5993 | dml_print("DML::%s: NumberOfStutterBurstsPerFrame = %d\n" , __func__, *NumberOfStutterBurstsPerFrame); |
5994 | dml_print("DML::%s: Z8NumberOfStutterBurstsPerFrame = %d\n" , __func__, *Z8NumberOfStutterBurstsPerFrame); |
5995 | #endif |
5996 | |
5997 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
5998 | if (UseMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) { |
5999 | if (BlendingAndTiming[k] == k) { |
6000 | if (TotalNumberOfActiveOTG == 0) { |
6001 | doublePixelClock = PixelClock[k]; |
6002 | doubleHTotal = HTotal[k]; |
6003 | doubleVTotal = VTotal[k]; |
6004 | } else if (doublePixelClock != PixelClock[k] || doubleHTotal != HTotal[k] |
6005 | || doubleVTotal != VTotal[k]) { |
6006 | SameTiming = false; |
6007 | } |
6008 | TotalNumberOfActiveOTG = TotalNumberOfActiveOTG + 1; |
6009 | } |
6010 | } |
6011 | } |
6012 | |
6013 | if (*StutterEfficiencyNotIncludingVBlank > 0) { |
6014 | LastStutterPeriod = VActiveTimeCriticalSurface - (*NumberOfStutterBurstsPerFrame - 1) * *StutterPeriod; |
6015 | |
6016 | if ((SynchronizeTimingsFinal || TotalNumberOfActiveOTG == 1) && SameTiming |
6017 | && LastStutterPeriod + MinTTUVBlankCriticalSurface > StutterEnterPlusExitWatermark) { |
6018 | *StutterEfficiency = (1 - (*NumberOfStutterBurstsPerFrame * SRExitTime |
6019 | + StutterBurstTime * VActiveTimeCriticalSurface |
6020 | / *StutterPeriod) / FrameTimeCriticalSurface) * 100; |
6021 | } else { |
6022 | *StutterEfficiency = *StutterEfficiencyNotIncludingVBlank; |
6023 | } |
6024 | } else { |
6025 | *StutterEfficiency = 0; |
6026 | } |
6027 | |
6028 | if (*Z8StutterEfficiencyNotIncludingVBlank > 0) { |
6029 | LastZ8StutterPeriod = VActiveTimeCriticalSurface |
6030 | - (*NumberOfStutterBurstsPerFrame - 1) * *StutterPeriod; |
6031 | if ((SynchronizeTimingsFinal || TotalNumberOfActiveOTG == 1) && SameTiming && LastZ8StutterPeriod + |
6032 | MinTTUVBlankCriticalSurface > Z8StutterEnterPlusExitWatermark) { |
6033 | *Z8StutterEfficiency = (1 - (*NumberOfStutterBurstsPerFrame * SRExitZ8Time + StutterBurstTime |
6034 | * VActiveTimeCriticalSurface / *StutterPeriod) / FrameTimeCriticalSurface) * 100; |
6035 | } else { |
6036 | *Z8StutterEfficiency = *Z8StutterEfficiencyNotIncludingVBlank; |
6037 | } |
6038 | } else { |
6039 | *Z8StutterEfficiency = 0.; |
6040 | } |
6041 | |
6042 | #ifdef __DML_VBA_DEBUG__ |
6043 | dml_print("DML::%s: LastZ8StutterPeriod = %f\n" , __func__, LastZ8StutterPeriod); |
6044 | dml_print("DML::%s: Z8StutterEnterPlusExitWatermark = %f\n" , __func__, Z8StutterEnterPlusExitWatermark); |
6045 | dml_print("DML::%s: StutterBurstTime = %f\n" , __func__, StutterBurstTime); |
6046 | dml_print("DML::%s: StutterPeriod = %f\n" , __func__, *StutterPeriod); |
6047 | dml_print("DML::%s: StutterEfficiency = %f\n" , __func__, *StutterEfficiency); |
6048 | dml_print("DML::%s: Z8StutterEfficiency = %f\n" , __func__, *Z8StutterEfficiency); |
6049 | dml_print("DML::%s: StutterEfficiencyNotIncludingVBlank = %f\n" , |
6050 | __func__, *StutterEfficiencyNotIncludingVBlank); |
6051 | dml_print("DML::%s: Z8NumberOfStutterBurstsPerFrame = %d\n" , __func__, *Z8NumberOfStutterBurstsPerFrame); |
6052 | #endif |
6053 | |
6054 | SwathSizeCriticalSurface = BytePerPixelYCriticalSurface * SwathHeightYCriticalSurface |
6055 | * dml_ceil(a: SwathWidthYCriticalSurface, granularity: BlockWidth256BytesYCriticalSurface); |
6056 | LastChunkOfSwathSize = SwathSizeCriticalSurface % (PixelChunkSizeInKByte * 1024); |
6057 | MissingPartOfLastSwathOfDETSize = dml_ceil(a: DETBufferSizeYCriticalSurface, granularity: SwathSizeCriticalSurface) |
6058 | - DETBufferSizeYCriticalSurface; |
6059 | |
6060 | *DCHUBBUB_ARB_CSTATE_MAX_CAP_MODE = !(!UnboundedRequestEnabled && (NumberOfActiveSurfaces == 1) |
6061 | && doublePlaneCriticalSurface && doublePipeCriticalSurface && (LastChunkOfSwathSize > 0) |
6062 | && (LastChunkOfSwathSize <= 4096) && (MissingPartOfLastSwathOfDETSize > 0) |
6063 | && (MissingPartOfLastSwathOfDETSize <= LastChunkOfSwathSize)); |
6064 | |
6065 | #ifdef __DML_VBA_DEBUG__ |
6066 | dml_print("DML::%s: SwathSizeCriticalSurface = %d\n" , __func__, SwathSizeCriticalSurface); |
6067 | dml_print("DML::%s: LastChunkOfSwathSize = %d\n" , __func__, LastChunkOfSwathSize); |
6068 | dml_print("DML::%s: MissingPartOfLastSwathOfDETSize = %d\n" , __func__, MissingPartOfLastSwathOfDETSize); |
6069 | dml_print("DML::%s: DCHUBBUB_ARB_CSTATE_MAX_CAP_MODE = %d\n" , __func__, *DCHUBBUB_ARB_CSTATE_MAX_CAP_MODE); |
6070 | #endif |
6071 | } // CalculateStutterEfficiency |
6072 | |
6073 | void dml32_CalculateMaxDETAndMinCompressedBufferSize( |
6074 | unsigned int ConfigReturnBufferSizeInKByte, |
6075 | unsigned int ROBBufferSizeInKByte, |
6076 | unsigned int MaxNumDPP, |
6077 | bool nomDETInKByteOverrideEnable, // VBA_DELTA, allow DV to override default DET size |
6078 | unsigned int nomDETInKByteOverrideValue, // VBA_DELTA |
6079 | |
6080 | /* Output */ |
6081 | unsigned int *MaxTotalDETInKByte, |
6082 | unsigned int *nomDETInKByte, |
6083 | unsigned int *MinCompressedBufferSizeInKByte) |
6084 | { |
6085 | bool det_buff_size_override_en = nomDETInKByteOverrideEnable; |
6086 | unsigned int det_buff_size_override_val = nomDETInKByteOverrideValue; |
6087 | |
6088 | *MaxTotalDETInKByte = dml_ceil(a: ((double)ConfigReturnBufferSizeInKByte + |
6089 | (double) ROBBufferSizeInKByte) * 4.0 / 5.0, granularity: 64); |
6090 | *nomDETInKByte = dml_floor(a: (double) *MaxTotalDETInKByte / (double) MaxNumDPP, granularity: 64); |
6091 | *MinCompressedBufferSizeInKByte = ConfigReturnBufferSizeInKByte - *MaxTotalDETInKByte; |
6092 | |
6093 | #ifdef __DML_VBA_DEBUG__ |
6094 | dml_print("DML::%s: ConfigReturnBufferSizeInKByte = %0d\n" , __func__, ConfigReturnBufferSizeInKByte); |
6095 | dml_print("DML::%s: ROBBufferSizeInKByte = %0d\n" , __func__, ROBBufferSizeInKByte); |
6096 | dml_print("DML::%s: MaxNumDPP = %0d\n" , __func__, MaxNumDPP); |
6097 | dml_print("DML::%s: MaxTotalDETInKByte = %0d\n" , __func__, *MaxTotalDETInKByte); |
6098 | dml_print("DML::%s: nomDETInKByte = %0d\n" , __func__, *nomDETInKByte); |
6099 | dml_print("DML::%s: MinCompressedBufferSizeInKByte = %0d\n" , __func__, *MinCompressedBufferSizeInKByte); |
6100 | #endif |
6101 | |
6102 | if (det_buff_size_override_en) { |
6103 | *nomDETInKByte = det_buff_size_override_val; |
6104 | #ifdef __DML_VBA_DEBUG__ |
6105 | dml_print("DML::%s: nomDETInKByte = %0d (override)\n" , __func__, *nomDETInKByte); |
6106 | #endif |
6107 | } |
6108 | } // CalculateMaxDETAndMinCompressedBufferSize |
6109 | |
6110 | bool dml32_CalculateVActiveBandwithSupport(unsigned int NumberOfActiveSurfaces, |
6111 | double ReturnBW, |
6112 | bool NotUrgentLatencyHiding[], |
6113 | double ReadBandwidthLuma[], |
6114 | double ReadBandwidthChroma[], |
6115 | double cursor_bw[], |
6116 | double meta_row_bandwidth[], |
6117 | double dpte_row_bandwidth[], |
6118 | unsigned int NumberOfDPP[], |
6119 | double UrgentBurstFactorLuma[], |
6120 | double UrgentBurstFactorChroma[], |
6121 | double UrgentBurstFactorCursor[]) |
6122 | { |
6123 | unsigned int k; |
6124 | bool NotEnoughUrgentLatencyHiding = false; |
6125 | bool CalculateVActiveBandwithSupport_val = false; |
6126 | double VActiveBandwith = 0; |
6127 | |
6128 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6129 | if (NotUrgentLatencyHiding[k]) { |
6130 | NotEnoughUrgentLatencyHiding = true; |
6131 | } |
6132 | } |
6133 | |
6134 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6135 | VActiveBandwith = VActiveBandwith + ReadBandwidthLuma[k] * UrgentBurstFactorLuma[k] + ReadBandwidthChroma[k] * UrgentBurstFactorChroma[k] + cursor_bw[k] * UrgentBurstFactorCursor[k] + NumberOfDPP[k] * meta_row_bandwidth[k] + NumberOfDPP[k] * dpte_row_bandwidth[k]; |
6136 | } |
6137 | |
6138 | CalculateVActiveBandwithSupport_val = (VActiveBandwith <= ReturnBW) && !NotEnoughUrgentLatencyHiding; |
6139 | |
6140 | #ifdef __DML_VBA_DEBUG__ |
6141 | dml_print("DML::%s: NotEnoughUrgentLatencyHiding = %d\n" , __func__, NotEnoughUrgentLatencyHiding); |
6142 | dml_print("DML::%s: VActiveBandwith = %f\n" , __func__, VActiveBandwith); |
6143 | dml_print("DML::%s: ReturnBW = %f\n" , __func__, ReturnBW); |
6144 | dml_print("DML::%s: CalculateVActiveBandwithSupport_val = %d\n" , __func__, CalculateVActiveBandwithSupport_val); |
6145 | #endif |
6146 | return CalculateVActiveBandwithSupport_val; |
6147 | } |
6148 | |
6149 | void dml32_CalculatePrefetchBandwithSupport(unsigned int NumberOfActiveSurfaces, |
6150 | double ReturnBW, |
6151 | bool NotUrgentLatencyHiding[], |
6152 | double ReadBandwidthLuma[], |
6153 | double ReadBandwidthChroma[], |
6154 | double PrefetchBandwidthLuma[], |
6155 | double PrefetchBandwidthChroma[], |
6156 | double cursor_bw[], |
6157 | double meta_row_bandwidth[], |
6158 | double dpte_row_bandwidth[], |
6159 | double cursor_bw_pre[], |
6160 | double prefetch_vmrow_bw[], |
6161 | unsigned int NumberOfDPP[], |
6162 | double UrgentBurstFactorLuma[], |
6163 | double UrgentBurstFactorChroma[], |
6164 | double UrgentBurstFactorCursor[], |
6165 | double UrgentBurstFactorLumaPre[], |
6166 | double UrgentBurstFactorChromaPre[], |
6167 | double UrgentBurstFactorCursorPre[], |
6168 | double PrefetchBW[], |
6169 | double VRatio[], |
6170 | double MaxVRatioPre, |
6171 | |
6172 | /* output */ |
6173 | double *MaxPrefetchBandwidth, |
6174 | double *FractionOfUrgentBandwidth, |
6175 | bool *PrefetchBandwidthSupport) |
6176 | { |
6177 | unsigned int k; |
6178 | double ActiveBandwidthPerSurface; |
6179 | bool NotEnoughUrgentLatencyHiding = false; |
6180 | double TotalActiveBandwidth = 0; |
6181 | double TotalPrefetchBandwidth = 0; |
6182 | |
6183 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6184 | if (NotUrgentLatencyHiding[k]) { |
6185 | NotEnoughUrgentLatencyHiding = true; |
6186 | } |
6187 | } |
6188 | |
6189 | *MaxPrefetchBandwidth = 0; |
6190 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6191 | ActiveBandwidthPerSurface = ReadBandwidthLuma[k] * UrgentBurstFactorLuma[k] + ReadBandwidthChroma[k] * UrgentBurstFactorChroma[k] + cursor_bw[k] * UrgentBurstFactorCursor[k] + NumberOfDPP[k] * (meta_row_bandwidth[k] + dpte_row_bandwidth[k]); |
6192 | |
6193 | TotalActiveBandwidth += ActiveBandwidthPerSurface; |
6194 | |
6195 | TotalPrefetchBandwidth = TotalPrefetchBandwidth + PrefetchBW[k] * VRatio[k]; |
6196 | |
6197 | *MaxPrefetchBandwidth = *MaxPrefetchBandwidth + dml_max3(a: NumberOfDPP[k] * prefetch_vmrow_bw[k], |
6198 | b: ActiveBandwidthPerSurface, |
6199 | c: NumberOfDPP[k] * (PrefetchBandwidthLuma[k] * UrgentBurstFactorLumaPre[k] + PrefetchBandwidthChroma[k] * UrgentBurstFactorChromaPre[k]) + cursor_bw_pre[k] * UrgentBurstFactorCursorPre[k]); |
6200 | } |
6201 | |
6202 | if (MaxVRatioPre == __DML_MAX_VRATIO_PRE__) |
6203 | *PrefetchBandwidthSupport = (*MaxPrefetchBandwidth <= ReturnBW) && (TotalPrefetchBandwidth <= TotalActiveBandwidth * __DML_MAX_BW_RATIO_PRE__) && !NotEnoughUrgentLatencyHiding; |
6204 | else |
6205 | *PrefetchBandwidthSupport = (*MaxPrefetchBandwidth <= ReturnBW) && !NotEnoughUrgentLatencyHiding; |
6206 | |
6207 | *FractionOfUrgentBandwidth = *MaxPrefetchBandwidth / ReturnBW; |
6208 | } |
6209 | |
6210 | double dml32_CalculateBandwidthAvailableForImmediateFlip(unsigned int NumberOfActiveSurfaces, |
6211 | double ReturnBW, |
6212 | double ReadBandwidthLuma[], |
6213 | double ReadBandwidthChroma[], |
6214 | double PrefetchBandwidthLuma[], |
6215 | double PrefetchBandwidthChroma[], |
6216 | double cursor_bw[], |
6217 | double cursor_bw_pre[], |
6218 | unsigned int NumberOfDPP[], |
6219 | double UrgentBurstFactorLuma[], |
6220 | double UrgentBurstFactorChroma[], |
6221 | double UrgentBurstFactorCursor[], |
6222 | double UrgentBurstFactorLumaPre[], |
6223 | double UrgentBurstFactorChromaPre[], |
6224 | double UrgentBurstFactorCursorPre[]) |
6225 | { |
6226 | unsigned int k; |
6227 | double CalculateBandwidthAvailableForImmediateFlip_val = ReturnBW; |
6228 | |
6229 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6230 | CalculateBandwidthAvailableForImmediateFlip_val = CalculateBandwidthAvailableForImmediateFlip_val - dml_max(a: ReadBandwidthLuma[k] * UrgentBurstFactorLuma[k] + ReadBandwidthChroma[k] * UrgentBurstFactorChroma[k] + cursor_bw[k] * UrgentBurstFactorCursor[k], |
6231 | b: NumberOfDPP[k] * (PrefetchBandwidthLuma[k] * UrgentBurstFactorLumaPre[k] + PrefetchBandwidthChroma[k] * UrgentBurstFactorChromaPre[k]) + cursor_bw_pre[k] * UrgentBurstFactorCursorPre[k]); |
6232 | } |
6233 | |
6234 | return CalculateBandwidthAvailableForImmediateFlip_val; |
6235 | } |
6236 | |
6237 | void dml32_CalculateImmediateFlipBandwithSupport(unsigned int NumberOfActiveSurfaces, |
6238 | double ReturnBW, |
6239 | enum immediate_flip_requirement ImmediateFlipRequirement[], |
6240 | double final_flip_bw[], |
6241 | double ReadBandwidthLuma[], |
6242 | double ReadBandwidthChroma[], |
6243 | double PrefetchBandwidthLuma[], |
6244 | double PrefetchBandwidthChroma[], |
6245 | double cursor_bw[], |
6246 | double meta_row_bandwidth[], |
6247 | double dpte_row_bandwidth[], |
6248 | double cursor_bw_pre[], |
6249 | double prefetch_vmrow_bw[], |
6250 | unsigned int NumberOfDPP[], |
6251 | double UrgentBurstFactorLuma[], |
6252 | double UrgentBurstFactorChroma[], |
6253 | double UrgentBurstFactorCursor[], |
6254 | double UrgentBurstFactorLumaPre[], |
6255 | double UrgentBurstFactorChromaPre[], |
6256 | double UrgentBurstFactorCursorPre[], |
6257 | |
6258 | /* output */ |
6259 | double *TotalBandwidth, |
6260 | double *FractionOfUrgentBandwidth, |
6261 | bool *ImmediateFlipBandwidthSupport) |
6262 | { |
6263 | unsigned int k; |
6264 | *TotalBandwidth = 0; |
6265 | for (k = 0; k < NumberOfActiveSurfaces; ++k) { |
6266 | if (ImmediateFlipRequirement[k] != dm_immediate_flip_not_required) { |
6267 | *TotalBandwidth = *TotalBandwidth + dml_max3(a: NumberOfDPP[k] * prefetch_vmrow_bw[k], |
6268 | b: NumberOfDPP[k] * final_flip_bw[k] + ReadBandwidthLuma[k] * UrgentBurstFactorLuma[k] + ReadBandwidthChroma[k] * UrgentBurstFactorChroma[k] + cursor_bw[k] * UrgentBurstFactorCursor[k], |
6269 | c: NumberOfDPP[k] * (final_flip_bw[k] + PrefetchBandwidthLuma[k] * UrgentBurstFactorLumaPre[k] + PrefetchBandwidthChroma[k] * UrgentBurstFactorChromaPre[k]) + cursor_bw_pre[k] * UrgentBurstFactorCursorPre[k]); |
6270 | } else { |
6271 | *TotalBandwidth = *TotalBandwidth + dml_max3(a: NumberOfDPP[k] * prefetch_vmrow_bw[k], |
6272 | b: NumberOfDPP[k] * (meta_row_bandwidth[k] + dpte_row_bandwidth[k]) + ReadBandwidthLuma[k] * UrgentBurstFactorLuma[k] + ReadBandwidthChroma[k] * UrgentBurstFactorChroma[k] + cursor_bw[k] * UrgentBurstFactorCursor[k], |
6273 | c: NumberOfDPP[k] * (PrefetchBandwidthLuma[k] * UrgentBurstFactorLumaPre[k] + PrefetchBandwidthChroma[k] * UrgentBurstFactorChromaPre[k]) + cursor_bw_pre[k] * UrgentBurstFactorCursorPre[k]); |
6274 | } |
6275 | } |
6276 | *ImmediateFlipBandwidthSupport = (*TotalBandwidth <= ReturnBW); |
6277 | *FractionOfUrgentBandwidth = *TotalBandwidth / ReturnBW; |
6278 | } |
6279 | |
6280 | bool dml32_CalculateDETSwathFillLatencyHiding(unsigned int NumberOfActiveSurfaces, |
6281 | double ReturnBW, |
6282 | double UrgentLatency, |
6283 | unsigned int SwathHeightY[], |
6284 | unsigned int SwathHeightC[], |
6285 | unsigned int SwathWidthY[], |
6286 | unsigned int SwathWidthC[], |
6287 | double BytePerPixelInDETY[], |
6288 | double BytePerPixelInDETC[], |
6289 | unsigned int DETBufferSizeY[], |
6290 | unsigned int DETBufferSizeC[], |
6291 | unsigned int NumOfDPP[], |
6292 | unsigned int HTotal[], |
6293 | double PixelClock[], |
6294 | double VRatioY[], |
6295 | double VRatioC[], |
6296 | enum dm_use_mall_for_pstate_change_mode UsesMALLForPStateChange[], |
6297 | enum unbounded_requesting_policy UseUnboundedRequesting) |
6298 | { |
6299 | int k; |
6300 | double SwathSizeAllSurfaces = 0; |
6301 | double SwathSizeAllSurfacesInFetchTimeUs; |
6302 | double DETSwathLatencyHidingUs; |
6303 | double DETSwathLatencyHidingYUs; |
6304 | double DETSwathLatencyHidingCUs; |
6305 | double SwathSizePerSurfaceY[DC__NUM_DPP__MAX]; |
6306 | double SwathSizePerSurfaceC[DC__NUM_DPP__MAX]; |
6307 | bool NotEnoughDETSwathFillLatencyHiding = false; |
6308 | |
6309 | if (UseUnboundedRequesting == dm_unbounded_requesting) |
6310 | return false; |
6311 | |
6312 | /* calculate sum of single swath size for all pipes in bytes */ |
6313 | for (k = 0; k < NumberOfActiveSurfaces; k++) { |
6314 | SwathSizePerSurfaceY[k] = SwathHeightY[k] * SwathWidthY[k] * BytePerPixelInDETY[k] * NumOfDPP[k]; |
6315 | |
6316 | if (SwathHeightC[k] != 0) |
6317 | SwathSizePerSurfaceC[k] = SwathHeightC[k] * SwathWidthC[k] * BytePerPixelInDETC[k] * NumOfDPP[k]; |
6318 | else |
6319 | SwathSizePerSurfaceC[k] = 0; |
6320 | |
6321 | SwathSizeAllSurfaces += SwathSizePerSurfaceY[k] + SwathSizePerSurfaceC[k]; |
6322 | } |
6323 | |
6324 | SwathSizeAllSurfacesInFetchTimeUs = SwathSizeAllSurfaces / ReturnBW + UrgentLatency; |
6325 | |
6326 | /* ensure all DET - 1 swath can hide a fetch for all surfaces */ |
6327 | for (k = 0; k < NumberOfActiveSurfaces; k++) { |
6328 | double LineTime = HTotal[k] / PixelClock[k]; |
6329 | |
6330 | /* only care if surface is not phantom */ |
6331 | if (UsesMALLForPStateChange[k] != dm_use_mall_pstate_change_phantom_pipe) { |
6332 | DETSwathLatencyHidingYUs = (dml_floor(a: DETBufferSizeY[k] / BytePerPixelInDETY[k] / SwathWidthY[k], granularity: 1.0) - SwathHeightY[k]) / VRatioY[k] * LineTime; |
6333 | |
6334 | if (SwathHeightC[k] != 0) { |
6335 | DETSwathLatencyHidingCUs = (dml_floor(a: DETBufferSizeC[k] / BytePerPixelInDETC[k] / SwathWidthC[k], granularity: 1.0) - SwathHeightC[k]) / VRatioC[k] * LineTime; |
6336 | |
6337 | DETSwathLatencyHidingUs = dml_min(a: DETSwathLatencyHidingYUs, b: DETSwathLatencyHidingCUs); |
6338 | } else { |
6339 | DETSwathLatencyHidingUs = DETSwathLatencyHidingYUs; |
6340 | } |
6341 | |
6342 | /* DET must be able to hide time to fetch 1 swath for each surface */ |
6343 | if (DETSwathLatencyHidingUs < SwathSizeAllSurfacesInFetchTimeUs) { |
6344 | NotEnoughDETSwathFillLatencyHiding = true; |
6345 | break; |
6346 | } |
6347 | } |
6348 | } |
6349 | |
6350 | return NotEnoughDETSwathFillLatencyHiding; |
6351 | } |
6352 | |