2 * Copyright © 2007-2019 Advanced Micro Devices, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining
6 * a copy of this software and associated documentation files (the
7 * "Software"), to deal in the Software without restriction, including
8 * without limitation the rights to use, copy, modify, merge, publish,
9 * distribute, sub license, and/or sell copies of the Software, and to
10 * permit persons to whom the Software is furnished to do so, subject to
11 * the following conditions:
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
14 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
15 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
16 * NON-INFRINGEMENT. IN NO EVENT SHALL THE COPYRIGHT HOLDERS, AUTHORS
17 * AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
20 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 * The above copyright notice and this permission notice (including the
23 * next paragraph) shall be included in all copies or substantial portions
28 ************************************************************************************************************************
29 * @file gfx9addrlib.cpp
30 * @brief Contgfx9ns the implementation for the Gfx9Lib class.
31 ************************************************************************************************************************
34 #include "gfx9addrlib.h"
36 #include "gfx9_gb_reg.h"
38 #include "amdgpu_asic_addr.h"
40 #include "util/macros.h"
42 ////////////////////////////////////////////////////////////////////////////////////////////////////
43 ////////////////////////////////////////////////////////////////////////////////////////////////////
49 ************************************************************************************************************************
53 * Creates an Gfx9Lib object.
56 * Returns an Gfx9Lib object pointer.
57 ************************************************************************************************************************
59 Addr::Lib
* Gfx9HwlInit(const Client
* pClient
)
61 return V2::Gfx9Lib::CreateObj(pClient
);
67 ////////////////////////////////////////////////////////////////////////////////////////////////////
68 // Static Const Member
69 ////////////////////////////////////////////////////////////////////////////////////////////////////
71 const SwizzleModeFlags
Gfx9Lib::SwizzleModeTable
[ADDR_SW_MAX_TYPE
] =
72 {//Linear 256B 4KB 64KB Var Z Std Disp Rot XOR T RtOpt
73 {1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, // ADDR_SW_LINEAR
74 {0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0}, // ADDR_SW_256B_S
75 {0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0}, // ADDR_SW_256B_D
76 {0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0}, // ADDR_SW_256B_R
78 {0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0}, // ADDR_SW_4KB_Z
79 {0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0}, // ADDR_SW_4KB_S
80 {0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0}, // ADDR_SW_4KB_D
81 {0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0}, // ADDR_SW_4KB_R
83 {0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0}, // ADDR_SW_64KB_Z
84 {0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0}, // ADDR_SW_64KB_S
85 {0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0}, // ADDR_SW_64KB_D
86 {0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0}, // ADDR_SW_64KB_R
88 {0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0}, // ADDR_SW_VAR_Z
89 {0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0}, // ADDR_SW_VAR_S
90 {0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0}, // ADDR_SW_VAR_D
91 {0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0}, // ADDR_SW_VAR_R
93 {0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 1, 0}, // ADDR_SW_64KB_Z_T
94 {0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 1, 0}, // ADDR_SW_64KB_S_T
95 {0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 1, 0}, // ADDR_SW_64KB_D_T
96 {0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 1, 0}, // ADDR_SW_64KB_R_T
98 {0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0}, // ADDR_SW_4KB_Z_x
99 {0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0}, // ADDR_SW_4KB_S_x
100 {0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 0, 0}, // ADDR_SW_4KB_D_x
101 {0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0}, // ADDR_SW_4KB_R_x
103 {0, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0}, // ADDR_SW_64KB_Z_X
104 {0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0}, // ADDR_SW_64KB_S_X
105 {0, 0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 0}, // ADDR_SW_64KB_D_X
106 {0, 0, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0}, // ADDR_SW_64KB_R_X
108 {0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 0}, // ADDR_SW_VAR_Z_X
109 {0, 0, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0}, // ADDR_SW_VAR_S_X
110 {0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0}, // ADDR_SW_VAR_D_X
111 {0, 0, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0}, // ADDR_SW_VAR_R_X
112 {1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, // ADDR_SW_LINEAR_GENERAL
115 const UINT_32
Gfx9Lib::MipTailOffset256B
[] = {2048, 1024, 512, 256, 128, 64, 32, 16,
116 8, 6, 5, 4, 3, 2, 1, 0};
118 const Dim3d
Gfx9Lib::Block256_3dS
[] = {{16, 4, 4}, {8, 4, 4}, {4, 4, 4}, {2, 4, 4}, {1, 4, 4}};
120 const Dim3d
Gfx9Lib::Block256_3dZ
[] = {{8, 4, 8}, {4, 4, 8}, {4, 4, 4}, {4, 2, 4}, {2, 2, 4}};
123 ************************************************************************************************************************
129 ************************************************************************************************************************
131 Gfx9Lib::Gfx9Lib(const Client
* pClient
)
136 m_class
= AI_ADDRLIB
;
137 memset(&m_settings
, 0, sizeof(m_settings
));
138 memcpy(m_swizzleModeTable
, SwizzleModeTable
, sizeof(SwizzleModeTable
));
139 memset(m_cachedMetaEqKey
, 0, sizeof(m_cachedMetaEqKey
));
140 m_metaEqOverrideIndex
= 0;
144 ************************************************************************************************************************
149 ************************************************************************************************************************
156 ************************************************************************************************************************
157 * Gfx9Lib::HwlComputeHtileInfo
160 * Interface function stub of AddrComputeHtilenfo
164 ************************************************************************************************************************
166 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeHtileInfo(
167 const ADDR2_COMPUTE_HTILE_INFO_INPUT
* pIn
, ///< [in] input structure
168 ADDR2_COMPUTE_HTILE_INFO_OUTPUT
* pOut
///< [out] output structure
171 UINT_32 numPipeTotal
= GetPipeNumForMetaAddressing(pIn
->hTileFlags
.pipeAligned
,
174 UINT_32 numRbTotal
= pIn
->hTileFlags
.rbAligned
? m_se
* m_rbPerSe
: 1;
176 UINT_32 numCompressBlkPerMetaBlk
, numCompressBlkPerMetaBlkLog2
;
178 if ((numPipeTotal
== 1) && (numRbTotal
== 1))
180 numCompressBlkPerMetaBlkLog2
= 10;
184 if (m_settings
.applyAliasFix
)
186 numCompressBlkPerMetaBlkLog2
= m_seLog2
+ m_rbPerSeLog2
+ Max(10u, m_pipeInterleaveLog2
);
190 numCompressBlkPerMetaBlkLog2
= m_seLog2
+ m_rbPerSeLog2
+ 10;
194 numCompressBlkPerMetaBlk
= 1 << numCompressBlkPerMetaBlkLog2
;
196 Dim3d metaBlkDim
= {8, 8, 1};
197 UINT_32 totalAmpBits
= numCompressBlkPerMetaBlkLog2
;
198 UINT_32 widthAmp
= (pIn
->numMipLevels
> 1) ? (totalAmpBits
>> 1) : RoundHalf(totalAmpBits
);
199 UINT_32 heightAmp
= totalAmpBits
- widthAmp
;
200 metaBlkDim
.w
<<= widthAmp
;
201 metaBlkDim
.h
<<= heightAmp
;
204 Dim3d metaBlkDimDbg
= {8, 8, 1};
205 for (UINT_32 index
= 0; index
< numCompressBlkPerMetaBlkLog2
; index
++)
207 if ((metaBlkDimDbg
.h
< metaBlkDimDbg
.w
) ||
208 ((pIn
->numMipLevels
> 1) && (metaBlkDimDbg
.h
== metaBlkDimDbg
.w
)))
210 metaBlkDimDbg
.h
<<= 1;
214 metaBlkDimDbg
.w
<<= 1;
217 ADDR_ASSERT((metaBlkDimDbg
.w
== metaBlkDim
.w
) && (metaBlkDimDbg
.h
== metaBlkDim
.h
));
224 GetMetaMipInfo(pIn
->numMipLevels
, &metaBlkDim
, FALSE
, pOut
->pMipInfo
,
225 pIn
->unalignedWidth
, pIn
->unalignedHeight
, pIn
->numSlices
,
226 &numMetaBlkX
, &numMetaBlkY
, &numMetaBlkZ
);
228 const UINT_32 metaBlkSize
= numCompressBlkPerMetaBlk
<< 2;
229 UINT_32 align
= numPipeTotal
* numRbTotal
* m_pipeInterleaveBytes
;
231 if ((IsXor(pIn
->swizzleMode
) == FALSE
) && (numPipeTotal
> 2))
233 align
*= (numPipeTotal
>> 1);
236 align
= Max(align
, metaBlkSize
);
238 if (m_settings
.metaBaseAlignFix
)
240 align
= Max(align
, GetBlockSize(pIn
->swizzleMode
));
243 if (m_settings
.htileAlignFix
)
245 const INT_32 metaBlkSizeLog2
= numCompressBlkPerMetaBlkLog2
+ 2;
246 const INT_32 htileCachelineSizeLog2
= 11;
247 const INT_32 maxNumOfRbMaskBits
= 1 + Log2(numPipeTotal
) + Log2(numRbTotal
);
249 INT_32 rbMaskPadding
= Max(0, htileCachelineSizeLog2
- (metaBlkSizeLog2
- maxNumOfRbMaskBits
));
251 align
<<= rbMaskPadding
;
254 pOut
->pitch
= numMetaBlkX
* metaBlkDim
.w
;
255 pOut
->height
= numMetaBlkY
* metaBlkDim
.h
;
256 pOut
->sliceSize
= numMetaBlkX
* numMetaBlkY
* metaBlkSize
;
258 pOut
->metaBlkWidth
= metaBlkDim
.w
;
259 pOut
->metaBlkHeight
= metaBlkDim
.h
;
260 pOut
->metaBlkNumPerSlice
= numMetaBlkX
* numMetaBlkY
;
262 pOut
->baseAlign
= align
;
263 pOut
->htileBytes
= PowTwoAlign(pOut
->sliceSize
* numMetaBlkZ
, align
);
269 ************************************************************************************************************************
270 * Gfx9Lib::HwlComputeCmaskInfo
273 * Interface function stub of AddrComputeCmaskInfo
277 ************************************************************************************************************************
279 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeCmaskInfo(
280 const ADDR2_COMPUTE_CMASK_INFO_INPUT
* pIn
, ///< [in] input structure
281 ADDR2_COMPUTE_CMASK_INFO_OUTPUT
* pOut
///< [out] output structure
284 // TODO: Clarify with AddrLib team
285 // ADDR_ASSERT(pIn->resourceType == ADDR_RSRC_TEX_2D);
287 UINT_32 numPipeTotal
= GetPipeNumForMetaAddressing(pIn
->cMaskFlags
.pipeAligned
,
290 UINT_32 numRbTotal
= pIn
->cMaskFlags
.rbAligned
? m_se
* m_rbPerSe
: 1;
292 UINT_32 numCompressBlkPerMetaBlkLog2
, numCompressBlkPerMetaBlk
;
294 if ((numPipeTotal
== 1) && (numRbTotal
== 1))
296 numCompressBlkPerMetaBlkLog2
= 13;
300 if (m_settings
.applyAliasFix
)
302 numCompressBlkPerMetaBlkLog2
= m_seLog2
+ m_rbPerSeLog2
+ Max(10u, m_pipeInterleaveLog2
);
306 numCompressBlkPerMetaBlkLog2
= m_seLog2
+ m_rbPerSeLog2
+ 10;
309 numCompressBlkPerMetaBlkLog2
= Max(numCompressBlkPerMetaBlkLog2
, 13u);
312 numCompressBlkPerMetaBlk
= 1 << numCompressBlkPerMetaBlkLog2
;
314 Dim2d metaBlkDim
= {8, 8};
315 UINT_32 totalAmpBits
= numCompressBlkPerMetaBlkLog2
;
316 UINT_32 heightAmp
= totalAmpBits
>> 1;
317 UINT_32 widthAmp
= totalAmpBits
- heightAmp
;
318 metaBlkDim
.w
<<= widthAmp
;
319 metaBlkDim
.h
<<= heightAmp
;
322 Dim2d metaBlkDimDbg
= {8, 8};
323 for (UINT_32 index
= 0; index
< numCompressBlkPerMetaBlkLog2
; index
++)
325 if (metaBlkDimDbg
.h
< metaBlkDimDbg
.w
)
327 metaBlkDimDbg
.h
<<= 1;
331 metaBlkDimDbg
.w
<<= 1;
334 ADDR_ASSERT((metaBlkDimDbg
.w
== metaBlkDim
.w
) && (metaBlkDimDbg
.h
== metaBlkDim
.h
));
337 UINT_32 numMetaBlkX
= (pIn
->unalignedWidth
+ metaBlkDim
.w
- 1) / metaBlkDim
.w
;
338 UINT_32 numMetaBlkY
= (pIn
->unalignedHeight
+ metaBlkDim
.h
- 1) / metaBlkDim
.h
;
339 UINT_32 numMetaBlkZ
= Max(pIn
->numSlices
, 1u);
341 UINT_32 sizeAlign
= numPipeTotal
* numRbTotal
* m_pipeInterleaveBytes
;
343 if (m_settings
.metaBaseAlignFix
)
345 sizeAlign
= Max(sizeAlign
, GetBlockSize(pIn
->swizzleMode
));
348 pOut
->pitch
= numMetaBlkX
* metaBlkDim
.w
;
349 pOut
->height
= numMetaBlkY
* metaBlkDim
.h
;
350 pOut
->sliceSize
= (numMetaBlkX
* numMetaBlkY
* numCompressBlkPerMetaBlk
) >> 1;
351 pOut
->cmaskBytes
= PowTwoAlign(pOut
->sliceSize
* numMetaBlkZ
, sizeAlign
);
352 pOut
->baseAlign
= Max(numCompressBlkPerMetaBlk
>> 1, sizeAlign
);
354 pOut
->metaBlkWidth
= metaBlkDim
.w
;
355 pOut
->metaBlkHeight
= metaBlkDim
.h
;
357 pOut
->metaBlkNumPerSlice
= numMetaBlkX
* numMetaBlkY
;
363 ************************************************************************************************************************
364 * Gfx9Lib::GetMetaMipInfo
371 ************************************************************************************************************************
373 VOID
Gfx9Lib::GetMetaMipInfo(
374 UINT_32 numMipLevels
, ///< [in] number of mip levels
375 Dim3d
* pMetaBlkDim
, ///< [in] meta block dimension
376 BOOL_32 dataThick
, ///< [in] data surface is thick
377 ADDR2_META_MIP_INFO
* pInfo
, ///< [out] meta mip info
378 UINT_32 mip0Width
, ///< [in] mip0 width
379 UINT_32 mip0Height
, ///< [in] mip0 height
380 UINT_32 mip0Depth
, ///< [in] mip0 depth
381 UINT_32
* pNumMetaBlkX
, ///< [out] number of metablock X in mipchain
382 UINT_32
* pNumMetaBlkY
, ///< [out] number of metablock Y in mipchain
383 UINT_32
* pNumMetaBlkZ
) ///< [out] number of metablock Z in mipchain
386 UINT_32 numMetaBlkX
= (mip0Width
+ pMetaBlkDim
->w
- 1) / pMetaBlkDim
->w
;
387 UINT_32 numMetaBlkY
= (mip0Height
+ pMetaBlkDim
->h
- 1) / pMetaBlkDim
->h
;
388 UINT_32 numMetaBlkZ
= (mip0Depth
+ pMetaBlkDim
->d
- 1) / pMetaBlkDim
->d
;
389 UINT_32 tailWidth
= pMetaBlkDim
->w
;
390 UINT_32 tailHeight
= pMetaBlkDim
->h
>> 1;
391 UINT_32 tailDepth
= pMetaBlkDim
->d
;
392 BOOL_32 inTail
= FALSE
;
393 AddrMajorMode major
= ADDR_MAJOR_MAX_TYPE
;
395 if (numMipLevels
> 1)
397 if (dataThick
&& (numMetaBlkZ
> numMetaBlkX
) && (numMetaBlkZ
> numMetaBlkY
))
400 major
= ADDR_MAJOR_Z
;
402 else if (numMetaBlkX
>= numMetaBlkY
)
405 major
= ADDR_MAJOR_X
;
410 major
= ADDR_MAJOR_Y
;
413 inTail
= ((mip0Width
<= tailWidth
) &&
414 (mip0Height
<= tailHeight
) &&
415 ((dataThick
== FALSE
) || (mip0Depth
<= tailDepth
)));
423 if (major
== ADDR_MAJOR_Z
)
426 pMipDim
= &numMetaBlkY
;
427 pOrderDim
= &numMetaBlkZ
;
430 else if (major
== ADDR_MAJOR_X
)
433 pMipDim
= &numMetaBlkY
;
434 pOrderDim
= &numMetaBlkX
;
440 pMipDim
= &numMetaBlkX
;
441 pOrderDim
= &numMetaBlkY
;
445 if ((*pMipDim
< 3) && (*pOrderDim
> orderLimit
) && (numMipLevels
> 3))
451 *pMipDim
+= ((*pMipDim
/ 2) + (*pMipDim
& 1));
458 UINT_32 mipWidth
= mip0Width
;
459 UINT_32 mipHeight
= mip0Height
;
460 UINT_32 mipDepth
= mip0Depth
;
461 Dim3d mipCoord
= {0};
463 for (UINT_32 mip
= 0; mip
< numMipLevels
; mip
++)
467 GetMetaMiptailInfo(&pInfo
[mip
], mipCoord
, numMipLevels
- mip
,
473 mipWidth
= PowTwoAlign(mipWidth
, pMetaBlkDim
->w
);
474 mipHeight
= PowTwoAlign(mipHeight
, pMetaBlkDim
->h
);
475 mipDepth
= PowTwoAlign(mipDepth
, pMetaBlkDim
->d
);
477 pInfo
[mip
].inMiptail
= FALSE
;
478 pInfo
[mip
].startX
= mipCoord
.w
;
479 pInfo
[mip
].startY
= mipCoord
.h
;
480 pInfo
[mip
].startZ
= mipCoord
.d
;
481 pInfo
[mip
].width
= mipWidth
;
482 pInfo
[mip
].height
= mipHeight
;
483 pInfo
[mip
].depth
= dataThick
? mipDepth
: 1;
485 if ((mip
>= 3) || (mip
& 1))
490 mipCoord
.w
+= mipWidth
;
493 mipCoord
.h
+= mipHeight
;
496 mipCoord
.d
+= mipDepth
;
507 mipCoord
.h
+= mipHeight
;
510 mipCoord
.w
+= mipWidth
;
513 mipCoord
.h
+= mipHeight
;
520 mipWidth
= Max(mipWidth
>> 1, 1u);
521 mipHeight
= Max(mipHeight
>> 1, 1u);
522 mipDepth
= Max(mipDepth
>> 1, 1u);
524 inTail
= ((mipWidth
<= tailWidth
) &&
525 (mipHeight
<= tailHeight
) &&
526 ((dataThick
== FALSE
) || (mipDepth
<= tailDepth
)));
531 *pNumMetaBlkX
= numMetaBlkX
;
532 *pNumMetaBlkY
= numMetaBlkY
;
533 *pNumMetaBlkZ
= numMetaBlkZ
;
537 ************************************************************************************************************************
538 * Gfx9Lib::HwlComputeDccInfo
541 * Interface function to compute DCC key info
545 ************************************************************************************************************************
547 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeDccInfo(
548 const ADDR2_COMPUTE_DCCINFO_INPUT
* pIn
, ///< [in] input structure
549 ADDR2_COMPUTE_DCCINFO_OUTPUT
* pOut
///< [out] output structure
552 BOOL_32 dataLinear
= IsLinear(pIn
->swizzleMode
);
553 BOOL_32 metaLinear
= pIn
->dccKeyFlags
.linear
;
554 BOOL_32 pipeAligned
= pIn
->dccKeyFlags
.pipeAligned
;
560 else if (metaLinear
== TRUE
)
565 UINT_32 numPipeTotal
= GetPipeNumForMetaAddressing(pipeAligned
, pIn
->swizzleMode
);
569 // Linear metadata supporting was removed for GFX9! No one can use this feature on GFX9.
570 ADDR_ASSERT_ALWAYS();
572 pOut
->dccRamBaseAlign
= numPipeTotal
* m_pipeInterleaveBytes
;
573 pOut
->dccRamSize
= PowTwoAlign((pIn
->dataSurfaceSize
/ 256), pOut
->dccRamBaseAlign
);
577 BOOL_32 dataThick
= IsThick(pIn
->resourceType
, pIn
->swizzleMode
);
579 UINT_32 minMetaBlkSize
= dataThick
? 65536 : 4096;
581 UINT_32 numFrags
= Max(pIn
->numFrags
, 1u);
582 UINT_32 numSlices
= Max(pIn
->numSlices
, 1u);
584 minMetaBlkSize
/= numFrags
;
586 UINT_32 numCompressBlkPerMetaBlk
= minMetaBlkSize
;
588 UINT_32 numRbTotal
= pIn
->dccKeyFlags
.rbAligned
? m_se
* m_rbPerSe
: 1;
590 if ((numPipeTotal
> 1) || (numRbTotal
> 1))
592 const UINT_32 thinBlkSize
= 1 << (m_settings
.applyAliasFix
? Max(10u, m_pipeInterleaveLog2
) : 10);
594 numCompressBlkPerMetaBlk
=
595 Max(numCompressBlkPerMetaBlk
, m_se
* m_rbPerSe
* (dataThick
? 262144 : thinBlkSize
));
597 if (numCompressBlkPerMetaBlk
> 65536 * pIn
->bpp
)
599 numCompressBlkPerMetaBlk
= 65536 * pIn
->bpp
;
603 Dim3d compressBlkDim
= GetDccCompressBlk(pIn
->resourceType
, pIn
->swizzleMode
, pIn
->bpp
);
604 Dim3d metaBlkDim
= compressBlkDim
;
606 for (UINT_32 index
= 1; index
< numCompressBlkPerMetaBlk
; index
<<= 1)
608 if ((metaBlkDim
.h
< metaBlkDim
.w
) ||
609 ((pIn
->numMipLevels
> 1) && (metaBlkDim
.h
== metaBlkDim
.w
)))
611 if ((dataThick
== FALSE
) || (metaBlkDim
.h
<= metaBlkDim
.d
))
622 if ((dataThick
== FALSE
) || (metaBlkDim
.w
<= metaBlkDim
.d
))
637 GetMetaMipInfo(pIn
->numMipLevels
, &metaBlkDim
, dataThick
, pOut
->pMipInfo
,
638 pIn
->unalignedWidth
, pIn
->unalignedHeight
, numSlices
,
639 &numMetaBlkX
, &numMetaBlkY
, &numMetaBlkZ
);
641 UINT_32 sizeAlign
= numPipeTotal
* numRbTotal
* m_pipeInterleaveBytes
;
643 if (numFrags
> m_maxCompFrag
)
645 sizeAlign
*= (numFrags
/ m_maxCompFrag
);
648 if (m_settings
.metaBaseAlignFix
)
650 sizeAlign
= Max(sizeAlign
, GetBlockSize(pIn
->swizzleMode
));
653 pOut
->dccRamSize
= numMetaBlkX
* numMetaBlkY
* numMetaBlkZ
*
654 numCompressBlkPerMetaBlk
* numFrags
;
655 pOut
->dccRamSize
= PowTwoAlign(pOut
->dccRamSize
, sizeAlign
);
656 pOut
->dccRamBaseAlign
= Max(numCompressBlkPerMetaBlk
, sizeAlign
);
658 pOut
->pitch
= numMetaBlkX
* metaBlkDim
.w
;
659 pOut
->height
= numMetaBlkY
* metaBlkDim
.h
;
660 pOut
->depth
= numMetaBlkZ
* metaBlkDim
.d
;
662 pOut
->compressBlkWidth
= compressBlkDim
.w
;
663 pOut
->compressBlkHeight
= compressBlkDim
.h
;
664 pOut
->compressBlkDepth
= compressBlkDim
.d
;
666 pOut
->metaBlkWidth
= metaBlkDim
.w
;
667 pOut
->metaBlkHeight
= metaBlkDim
.h
;
668 pOut
->metaBlkDepth
= metaBlkDim
.d
;
670 pOut
->metaBlkNumPerSlice
= numMetaBlkX
* numMetaBlkY
;
671 pOut
->fastClearSizePerSlice
=
672 pOut
->metaBlkNumPerSlice
* numCompressBlkPerMetaBlk
* Min(numFrags
, m_maxCompFrag
);
679 ************************************************************************************************************************
680 * Gfx9Lib::HwlComputeMaxBaseAlignments
683 * Gets maximum alignments
686 ************************************************************************************************************************
688 UINT_32
Gfx9Lib::HwlComputeMaxBaseAlignments() const
690 return ComputeSurfaceBaseAlignTiled(ADDR_SW_64KB
);
694 ************************************************************************************************************************
695 * Gfx9Lib::HwlComputeMaxMetaBaseAlignments
698 * Gets maximum alignments for metadata
700 * maximum alignments for metadata
701 ************************************************************************************************************************
703 UINT_32
Gfx9Lib::HwlComputeMaxMetaBaseAlignments() const
705 // Max base alignment for Htile
706 const UINT_32 maxNumPipeTotal
= GetPipeNumForMetaAddressing(TRUE
, ADDR_SW_64KB_Z
);
707 const UINT_32 maxNumRbTotal
= m_se
* m_rbPerSe
;
709 // If applyAliasFix was set, the extra bits should be MAX(10u, m_pipeInterleaveLog2),
710 // but we never saw any ASIC whose m_pipeInterleaveLog2 != 8, so just put an assertion and simply the logic.
711 ADDR_ASSERT((m_settings
.applyAliasFix
== FALSE
) || (m_pipeInterleaveLog2
<= 10u));
712 const UINT_32 maxNumCompressBlkPerMetaBlk
= 1u << (m_seLog2
+ m_rbPerSeLog2
+ 10u);
714 UINT_32 maxBaseAlignHtile
= maxNumPipeTotal
* maxNumRbTotal
* m_pipeInterleaveBytes
;
716 if (maxNumPipeTotal
> 2)
718 maxBaseAlignHtile
*= (maxNumPipeTotal
>> 1);
721 maxBaseAlignHtile
= Max(maxNumCompressBlkPerMetaBlk
<< 2, maxBaseAlignHtile
);
723 if (m_settings
.metaBaseAlignFix
)
725 maxBaseAlignHtile
= Max(maxBaseAlignHtile
, GetBlockSize(ADDR_SW_64KB
));
728 if (m_settings
.htileAlignFix
)
730 maxBaseAlignHtile
*= maxNumPipeTotal
;
733 // Max base alignment for Cmask will not be larger than that for Htile, no need to calculate
735 // Max base alignment for 2D Dcc will not be larger than that for 3D, no need to calculate
736 UINT_32 maxBaseAlignDcc3D
= 65536;
738 if ((maxNumPipeTotal
> 1) || (maxNumRbTotal
> 1))
740 maxBaseAlignDcc3D
= Min(m_se
* m_rbPerSe
* 262144, 65536 * 128u);
743 // Max base alignment for Msaa Dcc
744 UINT_32 maxBaseAlignDccMsaa
= maxNumPipeTotal
* maxNumRbTotal
* m_pipeInterleaveBytes
* (8 / m_maxCompFrag
);
746 if (m_settings
.metaBaseAlignFix
)
748 maxBaseAlignDccMsaa
= Max(maxBaseAlignDccMsaa
, GetBlockSize(ADDR_SW_64KB
));
751 return Max(maxBaseAlignHtile
, Max(maxBaseAlignDccMsaa
, maxBaseAlignDcc3D
));
755 ************************************************************************************************************************
756 * Gfx9Lib::HwlComputeCmaskAddrFromCoord
759 * Interface function stub of AddrComputeCmaskAddrFromCoord
763 ************************************************************************************************************************
765 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeCmaskAddrFromCoord(
766 const ADDR2_COMPUTE_CMASK_ADDRFROMCOORD_INPUT
* pIn
, ///< [in] input structure
767 ADDR2_COMPUTE_CMASK_ADDRFROMCOORD_OUTPUT
* pOut
) ///< [out] output structure
769 ADDR2_COMPUTE_CMASK_INFO_INPUT input
= {0};
770 input
.size
= sizeof(input
);
771 input
.cMaskFlags
= pIn
->cMaskFlags
;
772 input
.colorFlags
= pIn
->colorFlags
;
773 input
.unalignedWidth
= Max(pIn
->unalignedWidth
, 1u);
774 input
.unalignedHeight
= Max(pIn
->unalignedHeight
, 1u);
775 input
.numSlices
= Max(pIn
->numSlices
, 1u);
776 input
.swizzleMode
= pIn
->swizzleMode
;
777 input
.resourceType
= pIn
->resourceType
;
779 ADDR2_COMPUTE_CMASK_INFO_OUTPUT output
= {0};
780 output
.size
= sizeof(output
);
782 ADDR_E_RETURNCODE returnCode
= ComputeCmaskInfo(&input
, &output
);
784 if (returnCode
== ADDR_OK
)
786 UINT_32 fmaskBpp
= GetFmaskBpp(pIn
->numSamples
, pIn
->numFrags
);
787 UINT_32 fmaskElementBytesLog2
= Log2(fmaskBpp
>> 3);
788 UINT_32 metaBlkWidthLog2
= Log2(output
.metaBlkWidth
);
789 UINT_32 metaBlkHeightLog2
= Log2(output
.metaBlkHeight
);
791 MetaEqParams metaEqParams
= {0, fmaskElementBytesLog2
, 0, pIn
->cMaskFlags
,
792 Gfx9DataFmask
, pIn
->swizzleMode
, pIn
->resourceType
,
793 metaBlkWidthLog2
, metaBlkHeightLog2
, 0, 3, 3, 0};
795 const CoordEq
* pMetaEq
= GetMetaEquation(metaEqParams
);
797 UINT_32 xb
= pIn
->x
/ output
.metaBlkWidth
;
798 UINT_32 yb
= pIn
->y
/ output
.metaBlkHeight
;
799 UINT_32 zb
= pIn
->slice
;
801 UINT_32 pitchInBlock
= output
.pitch
/ output
.metaBlkWidth
;
802 UINT_32 sliceSizeInBlock
= (output
.height
/ output
.metaBlkHeight
) * pitchInBlock
;
803 UINT_32 blockIndex
= zb
* sliceSizeInBlock
+ yb
* pitchInBlock
+ xb
;
805 UINT_64 address
= pMetaEq
->solve(pIn
->x
, pIn
->y
, pIn
->slice
, 0, blockIndex
);
807 pOut
->addr
= address
>> 1;
808 pOut
->bitPosition
= static_cast<UINT_32
>((address
& 1) << 2);
810 UINT_32 numPipeBits
= GetPipeLog2ForMetaAddressing(pIn
->cMaskFlags
.pipeAligned
,
813 UINT_64 pipeXor
= static_cast<UINT_64
>(pIn
->pipeXor
& ((1 << numPipeBits
) - 1));
815 pOut
->addr
^= (pipeXor
<< m_pipeInterleaveLog2
);
822 ************************************************************************************************************************
823 * Gfx9Lib::HwlComputeHtileAddrFromCoord
826 * Interface function stub of AddrComputeHtileAddrFromCoord
830 ************************************************************************************************************************
832 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeHtileAddrFromCoord(
833 const ADDR2_COMPUTE_HTILE_ADDRFROMCOORD_INPUT
* pIn
, ///< [in] input structure
834 ADDR2_COMPUTE_HTILE_ADDRFROMCOORD_OUTPUT
* pOut
) ///< [out] output structure
836 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
838 if (pIn
->numMipLevels
> 1)
840 returnCode
= ADDR_NOTIMPLEMENTED
;
844 ADDR2_COMPUTE_HTILE_INFO_INPUT input
= {0};
845 input
.size
= sizeof(input
);
846 input
.hTileFlags
= pIn
->hTileFlags
;
847 input
.depthFlags
= pIn
->depthflags
;
848 input
.swizzleMode
= pIn
->swizzleMode
;
849 input
.unalignedWidth
= Max(pIn
->unalignedWidth
, 1u);
850 input
.unalignedHeight
= Max(pIn
->unalignedHeight
, 1u);
851 input
.numSlices
= Max(pIn
->numSlices
, 1u);
852 input
.numMipLevels
= Max(pIn
->numMipLevels
, 1u);
854 ADDR2_COMPUTE_HTILE_INFO_OUTPUT output
= {0};
855 output
.size
= sizeof(output
);
857 returnCode
= ComputeHtileInfo(&input
, &output
);
859 if (returnCode
== ADDR_OK
)
861 UINT_32 elementBytesLog2
= Log2(pIn
->bpp
>> 3);
862 UINT_32 metaBlkWidthLog2
= Log2(output
.metaBlkWidth
);
863 UINT_32 metaBlkHeightLog2
= Log2(output
.metaBlkHeight
);
864 UINT_32 numSamplesLog2
= Log2(pIn
->numSamples
);
866 MetaEqParams metaEqParams
= {0, elementBytesLog2
, numSamplesLog2
, pIn
->hTileFlags
,
867 Gfx9DataDepthStencil
, pIn
->swizzleMode
, ADDR_RSRC_TEX_2D
,
868 metaBlkWidthLog2
, metaBlkHeightLog2
, 0, 3, 3, 0};
870 const CoordEq
* pMetaEq
= GetMetaEquation(metaEqParams
);
872 UINT_32 xb
= pIn
->x
/ output
.metaBlkWidth
;
873 UINT_32 yb
= pIn
->y
/ output
.metaBlkHeight
;
874 UINT_32 zb
= pIn
->slice
;
876 UINT_32 pitchInBlock
= output
.pitch
/ output
.metaBlkWidth
;
877 UINT_32 sliceSizeInBlock
= (output
.height
/ output
.metaBlkHeight
) * pitchInBlock
;
878 UINT_32 blockIndex
= zb
* sliceSizeInBlock
+ yb
* pitchInBlock
+ xb
;
880 UINT_64 address
= pMetaEq
->solve(pIn
->x
, pIn
->y
, pIn
->slice
, 0, blockIndex
);
882 pOut
->addr
= address
>> 1;
884 UINT_32 numPipeBits
= GetPipeLog2ForMetaAddressing(pIn
->hTileFlags
.pipeAligned
,
887 UINT_64 pipeXor
= static_cast<UINT_64
>(pIn
->pipeXor
& ((1 << numPipeBits
) - 1));
889 pOut
->addr
^= (pipeXor
<< m_pipeInterleaveLog2
);
897 ************************************************************************************************************************
898 * Gfx9Lib::HwlComputeHtileCoordFromAddr
901 * Interface function stub of AddrComputeHtileCoordFromAddr
905 ************************************************************************************************************************
907 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeHtileCoordFromAddr(
908 const ADDR2_COMPUTE_HTILE_COORDFROMADDR_INPUT
* pIn
, ///< [in] input structure
909 ADDR2_COMPUTE_HTILE_COORDFROMADDR_OUTPUT
* pOut
) ///< [out] output structure
911 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
913 if (pIn
->numMipLevels
> 1)
915 returnCode
= ADDR_NOTIMPLEMENTED
;
919 ADDR2_COMPUTE_HTILE_INFO_INPUT input
= {0};
920 input
.size
= sizeof(input
);
921 input
.hTileFlags
= pIn
->hTileFlags
;
922 input
.swizzleMode
= pIn
->swizzleMode
;
923 input
.unalignedWidth
= Max(pIn
->unalignedWidth
, 1u);
924 input
.unalignedHeight
= Max(pIn
->unalignedHeight
, 1u);
925 input
.numSlices
= Max(pIn
->numSlices
, 1u);
926 input
.numMipLevels
= Max(pIn
->numMipLevels
, 1u);
928 ADDR2_COMPUTE_HTILE_INFO_OUTPUT output
= {0};
929 output
.size
= sizeof(output
);
931 returnCode
= ComputeHtileInfo(&input
, &output
);
933 if (returnCode
== ADDR_OK
)
935 UINT_32 elementBytesLog2
= Log2(pIn
->bpp
>> 3);
936 UINT_32 metaBlkWidthLog2
= Log2(output
.metaBlkWidth
);
937 UINT_32 metaBlkHeightLog2
= Log2(output
.metaBlkHeight
);
938 UINT_32 numSamplesLog2
= Log2(pIn
->numSamples
);
940 MetaEqParams metaEqParams
= {0, elementBytesLog2
, numSamplesLog2
, pIn
->hTileFlags
,
941 Gfx9DataDepthStencil
, pIn
->swizzleMode
, ADDR_RSRC_TEX_2D
,
942 metaBlkWidthLog2
, metaBlkHeightLog2
, 0, 3, 3, 0};
944 const CoordEq
* pMetaEq
= GetMetaEquation(metaEqParams
);
946 UINT_32 numPipeBits
= GetPipeLog2ForMetaAddressing(pIn
->hTileFlags
.pipeAligned
,
949 UINT_64 pipeXor
= static_cast<UINT_64
>(pIn
->pipeXor
& ((1 << numPipeBits
) - 1));
951 UINT_64 nibbleAddress
= (pIn
->addr
^ (pipeXor
<< m_pipeInterleaveLog2
)) << 1;
953 UINT_32 pitchInBlock
= output
.pitch
/ output
.metaBlkWidth
;
954 UINT_32 sliceSizeInBlock
= (output
.height
/ output
.metaBlkHeight
) * pitchInBlock
;
956 UINT_32 x
, y
, z
, s
, m
;
957 pMetaEq
->solveAddr(nibbleAddress
, sliceSizeInBlock
, x
, y
, z
, s
, m
);
959 pOut
->slice
= m
/ sliceSizeInBlock
;
960 pOut
->y
= ((m
% sliceSizeInBlock
) / pitchInBlock
) * output
.metaBlkHeight
+ y
;
961 pOut
->x
= (m
% pitchInBlock
) * output
.metaBlkWidth
+ x
;
969 ************************************************************************************************************************
970 * Gfx9Lib::HwlComputeDccAddrFromCoord
973 * Interface function stub of AddrComputeDccAddrFromCoord
977 ************************************************************************************************************************
979 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeDccAddrFromCoord(
980 const ADDR2_COMPUTE_DCC_ADDRFROMCOORD_INPUT
* pIn
,
981 ADDR2_COMPUTE_DCC_ADDRFROMCOORD_OUTPUT
* pOut
)
983 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
985 if ((pIn
->numMipLevels
> 1) || (pIn
->mipId
> 1) || pIn
->dccKeyFlags
.linear
)
987 returnCode
= ADDR_NOTIMPLEMENTED
;
991 ADDR2_COMPUTE_DCCINFO_INPUT input
= {0};
992 input
.size
= sizeof(input
);
993 input
.dccKeyFlags
= pIn
->dccKeyFlags
;
994 input
.colorFlags
= pIn
->colorFlags
;
995 input
.swizzleMode
= pIn
->swizzleMode
;
996 input
.resourceType
= pIn
->resourceType
;
997 input
.bpp
= pIn
->bpp
;
998 input
.unalignedWidth
= Max(pIn
->unalignedWidth
, 1u);
999 input
.unalignedHeight
= Max(pIn
->unalignedHeight
, 1u);
1000 input
.numSlices
= Max(pIn
->numSlices
, 1u);
1001 input
.numFrags
= Max(pIn
->numFrags
, 1u);
1002 input
.numMipLevels
= Max(pIn
->numMipLevels
, 1u);
1004 ADDR2_COMPUTE_DCCINFO_OUTPUT output
= {0};
1005 output
.size
= sizeof(output
);
1007 returnCode
= ComputeDccInfo(&input
, &output
);
1009 if (returnCode
== ADDR_OK
)
1011 UINT_32 elementBytesLog2
= Log2(pIn
->bpp
>> 3);
1012 UINT_32 numSamplesLog2
= Log2(pIn
->numFrags
);
1013 UINT_32 metaBlkWidthLog2
= Log2(output
.metaBlkWidth
);
1014 UINT_32 metaBlkHeightLog2
= Log2(output
.metaBlkHeight
);
1015 UINT_32 metaBlkDepthLog2
= Log2(output
.metaBlkDepth
);
1016 UINT_32 compBlkWidthLog2
= Log2(output
.compressBlkWidth
);
1017 UINT_32 compBlkHeightLog2
= Log2(output
.compressBlkHeight
);
1018 UINT_32 compBlkDepthLog2
= Log2(output
.compressBlkDepth
);
1020 MetaEqParams metaEqParams
= {pIn
->mipId
, elementBytesLog2
, numSamplesLog2
, pIn
->dccKeyFlags
,
1021 Gfx9DataColor
, pIn
->swizzleMode
, pIn
->resourceType
,
1022 metaBlkWidthLog2
, metaBlkHeightLog2
, metaBlkDepthLog2
,
1023 compBlkWidthLog2
, compBlkHeightLog2
, compBlkDepthLog2
};
1025 const CoordEq
* pMetaEq
= GetMetaEquation(metaEqParams
);
1027 UINT_32 xb
= pIn
->x
/ output
.metaBlkWidth
;
1028 UINT_32 yb
= pIn
->y
/ output
.metaBlkHeight
;
1029 UINT_32 zb
= pIn
->slice
/ output
.metaBlkDepth
;
1031 UINT_32 pitchInBlock
= output
.pitch
/ output
.metaBlkWidth
;
1032 UINT_32 sliceSizeInBlock
= (output
.height
/ output
.metaBlkHeight
) * pitchInBlock
;
1033 UINT_32 blockIndex
= zb
* sliceSizeInBlock
+ yb
* pitchInBlock
+ xb
;
1035 UINT_64 address
= pMetaEq
->solve(pIn
->x
, pIn
->y
, pIn
->slice
, pIn
->sample
, blockIndex
);
1037 pOut
->addr
= address
>> 1;
1039 UINT_32 numPipeBits
= GetPipeLog2ForMetaAddressing(pIn
->dccKeyFlags
.pipeAligned
,
1042 UINT_64 pipeXor
= static_cast<UINT_64
>(pIn
->pipeXor
& ((1 << numPipeBits
) - 1));
1044 pOut
->addr
^= (pipeXor
<< m_pipeInterleaveLog2
);
1052 ************************************************************************************************************************
1053 * Gfx9Lib::HwlInitGlobalParams
1056 * Initializes global parameters
1059 * TRUE if all settings are valid
1061 ************************************************************************************************************************
1063 BOOL_32
Gfx9Lib::HwlInitGlobalParams(
1064 const ADDR_CREATE_INPUT
* pCreateIn
) ///< [in] create input
1066 BOOL_32 valid
= TRUE
;
1068 if (m_settings
.isArcticIsland
)
1070 GB_ADDR_CONFIG gbAddrConfig
;
1072 gbAddrConfig
.u32All
= pCreateIn
->regValue
.gbAddrConfig
;
1074 // These values are copied from CModel code
1075 switch (gbAddrConfig
.bits
.NUM_PIPES
)
1077 case ADDR_CONFIG_1_PIPE
:
1081 case ADDR_CONFIG_2_PIPE
:
1085 case ADDR_CONFIG_4_PIPE
:
1089 case ADDR_CONFIG_8_PIPE
:
1093 case ADDR_CONFIG_16_PIPE
:
1097 case ADDR_CONFIG_32_PIPE
:
1102 ADDR_ASSERT_ALWAYS();
1106 switch (gbAddrConfig
.bits
.PIPE_INTERLEAVE_SIZE
)
1108 case ADDR_CONFIG_PIPE_INTERLEAVE_256B
:
1109 m_pipeInterleaveBytes
= ADDR_PIPEINTERLEAVE_256B
;
1110 m_pipeInterleaveLog2
= 8;
1112 case ADDR_CONFIG_PIPE_INTERLEAVE_512B
:
1113 m_pipeInterleaveBytes
= ADDR_PIPEINTERLEAVE_512B
;
1114 m_pipeInterleaveLog2
= 9;
1116 case ADDR_CONFIG_PIPE_INTERLEAVE_1KB
:
1117 m_pipeInterleaveBytes
= ADDR_PIPEINTERLEAVE_1KB
;
1118 m_pipeInterleaveLog2
= 10;
1120 case ADDR_CONFIG_PIPE_INTERLEAVE_2KB
:
1121 m_pipeInterleaveBytes
= ADDR_PIPEINTERLEAVE_2KB
;
1122 m_pipeInterleaveLog2
= 11;
1125 ADDR_ASSERT_ALWAYS();
1129 // Addr::V2::Lib::ComputePipeBankXor()/ComputeSlicePipeBankXor() requires pipe interleave to be exactly 8 bits,
1130 // and any larger value requires a post-process (left shift) on the output pipeBankXor bits.
1131 ADDR_ASSERT(m_pipeInterleaveBytes
== ADDR_PIPEINTERLEAVE_256B
);
1133 switch (gbAddrConfig
.bits
.NUM_BANKS
)
1135 case ADDR_CONFIG_1_BANK
:
1139 case ADDR_CONFIG_2_BANK
:
1143 case ADDR_CONFIG_4_BANK
:
1147 case ADDR_CONFIG_8_BANK
:
1151 case ADDR_CONFIG_16_BANK
:
1156 ADDR_ASSERT_ALWAYS();
1160 switch (gbAddrConfig
.bits
.NUM_SHADER_ENGINES
)
1162 case ADDR_CONFIG_1_SHADER_ENGINE
:
1166 case ADDR_CONFIG_2_SHADER_ENGINE
:
1170 case ADDR_CONFIG_4_SHADER_ENGINE
:
1174 case ADDR_CONFIG_8_SHADER_ENGINE
:
1179 ADDR_ASSERT_ALWAYS();
1183 switch (gbAddrConfig
.bits
.NUM_RB_PER_SE
)
1185 case ADDR_CONFIG_1_RB_PER_SHADER_ENGINE
:
1189 case ADDR_CONFIG_2_RB_PER_SHADER_ENGINE
:
1193 case ADDR_CONFIG_4_RB_PER_SHADER_ENGINE
:
1198 ADDR_ASSERT_ALWAYS();
1202 switch (gbAddrConfig
.bits
.MAX_COMPRESSED_FRAGS
)
1204 case ADDR_CONFIG_1_MAX_COMPRESSED_FRAGMENTS
:
1206 m_maxCompFragLog2
= 0;
1208 case ADDR_CONFIG_2_MAX_COMPRESSED_FRAGMENTS
:
1210 m_maxCompFragLog2
= 1;
1212 case ADDR_CONFIG_4_MAX_COMPRESSED_FRAGMENTS
:
1214 m_maxCompFragLog2
= 2;
1216 case ADDR_CONFIG_8_MAX_COMPRESSED_FRAGMENTS
:
1218 m_maxCompFragLog2
= 3;
1221 ADDR_ASSERT_ALWAYS();
1225 m_blockVarSizeLog2
= pCreateIn
->regValue
.blockVarSizeLog2
;
1226 ADDR_ASSERT((m_blockVarSizeLog2
== 0) ||
1227 ((m_blockVarSizeLog2
>= 17u) && (m_blockVarSizeLog2
<= 20u)));
1228 m_blockVarSizeLog2
= Min(Max(17u, m_blockVarSizeLog2
), 20u);
1230 if ((m_rbPerSeLog2
== 1) &&
1231 (((m_pipesLog2
== 1) && ((m_seLog2
== 2) || (m_seLog2
== 3))) ||
1232 ((m_pipesLog2
== 2) && ((m_seLog2
== 1) || (m_seLog2
== 2)))))
1234 ADDR_ASSERT(m_settings
.isVega10
== FALSE
);
1235 ADDR_ASSERT(m_settings
.isRaven
== FALSE
);
1237 ADDR_ASSERT(m_settings
.isVega20
== FALSE
);
1239 if (m_settings
.isVega12
)
1241 m_settings
.htileCacheRbConflict
= 1;
1248 ADDR_NOT_IMPLEMENTED();
1253 InitEquationTable();
1260 ************************************************************************************************************************
1261 * Gfx9Lib::HwlConvertChipFamily
1264 * Convert familyID defined in atiid.h to ChipFamily and set m_chipFamily/m_chipRevision
1267 ************************************************************************************************************************
1269 ChipFamily
Gfx9Lib::HwlConvertChipFamily(
1270 UINT_32 uChipFamily
, ///< [in] chip family defined in atiih.h
1271 UINT_32 uChipRevision
) ///< [in] chip revision defined in "asic_family"_id.h
1273 ChipFamily family
= ADDR_CHIP_FAMILY_AI
;
1275 switch (uChipFamily
)
1278 m_settings
.isArcticIsland
= 1;
1279 m_settings
.isVega10
= ASICREV_IS_VEGA10_P(uChipRevision
);
1280 m_settings
.isVega12
= ASICREV_IS_VEGA12_P(uChipRevision
);
1281 m_settings
.isVega20
= ASICREV_IS_VEGA20_P(uChipRevision
);
1282 m_settings
.isDce12
= 1;
1284 if (m_settings
.isVega10
== 0)
1286 m_settings
.htileAlignFix
= 1;
1287 m_settings
.applyAliasFix
= 1;
1290 m_settings
.metaBaseAlignFix
= 1;
1292 m_settings
.depthPipeXorDisable
= 1;
1295 m_settings
.isArcticIsland
= 1;
1297 if (ASICREV_IS_RAVEN(uChipRevision
))
1299 m_settings
.isRaven
= 1;
1301 m_settings
.depthPipeXorDisable
= 1;
1304 if (ASICREV_IS_RAVEN2(uChipRevision
))
1306 m_settings
.isRaven
= 1;
1309 if (m_settings
.isRaven
== 0)
1311 m_settings
.htileAlignFix
= 1;
1312 m_settings
.applyAliasFix
= 1;
1315 if (ASICREV_IS_RENOIR(uChipRevision
))
1317 m_settings
.isRaven
= 1;
1320 m_settings
.isDcn1
= m_settings
.isRaven
;
1322 m_settings
.metaBaseAlignFix
= 1;
1326 ADDR_ASSERT(!"This should be a Fusion");
1334 ************************************************************************************************************************
1335 * Gfx9Lib::InitRbEquation
1341 ************************************************************************************************************************
1343 VOID
Gfx9Lib::GetRbEquation(
1344 CoordEq
* pRbEq
, ///< [out] rb equation
1345 UINT_32 numRbPerSeLog2
, ///< [in] number of rb per shader engine
1346 UINT_32 numSeLog2
) ///< [in] number of shader engine
1349 // RB's are distributed on 16x16, except when we have 1 rb per se, in which case its 32x32
1350 UINT_32 rbRegion
= (numRbPerSeLog2
== 0) ? 5 : 4;
1351 Coordinate
cx('x', rbRegion
);
1352 Coordinate
cy('y', rbRegion
);
1355 UINT_32 numRbTotalLog2
= numRbPerSeLog2
+ numSeLog2
;
1357 // Clear the rb equation
1359 pRbEq
->resize(numRbTotalLog2
);
1361 if ((numSeLog2
> 0) && (numRbPerSeLog2
== 1))
1363 // Special case when more than 1 SE, and 2 RB per SE
1364 (*pRbEq
)[0].add(cx
);
1365 (*pRbEq
)[0].add(cy
);
1369 if (m_settings
.applyAliasFix
== false)
1371 (*pRbEq
)[0].add(cy
);
1374 (*pRbEq
)[0].add(cy
);
1378 UINT_32 numBits
= 2 * (numRbTotalLog2
- start
);
1380 for (UINT_32 i
= 0; i
< numBits
; i
++)
1383 start
+ (((start
+ i
) >= numRbTotalLog2
) ? (2 * (numRbTotalLog2
- start
) - i
- 1) : i
);
1387 (*pRbEq
)[idx
].add(cx
);
1392 (*pRbEq
)[idx
].add(cy
);
1399 ************************************************************************************************************************
1400 * Gfx9Lib::GetDataEquation
1403 * Get data equation for fmask and Z
1406 ************************************************************************************************************************
1408 VOID
Gfx9Lib::GetDataEquation(
1409 CoordEq
* pDataEq
, ///< [out] data surface equation
1410 Gfx9DataType dataSurfaceType
, ///< [in] data surface type
1411 AddrSwizzleMode swizzleMode
, ///< [in] data surface swizzle mode
1412 AddrResourceType resourceType
, ///< [in] data surface resource type
1413 UINT_32 elementBytesLog2
, ///< [in] data surface element bytes
1414 UINT_32 numSamplesLog2
) ///< [in] data surface sample count
1417 Coordinate
cx('x', 0);
1418 Coordinate
cy('y', 0);
1419 Coordinate
cz('z', 0);
1420 Coordinate
cs('s', 0);
1422 // Clear the equation
1424 pDataEq
->resize(27);
1426 if (dataSurfaceType
== Gfx9DataColor
)
1428 if (IsLinear(swizzleMode
))
1430 Coordinate
cm('m', 0);
1432 pDataEq
->resize(49);
1434 for (UINT_32 i
= 0; i
< 49; i
++)
1436 (*pDataEq
)[i
].add(cm
);
1440 else if (IsThick(resourceType
, swizzleMode
))
1442 // Color 3d_S and 3d_Z modes, 3d_D is same as color 2d
1444 if (IsStandardSwizzle(resourceType
, swizzleMode
))
1446 // Standard 3d swizzle
1447 // Fill in bottom x bits
1448 for (i
= elementBytesLog2
; i
< 4; i
++)
1450 (*pDataEq
)[i
].add(cx
);
1453 // Fill in 2 bits of y and then z
1454 for (i
= 4; i
< 6; i
++)
1456 (*pDataEq
)[i
].add(cy
);
1459 for (i
= 6; i
< 8; i
++)
1461 (*pDataEq
)[i
].add(cz
);
1464 if (elementBytesLog2
< 2)
1466 // fill in z & y bit
1467 (*pDataEq
)[8].add(cz
);
1468 (*pDataEq
)[9].add(cy
);
1472 else if (elementBytesLog2
== 2)
1474 // fill in y and x bit
1475 (*pDataEq
)[8].add(cy
);
1476 (*pDataEq
)[9].add(cx
);
1483 (*pDataEq
)[8].add(cx
);
1485 (*pDataEq
)[9].add(cx
);
1492 UINT_32 m2dEnd
= (elementBytesLog2
==0) ? 3 : ((elementBytesLog2
< 4) ? 4 : 5);
1493 UINT_32 numZs
= (elementBytesLog2
== 0 || elementBytesLog2
== 4) ?
1494 2 : ((elementBytesLog2
== 1) ? 3 : 1);
1495 pDataEq
->mort2d(cx
, cy
, elementBytesLog2
, m2dEnd
);
1496 for (i
= m2dEnd
+ 1; i
<= m2dEnd
+ numZs
; i
++)
1498 (*pDataEq
)[i
].add(cz
);
1501 if ((elementBytesLog2
== 0) || (elementBytesLog2
== 3))
1504 (*pDataEq
)[6].add(cx
);
1505 (*pDataEq
)[7].add(cz
);
1509 else if (elementBytesLog2
== 2)
1512 (*pDataEq
)[6].add(cy
);
1513 (*pDataEq
)[7].add(cz
);
1518 (*pDataEq
)[8].add(cy
);
1519 (*pDataEq
)[9].add(cx
);
1523 // Fill in bit 10 and up
1524 pDataEq
->mort3d( cz
, cy
, cx
, 10 );
1526 else if (IsThin(resourceType
, swizzleMode
))
1528 UINT_32 blockSizeLog2
= GetBlockSizeLog2(swizzleMode
);
1530 UINT_32 microYBits
= (8 - elementBytesLog2
) / 2;
1531 UINT_32 tileSplitStart
= blockSizeLog2
- numSamplesLog2
;
1533 // Fill in bottom x bits
1534 for (i
= elementBytesLog2
; i
< 4; i
++)
1536 (*pDataEq
)[i
].add(cx
);
1539 // Fill in bottom y bits
1540 for (i
= 4; i
< 4 + microYBits
; i
++)
1542 (*pDataEq
)[i
].add(cy
);
1545 // Fill in last of the micro_x bits
1546 for (i
= 4 + microYBits
; i
< 8; i
++)
1548 (*pDataEq
)[i
].add(cx
);
1551 // Fill in x/y bits below sample split
1552 pDataEq
->mort2d(cy
, cx
, 8, tileSplitStart
- 1);
1553 // Fill in sample bits
1554 for (i
= 0; i
< numSamplesLog2
; i
++)
1557 (*pDataEq
)[tileSplitStart
+ i
].add(cs
);
1559 // Fill in x/y bits above sample split
1560 if ((numSamplesLog2
& 1) ^ (blockSizeLog2
& 1))
1562 pDataEq
->mort2d(cx
, cy
, blockSizeLog2
);
1566 pDataEq
->mort2d(cy
, cx
, blockSizeLog2
);
1571 ADDR_ASSERT_ALWAYS();
1577 UINT_32 sampleStart
= elementBytesLog2
;
1578 UINT_32 pixelStart
= elementBytesLog2
+ numSamplesLog2
;
1579 UINT_32 ymajStart
= 6 + numSamplesLog2
;
1581 for (UINT_32 s
= 0; s
< numSamplesLog2
; s
++)
1584 (*pDataEq
)[sampleStart
+ s
].add(cs
);
1587 // Put in the x-major order pixel bits
1588 pDataEq
->mort2d(cx
, cy
, pixelStart
, ymajStart
- 1);
1589 // Put in the y-major order pixel bits
1590 pDataEq
->mort2d(cy
, cx
, ymajStart
);
1595 ************************************************************************************************************************
1596 * Gfx9Lib::GetPipeEquation
1602 ************************************************************************************************************************
1604 VOID
Gfx9Lib::GetPipeEquation(
1605 CoordEq
* pPipeEq
, ///< [out] pipe equation
1606 CoordEq
* pDataEq
, ///< [in] data equation
1607 UINT_32 pipeInterleaveLog2
, ///< [in] pipe interleave
1608 UINT_32 numPipeLog2
, ///< [in] number of pipes
1609 UINT_32 numSamplesLog2
, ///< [in] data surface sample count
1610 Gfx9DataType dataSurfaceType
, ///< [in] data surface type
1611 AddrSwizzleMode swizzleMode
, ///< [in] data surface swizzle mode
1612 AddrResourceType resourceType
///< [in] data surface resource type
1615 UINT_32 blockSizeLog2
= GetBlockSizeLog2(swizzleMode
);
1618 pDataEq
->copy(dataEq
);
1620 if (dataSurfaceType
== Gfx9DataColor
)
1622 INT_32 shift
= static_cast<INT_32
>(numSamplesLog2
);
1623 dataEq
.shift(-shift
, blockSizeLog2
- numSamplesLog2
);
1626 dataEq
.copy(*pPipeEq
, pipeInterleaveLog2
, numPipeLog2
);
1628 // This section should only apply to z/stencil, maybe fmask
1629 // If the pipe bit is below the comp block size,
1630 // then keep moving up the address until we find a bit that is above
1631 UINT_32 pipeStart
= 0;
1633 if (dataSurfaceType
!= Gfx9DataColor
)
1635 Coordinate
tileMin('x', 3);
1637 while (dataEq
[pipeInterleaveLog2
+ pipeStart
][0] < tileMin
)
1642 // if pipe is 0, then the first pipe bit is above the comp block size,
1643 // so we don't need to do anything
1644 // Note, this if condition is not necessary, since if we execute the loop when pipe==0,
1645 // we will get the same pipe equation
1648 for (UINT_32 i
= 0; i
< numPipeLog2
; i
++)
1650 // Copy the jth bit above pipe interleave to the current pipe equation bit
1651 dataEq
[pipeInterleaveLog2
+ pipeStart
+ i
].copyto((*pPipeEq
)[i
]);
1656 if (IsPrt(swizzleMode
))
1658 // Clear out bits above the block size if prt's are enabled
1659 dataEq
.resize(blockSizeLog2
);
1663 if (IsXor(swizzleMode
))
1667 if (IsThick(resourceType
, swizzleMode
))
1671 dataEq
.copy(xorMask2
, pipeInterleaveLog2
+ numPipeLog2
, 2 * numPipeLog2
);
1673 xorMask
.resize(numPipeLog2
);
1675 for (UINT_32 pipeIdx
= 0; pipeIdx
< numPipeLog2
; pipeIdx
++)
1677 xorMask
[pipeIdx
].add(xorMask2
[2 * pipeIdx
]);
1678 xorMask
[pipeIdx
].add(xorMask2
[2 * pipeIdx
+ 1]);
1683 // Xor in the bits above the pipe+gpu bits
1684 dataEq
.copy(xorMask
, pipeInterleaveLog2
+ pipeStart
+ numPipeLog2
, numPipeLog2
);
1686 if ((numSamplesLog2
== 0) && (IsPrt(swizzleMode
) == FALSE
))
1690 // if 1xaa and not prt, then xor in the z bits
1692 xorMask2
.resize(numPipeLog2
);
1693 for (UINT_32 pipeIdx
= 0; pipeIdx
< numPipeLog2
; pipeIdx
++)
1695 co
.set('z', numPipeLog2
- 1 - pipeIdx
);
1696 xorMask2
[pipeIdx
].add(co
);
1699 pPipeEq
->xorin(xorMask2
);
1704 pPipeEq
->xorin(xorMask
);
1708 ************************************************************************************************************************
1709 * Gfx9Lib::GetMetaEquation
1712 * Get meta equation for cmask/htile/DCC
1714 * Pointer to a calculated meta equation
1715 ************************************************************************************************************************
1717 const CoordEq
* Gfx9Lib::GetMetaEquation(
1718 const MetaEqParams
& metaEqParams
)
1720 UINT_32 cachedMetaEqIndex
;
1722 for (cachedMetaEqIndex
= 0; cachedMetaEqIndex
< MaxCachedMetaEq
; cachedMetaEqIndex
++)
1724 if (memcmp(&metaEqParams
,
1725 &m_cachedMetaEqKey
[cachedMetaEqIndex
],
1726 static_cast<UINT_32
>(sizeof(metaEqParams
))) == 0)
1732 CoordEq
* pMetaEq
= NULL
;
1734 if (cachedMetaEqIndex
< MaxCachedMetaEq
)
1736 pMetaEq
= &m_cachedMetaEq
[cachedMetaEqIndex
];
1740 m_cachedMetaEqKey
[m_metaEqOverrideIndex
] = metaEqParams
;
1742 pMetaEq
= &m_cachedMetaEq
[m_metaEqOverrideIndex
++];
1744 m_metaEqOverrideIndex
%= MaxCachedMetaEq
;
1746 GenMetaEquation(pMetaEq
,
1747 metaEqParams
.maxMip
,
1748 metaEqParams
.elementBytesLog2
,
1749 metaEqParams
.numSamplesLog2
,
1750 metaEqParams
.metaFlag
,
1751 metaEqParams
.dataSurfaceType
,
1752 metaEqParams
.swizzleMode
,
1753 metaEqParams
.resourceType
,
1754 metaEqParams
.metaBlkWidthLog2
,
1755 metaEqParams
.metaBlkHeightLog2
,
1756 metaEqParams
.metaBlkDepthLog2
,
1757 metaEqParams
.compBlkWidthLog2
,
1758 metaEqParams
.compBlkHeightLog2
,
1759 metaEqParams
.compBlkDepthLog2
);
1766 ************************************************************************************************************************
1767 * Gfx9Lib::GenMetaEquation
1770 * Get meta equation for cmask/htile/DCC
1773 ************************************************************************************************************************
1775 VOID
Gfx9Lib::GenMetaEquation(
1776 CoordEq
* pMetaEq
, ///< [out] meta equation
1777 UINT_32 maxMip
, ///< [in] max mip Id
1778 UINT_32 elementBytesLog2
, ///< [in] data surface element bytes
1779 UINT_32 numSamplesLog2
, ///< [in] data surface sample count
1780 ADDR2_META_FLAGS metaFlag
, ///< [in] meta falg
1781 Gfx9DataType dataSurfaceType
, ///< [in] data surface type
1782 AddrSwizzleMode swizzleMode
, ///< [in] data surface swizzle mode
1783 AddrResourceType resourceType
, ///< [in] data surface resource type
1784 UINT_32 metaBlkWidthLog2
, ///< [in] meta block width
1785 UINT_32 metaBlkHeightLog2
, ///< [in] meta block height
1786 UINT_32 metaBlkDepthLog2
, ///< [in] meta block depth
1787 UINT_32 compBlkWidthLog2
, ///< [in] compress block width
1788 UINT_32 compBlkHeightLog2
, ///< [in] compress block height
1789 UINT_32 compBlkDepthLog2
) ///< [in] compress block depth
1792 UINT_32 numPipeTotalLog2
= GetPipeLog2ForMetaAddressing(metaFlag
.pipeAligned
, swizzleMode
);
1793 UINT_32 pipeInterleaveLog2
= m_pipeInterleaveLog2
;
1795 // Get the correct data address and rb equation
1797 GetDataEquation(&dataEq
, dataSurfaceType
, swizzleMode
, resourceType
,
1798 elementBytesLog2
, numSamplesLog2
);
1800 // Get pipe and rb equations
1801 CoordEq pipeEquation
;
1802 GetPipeEquation(&pipeEquation
, &dataEq
, pipeInterleaveLog2
, numPipeTotalLog2
,
1803 numSamplesLog2
, dataSurfaceType
, swizzleMode
, resourceType
);
1804 numPipeTotalLog2
= pipeEquation
.getsize();
1806 if (metaFlag
.linear
)
1808 // Linear metadata supporting was removed for GFX9! No one can use this feature.
1809 ADDR_ASSERT_ALWAYS();
1811 ADDR_ASSERT(dataSurfaceType
== Gfx9DataColor
);
1813 dataEq
.copy(*pMetaEq
);
1815 if (IsLinear(swizzleMode
))
1817 if (metaFlag
.pipeAligned
)
1819 // Remove the pipe bits
1820 INT_32 shift
= static_cast<INT_32
>(numPipeTotalLog2
);
1821 pMetaEq
->shift(-shift
, pipeInterleaveLog2
);
1823 // Divide by comp block size, which for linear (which is always color) is 256 B
1826 if (metaFlag
.pipeAligned
)
1828 // Put pipe bits back in
1829 pMetaEq
->shift(numPipeTotalLog2
, pipeInterleaveLog2
);
1831 for (UINT_32 i
= 0; i
< numPipeTotalLog2
; i
++)
1833 pipeEquation
[i
].copyto((*pMetaEq
)[pipeInterleaveLog2
+ i
]);
1842 UINT_32 maxCompFragLog2
= static_cast<INT_32
>(m_maxCompFragLog2
);
1843 UINT_32 compFragLog2
=
1844 ((dataSurfaceType
== Gfx9DataColor
) && (numSamplesLog2
> maxCompFragLog2
)) ?
1845 maxCompFragLog2
: numSamplesLog2
;
1847 UINT_32 uncompFragLog2
= numSamplesLog2
- compFragLog2
;
1849 // Make sure the metaaddr is cleared
1851 pMetaEq
->resize(27);
1853 if (IsThick(resourceType
, swizzleMode
))
1855 Coordinate
cx('x', 0);
1856 Coordinate
cy('y', 0);
1857 Coordinate
cz('z', 0);
1861 pMetaEq
->mort3d(cy
, cx
, cz
);
1865 pMetaEq
->mort3d(cx
, cy
, cz
);
1870 Coordinate
cx('x', 0);
1871 Coordinate
cy('y', 0);
1876 pMetaEq
->mort2d(cy
, cx
, compFragLog2
);
1880 pMetaEq
->mort2d(cx
, cy
, compFragLog2
);
1883 //------------------------------------------------------------------------------------------------------------------------
1884 // Put the compressible fragments at the lsb
1885 // the uncompressible frags will be at the msb of the micro address
1886 //------------------------------------------------------------------------------------------------------------------------
1887 for (UINT_32 s
= 0; s
< compFragLog2
; s
++)
1890 (*pMetaEq
)[s
].add(cs
);
1894 // Keep a copy of the pipe equations
1895 CoordEq origPipeEquation
;
1896 pipeEquation
.copy(origPipeEquation
);
1899 // filter out everything under the compressed block size
1900 co
.set('x', compBlkWidthLog2
);
1901 pMetaEq
->Filter('<', co
, 0, 'x');
1902 co
.set('y', compBlkHeightLog2
);
1903 pMetaEq
->Filter('<', co
, 0, 'y');
1904 co
.set('z', compBlkDepthLog2
);
1905 pMetaEq
->Filter('<', co
, 0, 'z');
1907 // For non-color, filter out sample bits
1908 if (dataSurfaceType
!= Gfx9DataColor
)
1911 pMetaEq
->Filter('<', co
, 0, 's');
1914 // filter out everything above the metablock size
1915 co
.set('x', metaBlkWidthLog2
- 1);
1916 pMetaEq
->Filter('>', co
, 0, 'x');
1917 co
.set('y', metaBlkHeightLog2
- 1);
1918 pMetaEq
->Filter('>', co
, 0, 'y');
1919 co
.set('z', metaBlkDepthLog2
- 1);
1920 pMetaEq
->Filter('>', co
, 0, 'z');
1922 // filter out everything above the metablock size for the channel bits
1923 co
.set('x', metaBlkWidthLog2
- 1);
1924 pipeEquation
.Filter('>', co
, 0, 'x');
1925 co
.set('y', metaBlkHeightLog2
- 1);
1926 pipeEquation
.Filter('>', co
, 0, 'y');
1927 co
.set('z', metaBlkDepthLog2
- 1);
1928 pipeEquation
.Filter('>', co
, 0, 'z');
1930 // Make sure we still have the same number of channel bits
1931 if (pipeEquation
.getsize() != numPipeTotalLog2
)
1933 ADDR_ASSERT_ALWAYS();
1936 // Loop through all channel and rb bits,
1937 // and make sure these components exist in the metadata address
1938 for (UINT_32 i
= 0; i
< numPipeTotalLog2
; i
++)
1940 for (UINT_32 j
= pipeEquation
[i
].getsize(); j
> 0; j
--)
1942 if (pMetaEq
->Exists(pipeEquation
[i
][j
- 1]) == FALSE
)
1944 ADDR_ASSERT_ALWAYS();
1949 const UINT_32 numSeLog2
= metaFlag
.rbAligned
? m_seLog2
: 0;
1950 const UINT_32 numRbPeSeLog2
= metaFlag
.rbAligned
? m_rbPerSeLog2
: 0;
1951 const UINT_32 numRbTotalLog2
= numRbPeSeLog2
+ numSeLog2
;
1952 CoordEq origRbEquation
;
1954 GetRbEquation(&origRbEquation
, numRbPeSeLog2
, numSeLog2
);
1956 CoordEq rbEquation
= origRbEquation
;
1958 for (UINT_32 i
= 0; i
< numRbTotalLog2
; i
++)
1960 for (UINT_32 j
= rbEquation
[i
].getsize(); j
> 0; j
--)
1962 if (pMetaEq
->Exists(rbEquation
[i
][j
- 1]) == FALSE
)
1964 ADDR_ASSERT_ALWAYS();
1969 if (m_settings
.applyAliasFix
)
1974 // Loop through each rb id bit; if it is equal to any of the filtered channel bits, clear it
1975 for (UINT_32 i
= 0; i
< numRbTotalLog2
; i
++)
1977 for (UINT_32 j
= 0; j
< numPipeTotalLog2
; j
++)
1979 BOOL_32 isRbEquationInPipeEquation
= FALSE
;
1981 if (m_settings
.applyAliasFix
)
1983 CoordTerm filteredPipeEq
;
1984 filteredPipeEq
= pipeEquation
[j
];
1986 filteredPipeEq
.Filter('>', co
, 0, 'z');
1988 isRbEquationInPipeEquation
= (rbEquation
[i
] == filteredPipeEq
);
1992 isRbEquationInPipeEquation
= (rbEquation
[i
] == pipeEquation
[j
]);
1995 if (isRbEquationInPipeEquation
)
1997 rbEquation
[i
].Clear();
2002 bool rbAppendedWithPipeBits
[1 << (MaxSeLog2
+ MaxRbPerSeLog2
)] = {};
2004 // Loop through each bit of the channel, get the smallest coordinate,
2005 // and remove it from the metaaddr, and rb_equation
2006 for (UINT_32 i
= 0; i
< numPipeTotalLog2
; i
++)
2008 pipeEquation
[i
].getsmallest(co
);
2010 UINT_32 old_size
= pMetaEq
->getsize();
2011 pMetaEq
->Filter('=', co
);
2012 UINT_32 new_size
= pMetaEq
->getsize();
2013 if (new_size
!= old_size
-1)
2015 ADDR_ASSERT_ALWAYS();
2017 pipeEquation
.remove(co
);
2018 for (UINT_32 j
= 0; j
< numRbTotalLog2
; j
++)
2020 if (rbEquation
[j
].remove(co
))
2022 // if we actually removed something from this bit, then add the remaining
2023 // channel bits, as these can be removed for this bit
2024 for (UINT_32 k
= 0; k
< pipeEquation
[i
].getsize(); k
++)
2026 if (pipeEquation
[i
][k
] != co
)
2028 rbEquation
[j
].add(pipeEquation
[i
][k
]);
2029 rbAppendedWithPipeBits
[j
] = true;
2036 // Loop through the rb bits and see what remain;
2037 // filter out the smallest coordinate if it remains
2038 UINT_32 rbBitsLeft
= 0;
2039 for (UINT_32 i
= 0; i
< numRbTotalLog2
; i
++)
2041 BOOL_32 isRbEqAppended
= FALSE
;
2043 if (m_settings
.applyAliasFix
)
2045 isRbEqAppended
= (rbEquation
[i
].getsize() > (rbAppendedWithPipeBits
[i
] ? 1 : 0));
2049 isRbEqAppended
= (rbEquation
[i
].getsize() > 0);
2055 rbEquation
[i
].getsmallest(co
);
2056 UINT_32 old_size
= pMetaEq
->getsize();
2057 pMetaEq
->Filter('=', co
);
2058 UINT_32 new_size
= pMetaEq
->getsize();
2059 if (new_size
!= old_size
- 1)
2063 for (UINT_32 j
= i
+ 1; j
< numRbTotalLog2
; j
++)
2065 if (rbEquation
[j
].remove(co
))
2067 // if we actually removed something from this bit, then add the remaining
2068 // rb bits, as these can be removed for this bit
2069 for (UINT_32 k
= 0; k
< rbEquation
[i
].getsize(); k
++)
2071 if (rbEquation
[i
][k
] != co
)
2073 rbEquation
[j
].add(rbEquation
[i
][k
]);
2074 rbAppendedWithPipeBits
[j
] |= rbAppendedWithPipeBits
[i
];
2082 // capture the size of the metaaddr
2083 UINT_32 metaSize
= pMetaEq
->getsize();
2084 // resize to 49 bits...make this a nibble address
2085 pMetaEq
->resize(49);
2086 // Concatenate the macro address above the current address
2087 for (UINT_32 i
= metaSize
, j
= 0; i
< 49; i
++, j
++)
2090 (*pMetaEq
)[i
].add(co
);
2093 // Multiply by meta element size (in nibbles)
2094 if (dataSurfaceType
== Gfx9DataColor
)
2098 else if (dataSurfaceType
== Gfx9DataDepthStencil
)
2103 //------------------------------------------------------------------------------------------
2104 // Note the pipeInterleaveLog2+1 is because address is a nibble address
2105 // Shift up from pipe interleave number of channel
2106 // and rb bits left, and uncompressed fragments
2107 //------------------------------------------------------------------------------------------
2109 pMetaEq
->shift(numPipeTotalLog2
+ rbBitsLeft
+ uncompFragLog2
, pipeInterleaveLog2
+ 1);
2111 // Put in the channel bits
2112 for (UINT_32 i
= 0; i
< numPipeTotalLog2
; i
++)
2114 origPipeEquation
[i
].copyto((*pMetaEq
)[pipeInterleaveLog2
+1 + i
]);
2117 // Put in remaining rb bits
2118 for (UINT_32 i
= 0, j
= 0; j
< rbBitsLeft
; i
= (i
+ 1) % numRbTotalLog2
)
2120 BOOL_32 isRbEqAppended
= FALSE
;
2122 if (m_settings
.applyAliasFix
)
2124 isRbEqAppended
= (rbEquation
[i
].getsize() > (rbAppendedWithPipeBits
[i
] ? 1 : 0));
2128 isRbEqAppended
= (rbEquation
[i
].getsize() > 0);
2133 origRbEquation
[i
].copyto((*pMetaEq
)[pipeInterleaveLog2
+ 1 + numPipeTotalLog2
+ j
]);
2134 // Mark any rb bit we add in to the rb mask
2139 //------------------------------------------------------------------------------------------
2140 // Put in the uncompressed fragment bits
2141 //------------------------------------------------------------------------------------------
2142 for (UINT_32 i
= 0; i
< uncompFragLog2
; i
++)
2144 co
.set('s', compFragLog2
+ i
);
2145 (*pMetaEq
)[pipeInterleaveLog2
+ 1 + numPipeTotalLog2
+ rbBitsLeft
+ i
].add(co
);
2151 ************************************************************************************************************************
2152 * Gfx9Lib::IsEquationSupported
2155 * Check if equation is supported for given swizzle mode and resource type.
2159 ************************************************************************************************************************
2161 BOOL_32
Gfx9Lib::IsEquationSupported(
2162 AddrResourceType rsrcType
,
2163 AddrSwizzleMode swMode
,
2164 UINT_32 elementBytesLog2
) const
2166 BOOL_32 supported
= (elementBytesLog2
< MaxElementBytesLog2
) &&
2167 (IsLinear(swMode
) == FALSE
) &&
2168 (((IsTex2d(rsrcType
) == TRUE
) &&
2169 ((elementBytesLog2
< 4) ||
2170 ((IsRotateSwizzle(swMode
) == FALSE
) &&
2171 (IsZOrderSwizzle(swMode
) == FALSE
)))) ||
2172 ((IsTex3d(rsrcType
) == TRUE
) &&
2173 (IsRotateSwizzle(swMode
) == FALSE
) &&
2174 (IsBlock256b(swMode
) == FALSE
)));
2180 ************************************************************************************************************************
2181 * Gfx9Lib::InitEquationTable
2184 * Initialize Equation table.
2188 ************************************************************************************************************************
2190 VOID
Gfx9Lib::InitEquationTable()
2192 memset(m_equationTable
, 0, sizeof(m_equationTable
));
2194 // Loop all possible resource type (2D/3D)
2195 for (UINT_32 rsrcTypeIdx
= 0; rsrcTypeIdx
< MaxRsrcType
; rsrcTypeIdx
++)
2197 AddrResourceType rsrcType
= static_cast<AddrResourceType
>(rsrcTypeIdx
+ ADDR_RSRC_TEX_2D
);
2199 // Loop all possible swizzle mode
2200 for (UINT_32 swModeIdx
= 0; swModeIdx
< MaxSwMode
; swModeIdx
++)
2202 AddrSwizzleMode swMode
= static_cast<AddrSwizzleMode
>(swModeIdx
);
2204 // Loop all possible bpp
2205 for (UINT_32 bppIdx
= 0; bppIdx
< MaxElementBytesLog2
; bppIdx
++)
2207 UINT_32 equationIndex
= ADDR_INVALID_EQUATION_INDEX
;
2209 // Check if the input is supported
2210 if (IsEquationSupported(rsrcType
, swMode
, bppIdx
))
2212 ADDR_EQUATION equation
;
2213 ADDR_E_RETURNCODE retCode
;
2215 memset(&equation
, 0, sizeof(ADDR_EQUATION
));
2217 // Generate the equation
2218 if (IsBlock256b(swMode
) && IsTex2d(rsrcType
))
2220 retCode
= ComputeBlock256Equation(rsrcType
, swMode
, bppIdx
, &equation
);
2222 else if (IsThin(rsrcType
, swMode
))
2224 retCode
= ComputeThinEquation(rsrcType
, swMode
, bppIdx
, &equation
);
2228 retCode
= ComputeThickEquation(rsrcType
, swMode
, bppIdx
, &equation
);
2231 // Only fill the equation into the table if the return code is ADDR_OK,
2232 // otherwise if the return code is not ADDR_OK, it indicates this is not
2233 // a valid input, we do nothing but just fill invalid equation index
2234 // into the lookup table.
2235 if (retCode
== ADDR_OK
)
2237 equationIndex
= m_numEquations
;
2238 ADDR_ASSERT(equationIndex
< EquationTableSize
);
2240 m_equationTable
[equationIndex
] = equation
;
2246 ADDR_ASSERT_ALWAYS();
2250 // Fill the index into the lookup table, if the combination is not supported
2251 // fill the invalid equation index
2252 m_equationLookupTable
[rsrcTypeIdx
][swModeIdx
][bppIdx
] = equationIndex
;
2259 ************************************************************************************************************************
2260 * Gfx9Lib::HwlGetEquationIndex
2263 * Interface function stub of GetEquationIndex
2267 ************************************************************************************************************************
2269 UINT_32
Gfx9Lib::HwlGetEquationIndex(
2270 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
,
2271 ADDR2_COMPUTE_SURFACE_INFO_OUTPUT
* pOut
2274 AddrResourceType rsrcType
= pIn
->resourceType
;
2275 AddrSwizzleMode swMode
= pIn
->swizzleMode
;
2276 UINT_32 elementBytesLog2
= Log2(pIn
->bpp
>> 3);
2277 UINT_32 index
= ADDR_INVALID_EQUATION_INDEX
;
2279 if (IsEquationSupported(rsrcType
, swMode
, elementBytesLog2
))
2281 UINT_32 rsrcTypeIdx
= static_cast<UINT_32
>(rsrcType
) - 1;
2282 UINT_32 swModeIdx
= static_cast<UINT_32
>(swMode
);
2284 index
= m_equationLookupTable
[rsrcTypeIdx
][swModeIdx
][elementBytesLog2
];
2287 if (pOut
->pMipInfo
!= NULL
)
2289 for (UINT_32 i
= 0; i
< pIn
->numMipLevels
; i
++)
2291 pOut
->pMipInfo
[i
].equationIndex
= index
;
2299 ************************************************************************************************************************
2300 * Gfx9Lib::HwlComputeBlock256Equation
2303 * Interface function stub of ComputeBlock256Equation
2307 ************************************************************************************************************************
2309 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeBlock256Equation(
2310 AddrResourceType rsrcType
,
2311 AddrSwizzleMode swMode
,
2312 UINT_32 elementBytesLog2
,
2313 ADDR_EQUATION
* pEquation
) const
2315 ADDR_E_RETURNCODE ret
= ADDR_OK
;
2317 pEquation
->numBits
= 8;
2320 for (; i
< elementBytesLog2
; i
++)
2322 InitChannel(1, 0 , i
, &pEquation
->addr
[i
]);
2325 ADDR_CHANNEL_SETTING
* pixelBit
= &pEquation
->addr
[elementBytesLog2
];
2327 const UINT_32 maxBitsUsed
= 4;
2328 ADDR_CHANNEL_SETTING x
[maxBitsUsed
] = {};
2329 ADDR_CHANNEL_SETTING y
[maxBitsUsed
] = {};
2331 for (i
= 0; i
< maxBitsUsed
; i
++)
2333 InitChannel(1, 0, elementBytesLog2
+ i
, &x
[i
]);
2334 InitChannel(1, 1, i
, &y
[i
]);
2337 if (IsStandardSwizzle(rsrcType
, swMode
))
2339 switch (elementBytesLog2
)
2382 ADDR_ASSERT_ALWAYS();
2383 ret
= ADDR_INVALIDPARAMS
;
2387 else if (IsDisplaySwizzle(rsrcType
, swMode
))
2389 switch (elementBytesLog2
)
2432 ADDR_ASSERT_ALWAYS();
2433 ret
= ADDR_INVALIDPARAMS
;
2437 else if (IsRotateSwizzle(swMode
))
2439 switch (elementBytesLog2
)
2476 ADDR_ASSERT_ALWAYS();
2478 ret
= ADDR_INVALIDPARAMS
;
2484 ADDR_ASSERT_ALWAYS();
2485 ret
= ADDR_INVALIDPARAMS
;
2491 ASSERTED Dim2d microBlockDim
= Block256_2d
[elementBytesLog2
];
2492 ADDR_ASSERT((2u << GetMaxValidChannelIndex(pEquation
->addr
, 8, 0)) ==
2493 (microBlockDim
.w
* (1 << elementBytesLog2
)));
2494 ADDR_ASSERT((2u << GetMaxValidChannelIndex(pEquation
->addr
, 8, 1)) == microBlockDim
.h
);
2501 ************************************************************************************************************************
2502 * Gfx9Lib::HwlComputeThinEquation
2505 * Interface function stub of ComputeThinEquation
2509 ************************************************************************************************************************
2511 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeThinEquation(
2512 AddrResourceType rsrcType
,
2513 AddrSwizzleMode swMode
,
2514 UINT_32 elementBytesLog2
,
2515 ADDR_EQUATION
* pEquation
) const
2517 ADDR_E_RETURNCODE ret
= ADDR_OK
;
2519 UINT_32 blockSizeLog2
= GetBlockSizeLog2(swMode
);
2521 UINT_32 maxXorBits
= blockSizeLog2
;
2522 if (IsNonPrtXor(swMode
))
2524 // For non-prt-xor, maybe need to initialize some more bits for xor
2525 // The highest xor bit used in equation will be max the following 3 items:
2526 // 1. m_pipeInterleaveLog2 + 2 * pipeXorBits
2527 // 2. m_pipeInterleaveLog2 + pipeXorBits + 2 * bankXorBits
2530 maxXorBits
= Max(maxXorBits
, m_pipeInterleaveLog2
+ 2 * GetPipeXorBits(blockSizeLog2
));
2531 maxXorBits
= Max(maxXorBits
, m_pipeInterleaveLog2
+
2532 GetPipeXorBits(blockSizeLog2
) +
2533 2 * GetBankXorBits(blockSizeLog2
));
2536 const UINT_32 maxBitsUsed
= 14;
2537 ADDR_ASSERT((2 * maxBitsUsed
) >= maxXorBits
);
2538 ADDR_CHANNEL_SETTING x
[maxBitsUsed
] = {};
2539 ADDR_CHANNEL_SETTING y
[maxBitsUsed
] = {};
2541 const UINT_32 extraXorBits
= 16;
2542 ADDR_ASSERT(extraXorBits
>= maxXorBits
- blockSizeLog2
);
2543 ADDR_CHANNEL_SETTING xorExtra
[extraXorBits
] = {};
2545 for (UINT_32 i
= 0; i
< maxBitsUsed
; i
++)
2547 InitChannel(1, 0, elementBytesLog2
+ i
, &x
[i
]);
2548 InitChannel(1, 1, i
, &y
[i
]);
2551 ADDR_CHANNEL_SETTING
* pixelBit
= pEquation
->addr
;
2553 for (UINT_32 i
= 0; i
< elementBytesLog2
; i
++)
2555 InitChannel(1, 0 , i
, &pixelBit
[i
]);
2560 UINT_32 lowBits
= 0;
2562 if (IsZOrderSwizzle(swMode
))
2564 if (elementBytesLog2
<= 3)
2566 for (UINT_32 i
= elementBytesLog2
; i
< 6; i
++)
2568 pixelBit
[i
] = (((i
- elementBytesLog2
) & 1) == 0) ? x
[xIdx
++] : y
[yIdx
++];
2575 ret
= ADDR_INVALIDPARAMS
;
2580 ret
= HwlComputeBlock256Equation(rsrcType
, swMode
, elementBytesLog2
, pEquation
);
2584 Dim2d microBlockDim
= Block256_2d
[elementBytesLog2
];
2585 xIdx
= Log2(microBlockDim
.w
);
2586 yIdx
= Log2(microBlockDim
.h
);
2593 for (UINT_32 i
= lowBits
; i
< blockSizeLog2
; i
++)
2595 pixelBit
[i
] = ((i
& 1) == 0) ? y
[yIdx
++] : x
[xIdx
++];
2598 for (UINT_32 i
= blockSizeLog2
; i
< maxXorBits
; i
++)
2600 xorExtra
[i
- blockSizeLog2
] = ((i
& 1) == 0) ? y
[yIdx
++] : x
[xIdx
++];
2606 UINT_32 pipeStart
= m_pipeInterleaveLog2
;
2607 UINT_32 pipeXorBits
= GetPipeXorBits(blockSizeLog2
);
2609 UINT_32 bankStart
= pipeStart
+ pipeXorBits
;
2610 UINT_32 bankXorBits
= GetBankXorBits(blockSizeLog2
);
2612 for (UINT_32 i
= 0; i
< pipeXorBits
; i
++)
2614 UINT_32 xor1BitPos
= pipeStart
+ 2 * pipeXorBits
- 1 - i
;
2615 ADDR_CHANNEL_SETTING
* pXor1Src
= (xor1BitPos
< blockSizeLog2
) ?
2616 &pEquation
->addr
[xor1BitPos
] : &xorExtra
[xor1BitPos
- blockSizeLog2
];
2618 InitChannel(&pEquation
->xor1
[pipeStart
+ i
], pXor1Src
);
2621 for (UINT_32 i
= 0; i
< bankXorBits
; i
++)
2623 UINT_32 xor1BitPos
= bankStart
+ 2 * bankXorBits
- 1 - i
;
2624 ADDR_CHANNEL_SETTING
* pXor1Src
= (xor1BitPos
< blockSizeLog2
) ?
2625 &pEquation
->addr
[xor1BitPos
] : &xorExtra
[xor1BitPos
- blockSizeLog2
];
2627 InitChannel(&pEquation
->xor1
[bankStart
+ i
], pXor1Src
);
2630 if (IsPrt(swMode
) == FALSE
)
2632 for (UINT_32 i
= 0; i
< pipeXorBits
; i
++)
2634 InitChannel(1, 2, pipeXorBits
- i
- 1, &pEquation
->xor2
[pipeStart
+ i
]);
2637 for (UINT_32 i
= 0; i
< bankXorBits
; i
++)
2639 InitChannel(1, 2, bankXorBits
- i
- 1 + pipeXorBits
, &pEquation
->xor2
[bankStart
+ i
]);
2644 pEquation
->numBits
= blockSizeLog2
;
2651 ************************************************************************************************************************
2652 * Gfx9Lib::HwlComputeThickEquation
2655 * Interface function stub of ComputeThickEquation
2659 ************************************************************************************************************************
2661 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeThickEquation(
2662 AddrResourceType rsrcType
,
2663 AddrSwizzleMode swMode
,
2664 UINT_32 elementBytesLog2
,
2665 ADDR_EQUATION
* pEquation
) const
2667 ADDR_E_RETURNCODE ret
= ADDR_OK
;
2669 ADDR_ASSERT(IsTex3d(rsrcType
));
2671 UINT_32 blockSizeLog2
= GetBlockSizeLog2(swMode
);
2673 UINT_32 maxXorBits
= blockSizeLog2
;
2674 if (IsNonPrtXor(swMode
))
2676 // For non-prt-xor, maybe need to initialize some more bits for xor
2677 // The highest xor bit used in equation will be max the following 3:
2678 // 1. m_pipeInterleaveLog2 + 3 * pipeXorBits
2679 // 2. m_pipeInterleaveLog2 + pipeXorBits + 3 * bankXorBits
2682 maxXorBits
= Max(maxXorBits
, m_pipeInterleaveLog2
+ 3 * GetPipeXorBits(blockSizeLog2
));
2683 maxXorBits
= Max(maxXorBits
, m_pipeInterleaveLog2
+
2684 GetPipeXorBits(blockSizeLog2
) +
2685 3 * GetBankXorBits(blockSizeLog2
));
2688 for (UINT_32 i
= 0; i
< elementBytesLog2
; i
++)
2690 InitChannel(1, 0 , i
, &pEquation
->addr
[i
]);
2693 ADDR_CHANNEL_SETTING
* pixelBit
= &pEquation
->addr
[elementBytesLog2
];
2695 const UINT_32 maxBitsUsed
= 12;
2696 ADDR_ASSERT((3 * maxBitsUsed
) >= maxXorBits
);
2697 ADDR_CHANNEL_SETTING x
[maxBitsUsed
] = {};
2698 ADDR_CHANNEL_SETTING y
[maxBitsUsed
] = {};
2699 ADDR_CHANNEL_SETTING z
[maxBitsUsed
] = {};
2701 const UINT_32 extraXorBits
= 24;
2702 ADDR_ASSERT(extraXorBits
>= maxXorBits
- blockSizeLog2
);
2703 ADDR_CHANNEL_SETTING xorExtra
[extraXorBits
] = {};
2705 for (UINT_32 i
= 0; i
< maxBitsUsed
; i
++)
2707 InitChannel(1, 0, elementBytesLog2
+ i
, &x
[i
]);
2708 InitChannel(1, 1, i
, &y
[i
]);
2709 InitChannel(1, 2, i
, &z
[i
]);
2712 if (IsZOrderSwizzle(swMode
))
2714 switch (elementBytesLog2
)
2767 ADDR_ASSERT_ALWAYS();
2768 ret
= ADDR_INVALIDPARAMS
;
2772 else if (IsStandardSwizzle(rsrcType
, swMode
))
2774 switch (elementBytesLog2
)
2827 ADDR_ASSERT_ALWAYS();
2828 ret
= ADDR_INVALIDPARAMS
;
2834 ADDR_ASSERT_ALWAYS();
2835 ret
= ADDR_INVALIDPARAMS
;
2840 Dim3d microBlockDim
= Block1K_3d
[elementBytesLog2
];
2841 UINT_32 xIdx
= Log2(microBlockDim
.w
);
2842 UINT_32 yIdx
= Log2(microBlockDim
.h
);
2843 UINT_32 zIdx
= Log2(microBlockDim
.d
);
2845 pixelBit
= pEquation
->addr
;
2847 const UINT_32 lowBits
= 10;
2848 ADDR_ASSERT(pEquation
->addr
[lowBits
- 1].valid
== 1);
2849 ADDR_ASSERT(pEquation
->addr
[lowBits
].valid
== 0);
2851 for (UINT_32 i
= lowBits
; i
< blockSizeLog2
; i
++)
2855 pixelBit
[i
] = x
[xIdx
++];
2857 else if ((i
% 3) == 1)
2859 pixelBit
[i
] = z
[zIdx
++];
2863 pixelBit
[i
] = y
[yIdx
++];
2867 for (UINT_32 i
= blockSizeLog2
; i
< maxXorBits
; i
++)
2871 xorExtra
[i
- blockSizeLog2
] = x
[xIdx
++];
2873 else if ((i
% 3) == 1)
2875 xorExtra
[i
- blockSizeLog2
] = z
[zIdx
++];
2879 xorExtra
[i
- blockSizeLog2
] = y
[yIdx
++];
2886 UINT_32 pipeStart
= m_pipeInterleaveLog2
;
2887 UINT_32 pipeXorBits
= GetPipeXorBits(blockSizeLog2
);
2888 for (UINT_32 i
= 0; i
< pipeXorBits
; i
++)
2890 UINT_32 xor1BitPos
= pipeStart
+ (3 * pipeXorBits
) - 1 - (2 * i
);
2891 ADDR_CHANNEL_SETTING
* pXor1Src
= (xor1BitPos
< blockSizeLog2
) ?
2892 &pEquation
->addr
[xor1BitPos
] : &xorExtra
[xor1BitPos
- blockSizeLog2
];
2894 InitChannel(&pEquation
->xor1
[pipeStart
+ i
], pXor1Src
);
2896 UINT_32 xor2BitPos
= pipeStart
+ (3 * pipeXorBits
) - 2 - (2 * i
);
2897 ADDR_CHANNEL_SETTING
* pXor2Src
= (xor2BitPos
< blockSizeLog2
) ?
2898 &pEquation
->addr
[xor2BitPos
] : &xorExtra
[xor2BitPos
- blockSizeLog2
];
2900 InitChannel(&pEquation
->xor2
[pipeStart
+ i
], pXor2Src
);
2903 UINT_32 bankStart
= pipeStart
+ pipeXorBits
;
2904 UINT_32 bankXorBits
= GetBankXorBits(blockSizeLog2
);
2905 for (UINT_32 i
= 0; i
< bankXorBits
; i
++)
2907 UINT_32 xor1BitPos
= bankStart
+ (3 * bankXorBits
) - 1 - (2 * i
);
2908 ADDR_CHANNEL_SETTING
* pXor1Src
= (xor1BitPos
< blockSizeLog2
) ?
2909 &pEquation
->addr
[xor1BitPos
] : &xorExtra
[xor1BitPos
- blockSizeLog2
];
2911 InitChannel(&pEquation
->xor1
[bankStart
+ i
], pXor1Src
);
2913 UINT_32 xor2BitPos
= bankStart
+ (3 * bankXorBits
) - 2 - (2 * i
);
2914 ADDR_CHANNEL_SETTING
* pXor2Src
= (xor2BitPos
< blockSizeLog2
) ?
2915 &pEquation
->addr
[xor2BitPos
] : &xorExtra
[xor2BitPos
- blockSizeLog2
];
2917 InitChannel(&pEquation
->xor2
[bankStart
+ i
], pXor2Src
);
2921 pEquation
->numBits
= blockSizeLog2
;
2928 ************************************************************************************************************************
2929 * Gfx9Lib::IsValidDisplaySwizzleMode
2932 * Check if a swizzle mode is supported by display engine
2935 * TRUE is swizzle mode is supported by display engine
2936 ************************************************************************************************************************
2938 BOOL_32
Gfx9Lib::IsValidDisplaySwizzleMode(
2939 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
) const
2941 BOOL_32 support
= FALSE
;
2943 if (m_settings
.isDce12
)
2945 switch (pIn
->swizzleMode
)
2947 case ADDR_SW_256B_D
:
2948 case ADDR_SW_256B_R
:
2949 support
= (pIn
->bpp
== 32);
2952 case ADDR_SW_LINEAR
:
2955 case ADDR_SW_64KB_D
:
2956 case ADDR_SW_64KB_R
:
2959 case ADDR_SW_4KB_D_X
:
2960 case ADDR_SW_4KB_R_X
:
2961 case ADDR_SW_64KB_D_X
:
2962 case ADDR_SW_64KB_R_X
:
2963 case ADDR_SW_VAR_D_X
:
2964 case ADDR_SW_VAR_R_X
:
2965 support
= (pIn
->bpp
<= 64);
2972 else if (m_settings
.isDcn1
)
2974 switch (pIn
->swizzleMode
)
2977 case ADDR_SW_64KB_D
:
2979 case ADDR_SW_64KB_D_T
:
2980 case ADDR_SW_4KB_D_X
:
2981 case ADDR_SW_64KB_D_X
:
2982 case ADDR_SW_VAR_D_X
:
2983 support
= (pIn
->bpp
== 64);
2986 case ADDR_SW_LINEAR
:
2988 case ADDR_SW_64KB_S
:
2990 case ADDR_SW_64KB_S_T
:
2991 case ADDR_SW_4KB_S_X
:
2992 case ADDR_SW_64KB_S_X
:
2993 case ADDR_SW_VAR_S_X
:
2994 support
= (pIn
->bpp
<= 64);
3003 ADDR_NOT_IMPLEMENTED();
3010 ************************************************************************************************************************
3011 * Gfx9Lib::HwlComputePipeBankXor
3014 * Generate a PipeBankXor value to be ORed into bits above pipeInterleaveBits of address
3018 ************************************************************************************************************************
3020 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputePipeBankXor(
3021 const ADDR2_COMPUTE_PIPEBANKXOR_INPUT
* pIn
,
3022 ADDR2_COMPUTE_PIPEBANKXOR_OUTPUT
* pOut
) const
3024 if (IsXor(pIn
->swizzleMode
))
3026 UINT_32 macroBlockBits
= GetBlockSizeLog2(pIn
->swizzleMode
);
3027 UINT_32 pipeBits
= GetPipeXorBits(macroBlockBits
);
3028 UINT_32 bankBits
= GetBankXorBits(macroBlockBits
);
3030 UINT_32 pipeXor
= 0;
3031 UINT_32 bankXor
= 0;
3033 const UINT_32 bankMask
= (1 << bankBits
) - 1;
3034 const UINT_32 index
= pIn
->surfIndex
& bankMask
;
3036 const UINT_32 bpp
= pIn
->flags
.fmask
?
3037 GetFmaskBpp(pIn
->numSamples
, pIn
->numFrags
) : GetElemLib()->GetBitsPerPixel(pIn
->format
);
3040 static const UINT_32 BankXorSmallBpp
[] = {0, 7, 4, 3, 8, 15, 12, 11, 1, 6, 5, 2, 9, 14, 13, 10};
3041 static const UINT_32 BankXorLargeBpp
[] = {0, 7, 8, 15, 4, 3, 12, 11, 1, 6, 9, 14, 5, 2, 13, 10};
3043 bankXor
= (bpp
<= 32) ? BankXorSmallBpp
[index
] : BankXorLargeBpp
[index
];
3045 else if (bankBits
> 0)
3047 UINT_32 bankIncrease
= (1 << (bankBits
- 1)) - 1;
3048 bankIncrease
= (bankIncrease
== 0) ? 1 : bankIncrease
;
3049 bankXor
= (index
* bankIncrease
) & bankMask
;
3052 pOut
->pipeBankXor
= (bankXor
<< pipeBits
) | pipeXor
;
3056 pOut
->pipeBankXor
= 0;
3063 ************************************************************************************************************************
3064 * Gfx9Lib::HwlComputeSlicePipeBankXor
3067 * Generate slice PipeBankXor value based on base PipeBankXor value and slice id
3071 ************************************************************************************************************************
3073 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSlicePipeBankXor(
3074 const ADDR2_COMPUTE_SLICE_PIPEBANKXOR_INPUT
* pIn
,
3075 ADDR2_COMPUTE_SLICE_PIPEBANKXOR_OUTPUT
* pOut
) const
3077 UINT_32 macroBlockBits
= GetBlockSizeLog2(pIn
->swizzleMode
);
3078 UINT_32 pipeBits
= GetPipeXorBits(macroBlockBits
);
3079 UINT_32 bankBits
= GetBankXorBits(macroBlockBits
);
3081 UINT_32 pipeXor
= ReverseBitVector(pIn
->slice
, pipeBits
);
3082 UINT_32 bankXor
= ReverseBitVector(pIn
->slice
>> pipeBits
, bankBits
);
3084 pOut
->pipeBankXor
= pIn
->basePipeBankXor
^ (pipeXor
| (bankXor
<< pipeBits
));
3090 ************************************************************************************************************************
3091 * Gfx9Lib::HwlComputeSubResourceOffsetForSwizzlePattern
3094 * Compute sub resource offset to support swizzle pattern
3098 ************************************************************************************************************************
3100 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSubResourceOffsetForSwizzlePattern(
3101 const ADDR2_COMPUTE_SUBRESOURCE_OFFSET_FORSWIZZLEPATTERN_INPUT
* pIn
,
3102 ADDR2_COMPUTE_SUBRESOURCE_OFFSET_FORSWIZZLEPATTERN_OUTPUT
* pOut
) const
3104 ADDR_ASSERT(IsThin(pIn
->resourceType
, pIn
->swizzleMode
));
3106 UINT_32 macroBlockBits
= GetBlockSizeLog2(pIn
->swizzleMode
);
3107 UINT_32 pipeBits
= GetPipeXorBits(macroBlockBits
);
3108 UINT_32 bankBits
= GetBankXorBits(macroBlockBits
);
3109 UINT_32 pipeXor
= ReverseBitVector(pIn
->slice
, pipeBits
);
3110 UINT_32 bankXor
= ReverseBitVector(pIn
->slice
>> pipeBits
, bankBits
);
3111 UINT_32 pipeBankXor
= ((pipeXor
| (bankXor
<< pipeBits
)) ^ (pIn
->pipeBankXor
)) << m_pipeInterleaveLog2
;
3113 pOut
->offset
= pIn
->slice
* pIn
->sliceSize
+
3114 pIn
->macroBlockOffset
+
3115 (pIn
->mipTailOffset
^ pipeBankXor
) -
3116 static_cast<UINT_64
>(pipeBankXor
);
3121 ************************************************************************************************************************
3122 * Gfx9Lib::ValidateNonSwModeParams
3125 * Validate compute surface info params except swizzle mode
3128 * TRUE if parameters are valid, FALSE otherwise
3129 ************************************************************************************************************************
3131 BOOL_32
Gfx9Lib::ValidateNonSwModeParams(
3132 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
) const
3134 BOOL_32 valid
= TRUE
;
3136 if ((pIn
->bpp
== 0) || (pIn
->bpp
> 128) || (pIn
->width
== 0) || (pIn
->numFrags
> 8) || (pIn
->numSamples
> 16))
3138 ADDR_ASSERT_ALWAYS();
3142 if (pIn
->resourceType
>= ADDR_RSRC_MAX_TYPE
)
3144 ADDR_ASSERT_ALWAYS();
3148 const BOOL_32 mipmap
= (pIn
->numMipLevels
> 1);
3149 const BOOL_32 msaa
= (pIn
->numFrags
> 1);
3150 const BOOL_32 isBc
= ElemLib::IsBlockCompressed(pIn
->format
);
3152 const AddrResourceType rsrcType
= pIn
->resourceType
;
3153 const BOOL_32 tex3d
= IsTex3d(rsrcType
);
3154 const BOOL_32 tex2d
= IsTex2d(rsrcType
);
3155 const BOOL_32 tex1d
= IsTex1d(rsrcType
);
3157 const ADDR2_SURFACE_FLAGS flags
= pIn
->flags
;
3158 const BOOL_32 zbuffer
= flags
.depth
|| flags
.stencil
;
3159 const BOOL_32 display
= flags
.display
|| flags
.rotated
;
3160 const BOOL_32 stereo
= flags
.qbStereo
;
3161 const BOOL_32 fmask
= flags
.fmask
;
3163 // Resource type check
3166 if (msaa
|| zbuffer
|| display
|| stereo
|| isBc
|| fmask
)
3168 ADDR_ASSERT_ALWAYS();
3174 if ((msaa
&& mipmap
) || (stereo
&& msaa
) || (stereo
&& mipmap
))
3176 ADDR_ASSERT_ALWAYS();
3182 if (msaa
|| zbuffer
|| display
|| stereo
|| fmask
)
3184 ADDR_ASSERT_ALWAYS();
3190 ADDR_ASSERT_ALWAYS();
3198 ************************************************************************************************************************
3199 * Gfx9Lib::ValidateSwModeParams
3202 * Validate compute surface info related to swizzle mode
3205 * TRUE if parameters are valid, FALSE otherwise
3206 ************************************************************************************************************************
3208 BOOL_32
Gfx9Lib::ValidateSwModeParams(
3209 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
) const
3211 BOOL_32 valid
= TRUE
;
3213 if (pIn
->swizzleMode
>= ADDR_SW_MAX_TYPE
)
3215 ADDR_ASSERT_ALWAYS();
3219 const BOOL_32 mipmap
= (pIn
->numMipLevels
> 1);
3220 const BOOL_32 msaa
= (pIn
->numFrags
> 1);
3221 const BOOL_32 isBc
= ElemLib::IsBlockCompressed(pIn
->format
);
3222 const BOOL_32 is422
= ElemLib::IsMacroPixelPacked(pIn
->format
);
3224 const AddrResourceType rsrcType
= pIn
->resourceType
;
3225 const BOOL_32 tex3d
= IsTex3d(rsrcType
);
3226 const BOOL_32 tex2d
= IsTex2d(rsrcType
);
3227 const BOOL_32 tex1d
= IsTex1d(rsrcType
);
3229 const AddrSwizzleMode swizzle
= pIn
->swizzleMode
;
3230 const BOOL_32 linear
= IsLinear(swizzle
);
3231 const BOOL_32 blk256B
= IsBlock256b(swizzle
);
3232 const BOOL_32 blkVar
= IsBlockVariable(swizzle
);
3233 const BOOL_32 isNonPrtXor
= IsNonPrtXor(swizzle
);
3235 const ADDR2_SURFACE_FLAGS flags
= pIn
->flags
;
3236 const BOOL_32 zbuffer
= flags
.depth
|| flags
.stencil
;
3237 const BOOL_32 color
= flags
.color
;
3238 const BOOL_32 texture
= flags
.texture
;
3239 const BOOL_32 display
= flags
.display
|| flags
.rotated
;
3240 const BOOL_32 prt
= flags
.prt
;
3241 const BOOL_32 fmask
= flags
.fmask
;
3243 const BOOL_32 thin3d
= tex3d
&& flags
.view3dAs2dArray
;
3244 const BOOL_32 zMaxMip
= tex3d
&& mipmap
&&
3245 (pIn
->numSlices
>= pIn
->width
) && (pIn
->numSlices
>= pIn
->height
);
3248 if (msaa
&& (GetBlockSize(swizzle
) < (m_pipeInterleaveBytes
* pIn
->numFrags
)))
3250 // MSAA surface must have blk_bytes/pipe_interleave >= num_samples
3251 ADDR_ASSERT_ALWAYS();
3255 if (display
&& (IsValidDisplaySwizzleMode(pIn
) == FALSE
))
3257 ADDR_ASSERT_ALWAYS();
3261 if ((pIn
->bpp
== 96) && (linear
== FALSE
))
3263 ADDR_ASSERT_ALWAYS();
3267 if (prt
&& isNonPrtXor
)
3269 ADDR_ASSERT_ALWAYS();
3273 // Resource type check
3276 if (linear
== FALSE
)
3278 ADDR_ASSERT_ALWAYS();
3283 // Swizzle type check
3286 if (((tex1d
== FALSE
) && prt
) || zbuffer
|| msaa
|| (pIn
->bpp
== 0) ||
3287 ((pIn
->bpp
% 8) != 0) || (isBc
&& texture
) || fmask
)
3289 ADDR_ASSERT_ALWAYS();
3293 else if (IsZOrderSwizzle(swizzle
))
3295 if ((color
&& msaa
) || thin3d
|| isBc
|| is422
|| (tex2d
&& (pIn
->bpp
> 64)) || (msaa
&& (pIn
->bpp
> 32)))
3297 ADDR_ASSERT_ALWAYS();
3301 else if (IsStandardSwizzle(swizzle
))
3303 if (zbuffer
|| thin3d
|| (tex3d
&& (pIn
->bpp
== 128) && color
) || fmask
)
3305 ADDR_ASSERT_ALWAYS();
3309 else if (IsDisplaySwizzle(swizzle
))
3311 if (zbuffer
|| (prt
&& tex3d
) || fmask
|| zMaxMip
)
3313 ADDR_ASSERT_ALWAYS();
3317 else if (IsRotateSwizzle(swizzle
))
3319 if (zbuffer
|| (pIn
->bpp
> 64) || tex3d
|| isBc
|| fmask
)
3321 ADDR_ASSERT_ALWAYS();
3327 ADDR_ASSERT_ALWAYS();
3334 if (prt
|| zbuffer
|| tex3d
|| mipmap
|| msaa
)
3336 ADDR_ASSERT_ALWAYS();
3342 ADDR_ASSERT_ALWAYS();
3350 ************************************************************************************************************************
3351 * Gfx9Lib::HwlComputeSurfaceInfoSanityCheck
3354 * Compute surface info sanity check
3357 * ADDR_OK if parameters are valid, ADDR_INVALIDPARAMS otherwise
3358 ************************************************************************************************************************
3360 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSurfaceInfoSanityCheck(
3361 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
) const
3363 return ValidateNonSwModeParams(pIn
) && ValidateSwModeParams(pIn
) ? ADDR_OK
: ADDR_INVALIDPARAMS
;
3367 ************************************************************************************************************************
3368 * Gfx9Lib::HwlGetPreferredSurfaceSetting
3371 * Internal function to get suggested surface information for cliet to use
3375 ************************************************************************************************************************
3377 ADDR_E_RETURNCODE
Gfx9Lib::HwlGetPreferredSurfaceSetting(
3378 const ADDR2_GET_PREFERRED_SURF_SETTING_INPUT
* pIn
,
3379 ADDR2_GET_PREFERRED_SURF_SETTING_OUTPUT
* pOut
) const
3381 ADDR_E_RETURNCODE returnCode
= ADDR_INVALIDPARAMS
;
3382 ElemLib
* pElemLib
= GetElemLib();
3384 UINT_32 bpp
= pIn
->bpp
;
3385 UINT_32 width
= Max(pIn
->width
, 1u);
3386 UINT_32 height
= Max(pIn
->height
, 1u);
3387 UINT_32 numSamples
= Max(pIn
->numSamples
, 1u);
3388 UINT_32 numFrags
= (pIn
->numFrags
== 0) ? numSamples
: pIn
->numFrags
;
3390 if (pIn
->flags
.fmask
)
3392 bpp
= GetFmaskBpp(numSamples
, numFrags
);
3395 pOut
->resourceType
= ADDR_RSRC_TEX_2D
;
3399 // Set format to INVALID will skip this conversion
3400 if (pIn
->format
!= ADDR_FMT_INVALID
)
3402 UINT_32 expandX
, expandY
;
3404 // Don't care for this case
3405 ElemMode elemMode
= ADDR_UNCOMPRESSED
;
3407 // Get compression/expansion factors and element mode which indicates compression/expansion
3408 bpp
= pElemLib
->GetBitsPerPixel(pIn
->format
,
3413 UINT_32 basePitch
= 0;
3414 GetElemLib()->AdjustSurfaceInfo(elemMode
,
3423 // The output may get changed for volume(3D) texture resource in future
3424 pOut
->resourceType
= pIn
->resourceType
;
3427 const UINT_32 numSlices
= Max(pIn
->numSlices
, 1u);
3428 const UINT_32 numMipLevels
= Max(pIn
->numMipLevels
, 1u);
3429 const BOOL_32 msaa
= (numFrags
> 1) || (numSamples
> 1);
3430 const BOOL_32 displayRsrc
= pIn
->flags
.display
|| pIn
->flags
.rotated
;
3432 // Pre sanity check on non swizzle mode parameters
3433 ADDR2_COMPUTE_SURFACE_INFO_INPUT localIn
= {};
3434 localIn
.flags
= pIn
->flags
;
3435 localIn
.resourceType
= pOut
->resourceType
;
3436 localIn
.format
= pIn
->format
;
3438 localIn
.width
= width
;
3439 localIn
.height
= height
;
3440 localIn
.numSlices
= numSlices
;
3441 localIn
.numMipLevels
= numMipLevels
;
3442 localIn
.numSamples
= numSamples
;
3443 localIn
.numFrags
= numFrags
;
3445 if (ValidateNonSwModeParams(&localIn
))
3447 // Forbid swizzle mode(s) by client setting, for simplicity we never allow VAR swizzle mode for GFX9
3448 ADDR2_SWMODE_SET allowedSwModeSet
= {};
3449 allowedSwModeSet
.value
|= pIn
->forbiddenBlock
.linear
? 0 : Gfx9LinearSwModeMask
;
3450 allowedSwModeSet
.value
|= pIn
->forbiddenBlock
.micro
? 0 : Gfx9Blk256BSwModeMask
;
3451 allowedSwModeSet
.value
|= pIn
->forbiddenBlock
.macro4KB
? 0 : Gfx9Blk4KBSwModeMask
;
3452 allowedSwModeSet
.value
|= pIn
->forbiddenBlock
.macro64KB
? 0 : Gfx9Blk64KBSwModeMask
;
3454 if (pIn
->preferredSwSet
.value
!= 0)
3456 allowedSwModeSet
.value
&= pIn
->preferredSwSet
.sw_Z
? ~0 : ~Gfx9ZSwModeMask
;
3457 allowedSwModeSet
.value
&= pIn
->preferredSwSet
.sw_S
? ~0 : ~Gfx9StandardSwModeMask
;
3458 allowedSwModeSet
.value
&= pIn
->preferredSwSet
.sw_D
? ~0 : ~Gfx9DisplaySwModeMask
;
3459 allowedSwModeSet
.value
&= pIn
->preferredSwSet
.sw_R
? ~0 : ~Gfx9RotateSwModeMask
;
3464 allowedSwModeSet
.value
&= ~Gfx9XorSwModeMask
;
3467 if (pIn
->maxAlign
> 0)
3469 if (pIn
->maxAlign
< GetBlockSize(ADDR_SW_64KB
))
3471 allowedSwModeSet
.value
&= ~Gfx9Blk64KBSwModeMask
;
3474 if (pIn
->maxAlign
< GetBlockSize(ADDR_SW_4KB
))
3476 allowedSwModeSet
.value
&= ~Gfx9Blk4KBSwModeMask
;
3479 if (pIn
->maxAlign
< GetBlockSize(ADDR_SW_256B
))
3481 allowedSwModeSet
.value
&= ~Gfx9Blk256BSwModeMask
;
3485 // Filter out invalid swizzle mode(s) by image attributes and HW restrictions
3486 switch (pOut
->resourceType
)
3488 case ADDR_RSRC_TEX_1D
:
3489 allowedSwModeSet
.value
&= Gfx9Rsrc1dSwModeMask
;
3492 case ADDR_RSRC_TEX_2D
:
3493 allowedSwModeSet
.value
&= pIn
->flags
.prt
? Gfx9Rsrc2dPrtSwModeMask
: Gfx9Rsrc2dSwModeMask
;
3497 allowedSwModeSet
.value
&= ~(Gfx9RotateSwModeMask
| Gfx9ZSwModeMask
);
3501 case ADDR_RSRC_TEX_3D
:
3502 allowedSwModeSet
.value
&= pIn
->flags
.prt
? Gfx9Rsrc3dPrtSwModeMask
: Gfx9Rsrc3dSwModeMask
;
3504 if ((numMipLevels
> 1) && (numSlices
>= width
) && (numSlices
>= height
))
3506 // SW_*_D for 3D mipmaps (maxmip > 0) is only supported for Xmajor or Ymajor mipmap
3507 // When depth (Z) is the maximum dimension then must use one of the SW_*_S
3508 // or SW_*_Z modes if mipmapping is desired on a 3D surface
3509 allowedSwModeSet
.value
&= ~Gfx9DisplaySwModeMask
;
3512 if ((bpp
== 128) && pIn
->flags
.color
)
3514 allowedSwModeSet
.value
&= ~Gfx9StandardSwModeMask
;
3517 if (pIn
->flags
.view3dAs2dArray
)
3519 allowedSwModeSet
.value
&= Gfx9Rsrc3dThinSwModeMask
| Gfx9LinearSwModeMask
;
3524 ADDR_ASSERT_ALWAYS();
3525 allowedSwModeSet
.value
= 0;
3529 if (pIn
->format
== ADDR_FMT_32_32_32
)
3531 allowedSwModeSet
.value
&= Gfx9LinearSwModeMask
;
3534 if (ElemLib::IsBlockCompressed(pIn
->format
))
3536 if (pIn
->flags
.texture
)
3538 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
| Gfx9DisplaySwModeMask
;
3542 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
| Gfx9DisplaySwModeMask
| Gfx9LinearSwModeMask
;
3546 if (ElemLib::IsMacroPixelPacked(pIn
->format
) ||
3547 (msaa
&& ((bpp
> 32) || pIn
->flags
.color
|| pIn
->flags
.unordered
)))
3549 allowedSwModeSet
.value
&= ~Gfx9ZSwModeMask
;
3552 if (pIn
->flags
.fmask
|| pIn
->flags
.depth
|| pIn
->flags
.stencil
)
3554 allowedSwModeSet
.value
&= Gfx9ZSwModeMask
;
3556 if (pIn
->flags
.noMetadata
== FALSE
)
3558 if (pIn
->flags
.depth
&&
3559 pIn
->flags
.texture
&&
3560 (((bpp
== 16) && (numFrags
>= 4)) || ((bpp
== 32) && (numFrags
>= 2))))
3562 // When _X/_T swizzle mode was used for MSAA depth texture, TC will get zplane
3563 // equation from wrong address within memory range a tile covered and use the
3564 // garbage data for compressed Z reading which finally leads to corruption.
3565 allowedSwModeSet
.value
&= ~Gfx9XorSwModeMask
;
3568 if (m_settings
.htileCacheRbConflict
&&
3569 (pIn
->flags
.depth
|| pIn
->flags
.stencil
) &&
3571 (pIn
->flags
.metaRbUnaligned
== FALSE
) &&
3572 (pIn
->flags
.metaPipeUnaligned
== FALSE
))
3574 // Z_X 2D array with Rb/Pipe aligned HTile won't have metadata cache coherency
3575 allowedSwModeSet
.value
&= ~Gfx9XSwModeMask
;
3582 allowedSwModeSet
.value
&= Gfx9MsaaSwModeMask
;
3585 if ((numFrags
> 1) &&
3586 (GetBlockSize(ADDR_SW_4KB
) < (m_pipeInterleaveBytes
* numFrags
)))
3588 // MSAA surface must have blk_bytes/pipe_interleave >= num_samples
3589 allowedSwModeSet
.value
&= Gfx9Blk64KBSwModeMask
;
3592 if (numMipLevels
> 1)
3594 allowedSwModeSet
.value
&= ~Gfx9Blk256BSwModeMask
;
3599 if (m_settings
.isDce12
)
3601 allowedSwModeSet
.value
&= (bpp
== 32) ? Dce12Bpp32SwModeMask
: Dce12NonBpp32SwModeMask
;
3603 else if (m_settings
.isDcn1
)
3605 allowedSwModeSet
.value
&= (bpp
== 64) ? Dcn1Bpp64SwModeMask
: Dcn1NonBpp64SwModeMask
;
3609 ADDR_NOT_IMPLEMENTED();
3613 if (allowedSwModeSet
.value
!= 0)
3616 // Post sanity check, at least AddrLib should accept the output generated by its own
3617 UINT_32 validateSwModeSet
= allowedSwModeSet
.value
;
3619 for (UINT_32 i
= 0; validateSwModeSet
!= 0; i
++)
3621 if (validateSwModeSet
& 1)
3623 localIn
.swizzleMode
= static_cast<AddrSwizzleMode
>(i
);
3624 ADDR_ASSERT(ValidateSwModeParams(&localIn
));
3627 validateSwModeSet
>>= 1;
3631 pOut
->validSwModeSet
= allowedSwModeSet
;
3632 pOut
->canXor
= (allowedSwModeSet
.value
& Gfx9XorSwModeMask
) ? TRUE
: FALSE
;
3633 pOut
->validBlockSet
= GetAllowedBlockSet(allowedSwModeSet
);
3634 pOut
->validSwTypeSet
= GetAllowedSwSet(allowedSwModeSet
);
3636 pOut
->clientPreferredSwSet
= pIn
->preferredSwSet
;
3638 if (pOut
->clientPreferredSwSet
.value
== 0)
3640 pOut
->clientPreferredSwSet
.value
= AddrSwSetAll
;
3643 if (allowedSwModeSet
.value
== Gfx9LinearSwModeMask
)
3645 pOut
->swizzleMode
= ADDR_SW_LINEAR
;
3649 // Always ignore linear swizzle mode if there is other choice.
3650 allowedSwModeSet
.swLinear
= 0;
3652 ADDR2_BLOCK_SET allowedBlockSet
= GetAllowedBlockSet(allowedSwModeSet
);
3654 // Determine block size if there is 2 or more block type candidates
3655 if (IsPow2(allowedBlockSet
.value
) == FALSE
)
3657 const AddrSwizzleMode swMode
[AddrBlockMaxTiledType
] = {ADDR_SW_256B
, ADDR_SW_4KB
, ADDR_SW_64KB
};
3658 Dim3d blkDim
[AddrBlockMaxTiledType
] = {{0}, {0}, {0}};
3659 Dim3d padDim
[AddrBlockMaxTiledType
] = {{0}, {0}, {0}};
3660 UINT_64 padSize
[AddrBlockMaxTiledType
] = {0};
3662 const UINT_32 ratioLow
= pIn
->flags
.minimizeAlign
? 1 : (pIn
->flags
.opt4space
? 3 : 2);
3663 const UINT_32 ratioHi
= pIn
->flags
.minimizeAlign
? 1 : (pIn
->flags
.opt4space
? 2 : 1);
3664 const UINT_64 sizeAlignInElement
= Max(NextPow2(pIn
->minSizeAlign
) / (bpp
>> 3), 1u);
3665 UINT_32 minSizeBlk
= AddrBlockMicro
;
3666 UINT_64 minSize
= 0;
3668 for (UINT_32 i
= AddrBlockMicro
; i
< AddrBlockMaxTiledType
; i
++)
3670 if (allowedBlockSet
.value
& (1 << i
))
3672 ComputeBlockDimensionForSurf(&blkDim
[i
].w
,
3682 blkDim
[i
].w
= PowTwoAlign(blkDim
[i
].w
, 32);
3685 padSize
[i
] = ComputePadSize(&blkDim
[i
], width
, height
, numSlices
, &padDim
[i
]);
3686 padSize
[i
] = PowTwoAlign(padSize
[i
], sizeAlignInElement
);
3688 if ((minSize
== 0) ||
3689 ((padSize
[i
] * ratioHi
) <= (minSize
* ratioLow
)))
3691 minSize
= padSize
[i
];
3697 if ((allowedBlockSet
.micro
== TRUE
) &&
3698 (width
<= blkDim
[AddrBlockMicro
].w
) &&
3699 (height
<= blkDim
[AddrBlockMicro
].h
) &&
3700 (NextPow2(pIn
->minSizeAlign
) <= GetBlockSize(ADDR_SW_256B
)))
3702 minSizeBlk
= AddrBlockMicro
;
3705 if (minSizeBlk
== AddrBlockMicro
)
3707 allowedSwModeSet
.value
&= Gfx9Blk256BSwModeMask
;
3709 else if (minSizeBlk
== AddrBlock4KB
)
3711 allowedSwModeSet
.value
&= Gfx9Blk4KBSwModeMask
;
3715 ADDR_ASSERT(minSizeBlk
== AddrBlock64KB
);
3716 allowedSwModeSet
.value
&= Gfx9Blk64KBSwModeMask
;
3720 // Block type should be determined.
3721 ADDR_ASSERT(IsPow2(GetAllowedBlockSet(allowedSwModeSet
).value
));
3723 ADDR2_SWTYPE_SET allowedSwSet
= GetAllowedSwSet(allowedSwModeSet
);
3725 // Determine swizzle type if there is 2 or more swizzle type candidates
3726 if (IsPow2(allowedSwSet
.value
) == FALSE
)
3728 if (ElemLib::IsBlockCompressed(pIn
->format
))
3730 if (allowedSwSet
.sw_D
)
3732 allowedSwModeSet
.value
&= Gfx9DisplaySwModeMask
;
3736 ADDR_ASSERT(allowedSwSet
.sw_S
);
3737 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
;
3740 else if (ElemLib::IsMacroPixelPacked(pIn
->format
))
3742 if (allowedSwSet
.sw_S
)
3744 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
;
3746 else if (allowedSwSet
.sw_D
)
3748 allowedSwModeSet
.value
&= Gfx9DisplaySwModeMask
;
3752 ADDR_ASSERT(allowedSwSet
.sw_R
);
3753 allowedSwModeSet
.value
&= Gfx9RotateSwModeMask
;
3756 else if (pOut
->resourceType
== ADDR_RSRC_TEX_3D
)
3758 if (pIn
->flags
.color
&& allowedSwSet
.sw_D
)
3760 allowedSwModeSet
.value
&= Gfx9DisplaySwModeMask
;
3762 else if (allowedSwSet
.sw_Z
)
3764 allowedSwModeSet
.value
&= Gfx9ZSwModeMask
;
3768 ADDR_ASSERT(allowedSwSet
.sw_S
);
3769 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
;
3774 if (pIn
->flags
.rotated
&& allowedSwSet
.sw_R
)
3776 allowedSwModeSet
.value
&= Gfx9RotateSwModeMask
;
3778 else if (displayRsrc
&& allowedSwSet
.sw_D
)
3780 allowedSwModeSet
.value
&= Gfx9DisplaySwModeMask
;
3782 else if (allowedSwSet
.sw_S
)
3784 allowedSwModeSet
.value
&= Gfx9StandardSwModeMask
;
3788 ADDR_ASSERT(allowedSwSet
.sw_Z
);
3789 allowedSwModeSet
.value
&= Gfx9ZSwModeMask
;
3794 // Swizzle type should be determined.
3795 ADDR_ASSERT(IsPow2(GetAllowedSwSet(allowedSwModeSet
).value
));
3797 // Determine swizzle mode now - always select the "largest" swizzle mode for a given block type +
3798 // swizzle type combination. For example, for AddrBlock64KB + ADDR_SW_S, select SW_64KB_S_X(25) if it's
3799 // available, or otherwise select SW_64KB_S_T(17) if it's available, or otherwise select SW_64KB_S(9).
3800 pOut
->swizzleMode
= static_cast<AddrSwizzleMode
>(Log2NonPow2(allowedSwModeSet
.value
));
3803 returnCode
= ADDR_OK
;
3807 // Invalid combination...
3808 ADDR_ASSERT_ALWAYS();
3813 // Invalid combination...
3814 ADDR_ASSERT_ALWAYS();
3821 ************************************************************************************************************************
3822 * Gfx9Lib::ComputeStereoInfo
3825 * Compute height alignment and right eye pipeBankXor for stereo surface
3830 ************************************************************************************************************************
3832 ADDR_E_RETURNCODE
Gfx9Lib::ComputeStereoInfo(
3833 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
,
3834 ADDR2_COMPUTE_SURFACE_INFO_OUTPUT
* pOut
,
3835 UINT_32
* pHeightAlign
3838 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
3840 UINT_32 eqIndex
= HwlGetEquationIndex(pIn
, pOut
);
3842 if (eqIndex
< m_numEquations
)
3844 if (IsXor(pIn
->swizzleMode
))
3846 const UINT_32 blkSizeLog2
= GetBlockSizeLog2(pIn
->swizzleMode
);
3847 const UINT_32 numPipeBits
= GetPipeXorBits(blkSizeLog2
);
3848 const UINT_32 numBankBits
= GetBankXorBits(blkSizeLog2
);
3849 const UINT_32 bppLog2
= Log2(pIn
->bpp
>> 3);
3850 const UINT_32 maxYCoordBlock256
= Log2(Block256_2d
[bppLog2
].h
) - 1;
3851 ASSERTED
const ADDR_EQUATION
*pEqToCheck
= &m_equationTable
[eqIndex
];
3853 ADDR_ASSERT(maxYCoordBlock256
==
3854 GetMaxValidChannelIndex(&pEqToCheck
->addr
[0], GetBlockSizeLog2(ADDR_SW_256B
), 1));
3856 const UINT_32 maxYCoordInBaseEquation
=
3857 (blkSizeLog2
- GetBlockSizeLog2(ADDR_SW_256B
)) / 2 + maxYCoordBlock256
;
3859 ADDR_ASSERT(maxYCoordInBaseEquation
==
3860 GetMaxValidChannelIndex(&pEqToCheck
->addr
[0], blkSizeLog2
, 1));
3862 const UINT_32 maxYCoordInPipeXor
= (numPipeBits
== 0) ? 0 : maxYCoordBlock256
+ numPipeBits
;
3864 ADDR_ASSERT(maxYCoordInPipeXor
==
3865 GetMaxValidChannelIndex(&pEqToCheck
->xor1
[m_pipeInterleaveLog2
], numPipeBits
, 1));
3867 const UINT_32 maxYCoordInBankXor
= (numBankBits
== 0) ?
3868 0 : maxYCoordBlock256
+ (numPipeBits
+ 1) / 2 + numBankBits
;
3870 ADDR_ASSERT(maxYCoordInBankXor
==
3871 GetMaxValidChannelIndex(&pEqToCheck
->xor1
[m_pipeInterleaveLog2
+ numPipeBits
], numBankBits
, 1));
3873 const UINT_32 maxYCoordInPipeBankXor
= Max(maxYCoordInPipeXor
, maxYCoordInBankXor
);
3875 if (maxYCoordInPipeBankXor
> maxYCoordInBaseEquation
)
3877 *pHeightAlign
= 1u << maxYCoordInPipeBankXor
;
3879 if (pOut
->pStereoInfo
!= NULL
)
3881 pOut
->pStereoInfo
->rightSwizzle
= 0;
3883 if ((PowTwoAlign(pIn
->height
, *pHeightAlign
) % (*pHeightAlign
* 2)) != 0)
3885 if (maxYCoordInPipeXor
== maxYCoordInPipeBankXor
)
3887 pOut
->pStereoInfo
->rightSwizzle
|= (1u << 1);
3890 if (maxYCoordInBankXor
== maxYCoordInPipeBankXor
)
3892 pOut
->pStereoInfo
->rightSwizzle
|=
3893 1u << ((numPipeBits
% 2) ? numPipeBits
: numPipeBits
+ 1);
3896 ADDR_ASSERT(pOut
->pStereoInfo
->rightSwizzle
==
3897 GetCoordActiveMask(&pEqToCheck
->xor1
[m_pipeInterleaveLog2
],
3898 numPipeBits
+ numBankBits
, 1, maxYCoordInPipeBankXor
));
3906 ADDR_ASSERT_ALWAYS();
3907 returnCode
= ADDR_ERROR
;
3914 ************************************************************************************************************************
3915 * Gfx9Lib::HwlComputeSurfaceInfoTiled
3918 * Internal function to calculate alignment for tiled surface
3922 ************************************************************************************************************************
3924 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSurfaceInfoTiled(
3925 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
, ///< [in] input structure
3926 ADDR2_COMPUTE_SURFACE_INFO_OUTPUT
* pOut
///< [out] output structure
3929 ADDR_E_RETURNCODE returnCode
= ComputeBlockDimensionForSurf(&pOut
->blockWidth
,
3937 if (returnCode
== ADDR_OK
)
3939 UINT_32 pitchAlignInElement
= pOut
->blockWidth
;
3941 if ((IsTex2d(pIn
->resourceType
) == TRUE
) &&
3942 (pIn
->flags
.display
|| pIn
->flags
.rotated
) &&
3943 (pIn
->numMipLevels
<= 1) &&
3944 (pIn
->numSamples
<= 1) &&
3945 (pIn
->numFrags
<= 1))
3947 // Display engine needs pitch align to be at least 32 pixels.
3948 pitchAlignInElement
= PowTwoAlign(pitchAlignInElement
, 32);
3951 pOut
->pitch
= PowTwoAlign(pIn
->width
, pitchAlignInElement
);
3953 if ((pIn
->numMipLevels
<= 1) && (pIn
->pitchInElement
> 0))
3955 if ((pIn
->pitchInElement
% pitchAlignInElement
) != 0)
3957 returnCode
= ADDR_INVALIDPARAMS
;
3959 else if (pIn
->pitchInElement
< pOut
->pitch
)
3961 returnCode
= ADDR_INVALIDPARAMS
;
3965 pOut
->pitch
= pIn
->pitchInElement
;
3969 UINT_32 heightAlign
= 0;
3971 if (pIn
->flags
.qbStereo
)
3973 returnCode
= ComputeStereoInfo(pIn
, pOut
, &heightAlign
);
3976 if (returnCode
== ADDR_OK
)
3978 pOut
->height
= PowTwoAlign(pIn
->height
, pOut
->blockHeight
);
3980 if (heightAlign
> 1)
3982 pOut
->height
= PowTwoAlign(pOut
->height
, heightAlign
);
3985 pOut
->numSlices
= PowTwoAlign(pIn
->numSlices
, pOut
->blockSlices
);
3987 pOut
->epitchIsHeight
= FALSE
;
3988 pOut
->mipChainInTail
= FALSE
;
3989 pOut
->firstMipIdInTail
= pIn
->numMipLevels
;
3991 pOut
->mipChainPitch
= pOut
->pitch
;
3992 pOut
->mipChainHeight
= pOut
->height
;
3993 pOut
->mipChainSlice
= pOut
->numSlices
;
3995 if (pIn
->numMipLevels
> 1)
3997 pOut
->firstMipIdInTail
= GetMipChainInfo(pIn
->resourceType
,
4009 const UINT_32 endingMipId
= Min(pOut
->firstMipIdInTail
, pIn
->numMipLevels
- 1);
4011 if (endingMipId
== 0)
4013 const Dim3d tailMaxDim
= GetMipTailDim(pIn
->resourceType
,
4019 pOut
->epitchIsHeight
= TRUE
;
4020 pOut
->pitch
= tailMaxDim
.w
;
4021 pOut
->height
= tailMaxDim
.h
;
4022 pOut
->numSlices
= IsThick(pIn
->resourceType
, pIn
->swizzleMode
) ?
4023 tailMaxDim
.d
: pIn
->numSlices
;
4024 pOut
->mipChainInTail
= TRUE
;
4028 UINT_32 mip0WidthInBlk
= pOut
->pitch
/ pOut
->blockWidth
;
4029 UINT_32 mip0HeightInBlk
= pOut
->height
/ pOut
->blockHeight
;
4031 AddrMajorMode majorMode
= GetMajorMode(pIn
->resourceType
,
4035 pOut
->numSlices
/ pOut
->blockSlices
);
4036 if (majorMode
== ADDR_MAJOR_Y
)
4038 UINT_32 mip1WidthInBlk
= RoundHalf(mip0WidthInBlk
);
4040 if ((mip1WidthInBlk
== 1) && (endingMipId
> 2))
4045 pOut
->mipChainPitch
+= (mip1WidthInBlk
* pOut
->blockWidth
);
4047 pOut
->epitchIsHeight
= FALSE
;
4051 UINT_32 mip1HeightInBlk
= RoundHalf(mip0HeightInBlk
);
4053 if ((mip1HeightInBlk
== 1) && (endingMipId
> 2))
4058 pOut
->mipChainHeight
+= (mip1HeightInBlk
* pOut
->blockHeight
);
4060 pOut
->epitchIsHeight
= TRUE
;
4064 if (pOut
->pMipInfo
!= NULL
)
4066 UINT_32 elementBytesLog2
= Log2(pIn
->bpp
>> 3);
4068 for (UINT_32 i
= 0; i
< pIn
->numMipLevels
; i
++)
4070 Dim3d mipStartPos
= {0};
4071 UINT_32 mipTailOffsetInBytes
= 0;
4073 mipStartPos
= GetMipStartPos(pIn
->resourceType
,
4083 &mipTailOffsetInBytes
);
4085 UINT_32 pitchInBlock
=
4086 pOut
->mipChainPitch
/ pOut
->blockWidth
;
4087 UINT_32 sliceInBlock
=
4088 (pOut
->mipChainHeight
/ pOut
->blockHeight
) * pitchInBlock
;
4089 UINT_64 blockIndex
=
4090 mipStartPos
.d
* sliceInBlock
+ mipStartPos
.h
* pitchInBlock
+ mipStartPos
.w
;
4091 UINT_64 macroBlockOffset
=
4092 blockIndex
<< GetBlockSizeLog2(pIn
->swizzleMode
);
4094 pOut
->pMipInfo
[i
].macroBlockOffset
= macroBlockOffset
;
4095 pOut
->pMipInfo
[i
].mipTailOffset
= mipTailOffsetInBytes
;
4099 else if (pOut
->pMipInfo
!= NULL
)
4101 pOut
->pMipInfo
[0].pitch
= pOut
->pitch
;
4102 pOut
->pMipInfo
[0].height
= pOut
->height
;
4103 pOut
->pMipInfo
[0].depth
= IsTex3d(pIn
->resourceType
)? pOut
->numSlices
: 1;
4104 pOut
->pMipInfo
[0].offset
= 0;
4107 pOut
->sliceSize
= static_cast<UINT_64
>(pOut
->mipChainPitch
) * pOut
->mipChainHeight
*
4108 (pIn
->bpp
>> 3) * pIn
->numFrags
;
4109 pOut
->surfSize
= pOut
->sliceSize
* pOut
->mipChainSlice
;
4110 pOut
->baseAlign
= ComputeSurfaceBaseAlignTiled(pIn
->swizzleMode
);
4112 if ((IsBlock256b(pIn
->swizzleMode
) == FALSE
) &&
4113 (pIn
->flags
.color
|| pIn
->flags
.depth
|| pIn
->flags
.stencil
|| pIn
->flags
.fmask
) &&
4114 (pIn
->flags
.texture
== TRUE
) &&
4115 (pIn
->flags
.noMetadata
== FALSE
) &&
4116 (pIn
->flags
.metaPipeUnaligned
== FALSE
))
4118 // Assume client requires pipe aligned metadata, which is TcCompatible and will be accessed by TC...
4119 // Then we need extra padding for base surface. Otherwise, metadata and data surface for same pixel will
4120 // be flushed to different pipes, but texture engine only uses pipe id of data surface to fetch both of
4121 // them, which may cause invalid metadata to be fetched.
4122 pOut
->baseAlign
= Max(pOut
->baseAlign
, m_pipeInterleaveBytes
* m_pipes
* m_se
);
4127 pOut
->baseAlign
= Max(pOut
->baseAlign
, PrtAlignment
);
4136 ************************************************************************************************************************
4137 * Gfx9Lib::HwlComputeSurfaceInfoLinear
4140 * Internal function to calculate alignment for linear surface
4144 ************************************************************************************************************************
4146 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSurfaceInfoLinear(
4147 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
, ///< [in] input structure
4148 ADDR2_COMPUTE_SURFACE_INFO_OUTPUT
* pOut
///< [out] output structure
4151 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
4153 UINT_32 actualHeight
= 0;
4154 UINT_32 elementBytes
= pIn
->bpp
>> 3;
4155 const UINT_32 alignment
= pIn
->flags
.prt
? PrtAlignment
: 256;
4157 if (IsTex1d(pIn
->resourceType
))
4159 if (pIn
->height
> 1)
4161 returnCode
= ADDR_INVALIDPARAMS
;
4165 const UINT_32 pitchAlignInElement
= alignment
/ elementBytes
;
4167 pitch
= PowTwoAlign(pIn
->width
, pitchAlignInElement
);
4168 actualHeight
= pIn
->numMipLevels
;
4170 if (pIn
->flags
.prt
== FALSE
)
4172 returnCode
= ApplyCustomizedPitchHeight(pIn
, elementBytes
, pitchAlignInElement
,
4173 &pitch
, &actualHeight
);
4176 if (returnCode
== ADDR_OK
)
4178 if (pOut
->pMipInfo
!= NULL
)
4180 for (UINT_32 i
= 0; i
< pIn
->numMipLevels
; i
++)
4182 pOut
->pMipInfo
[i
].offset
= pitch
* elementBytes
* i
;
4183 pOut
->pMipInfo
[i
].pitch
= pitch
;
4184 pOut
->pMipInfo
[i
].height
= 1;
4185 pOut
->pMipInfo
[i
].depth
= 1;
4193 returnCode
= ComputeSurfaceLinearPadding(pIn
, &pitch
, &actualHeight
, pOut
->pMipInfo
);
4196 if ((pitch
== 0) || (actualHeight
== 0))
4198 returnCode
= ADDR_INVALIDPARAMS
;
4201 if (returnCode
== ADDR_OK
)
4203 pOut
->pitch
= pitch
;
4204 pOut
->height
= pIn
->height
;
4205 pOut
->numSlices
= pIn
->numSlices
;
4206 pOut
->mipChainPitch
= pitch
;
4207 pOut
->mipChainHeight
= actualHeight
;
4208 pOut
->mipChainSlice
= pOut
->numSlices
;
4209 pOut
->epitchIsHeight
= (pIn
->numMipLevels
> 1) ? TRUE
: FALSE
;
4210 pOut
->sliceSize
= static_cast<UINT_64
>(pOut
->pitch
) * actualHeight
* elementBytes
;
4211 pOut
->surfSize
= pOut
->sliceSize
* pOut
->numSlices
;
4212 pOut
->baseAlign
= (pIn
->swizzleMode
== ADDR_SW_LINEAR_GENERAL
) ? (pIn
->bpp
/ 8) : alignment
;
4213 pOut
->blockWidth
= (pIn
->swizzleMode
== ADDR_SW_LINEAR_GENERAL
) ? 1 : (256 / elementBytes
);
4214 pOut
->blockHeight
= 1;
4215 pOut
->blockSlices
= 1;
4218 // Post calculation validate
4219 ADDR_ASSERT(pOut
->sliceSize
> 0);
4225 ************************************************************************************************************************
4226 * Gfx9Lib::GetMipChainInfo
4229 * Internal function to get out information about mip chain
4232 * Smaller value between Id of first mip fitted in mip tail and max Id of mip being created
4233 ************************************************************************************************************************
4235 UINT_32
Gfx9Lib::GetMipChainInfo(
4236 AddrResourceType resourceType
,
4237 AddrSwizzleMode swizzleMode
,
4243 UINT_32 blockHeight
,
4245 UINT_32 numMipLevel
,
4246 ADDR2_MIP_INFO
* pMipInfo
) const
4248 const Dim3d tailMaxDim
=
4249 GetMipTailDim(resourceType
, swizzleMode
, blockWidth
, blockHeight
, blockDepth
);
4251 UINT_32 mipPitch
= mip0Width
;
4252 UINT_32 mipHeight
= mip0Height
;
4253 UINT_32 mipDepth
= IsTex3d(resourceType
) ? mip0Depth
: 1;
4255 UINT_32 firstMipIdInTail
= numMipLevel
;
4256 BOOL_32 inTail
= FALSE
;
4257 BOOL_32 finalDim
= FALSE
;
4258 BOOL_32 is3dThick
= IsThick(resourceType
, swizzleMode
);
4259 BOOL_32 is3dThin
= IsTex3d(resourceType
) && (is3dThick
== FALSE
);
4261 for (UINT_32 mipId
= 0; mipId
< numMipLevel
; mipId
++)
4265 if (finalDim
== FALSE
)
4271 mipSize
= mipPitch
* mipHeight
* mipDepth
* (bpp
>> 3);
4275 mipSize
= mipPitch
* mipHeight
* (bpp
>> 3);
4280 UINT_32 index
= Log2(bpp
>> 3);
4284 mipPitch
= Block256_3dZ
[index
].w
;
4285 mipHeight
= Block256_3dZ
[index
].h
;
4286 mipDepth
= Block256_3dZ
[index
].d
;
4290 mipPitch
= Block256_2d
[index
].w
;
4291 mipHeight
= Block256_2d
[index
].h
;
4300 inTail
= IsInMipTail(resourceType
, swizzleMode
, tailMaxDim
,
4301 mipPitch
, mipHeight
, mipDepth
);
4305 firstMipIdInTail
= mipId
;
4306 mipPitch
= tailMaxDim
.w
;
4307 mipHeight
= tailMaxDim
.h
;
4311 mipDepth
= tailMaxDim
.d
;
4316 mipPitch
= PowTwoAlign(mipPitch
, blockWidth
);
4317 mipHeight
= PowTwoAlign(mipHeight
, blockHeight
);
4321 mipDepth
= PowTwoAlign(mipDepth
, blockDepth
);
4326 if (pMipInfo
!= NULL
)
4328 pMipInfo
[mipId
].pitch
= mipPitch
;
4329 pMipInfo
[mipId
].height
= mipHeight
;
4330 pMipInfo
[mipId
].depth
= mipDepth
;
4331 pMipInfo
[mipId
].offset
= offset
;
4334 offset
+= (mipPitch
* mipHeight
* mipDepth
* (bpp
>> 3));
4340 mipDepth
= Max(mipDepth
>> 1, 1u);
4345 mipPitch
= Max(mipPitch
>> 1, 1u);
4346 mipHeight
= Max(mipHeight
>> 1, 1u);
4348 if (is3dThick
|| is3dThin
)
4350 mipDepth
= Max(mipDepth
>> 1, 1u);
4355 return firstMipIdInTail
;
4359 ************************************************************************************************************************
4360 * Gfx9Lib::GetMetaMiptailInfo
4363 * Get mip tail coordinate information.
4367 ************************************************************************************************************************
4369 VOID
Gfx9Lib::GetMetaMiptailInfo(
4370 ADDR2_META_MIP_INFO
* pInfo
, ///< [out] output structure to store per mip coord
4371 Dim3d mipCoord
, ///< [in] mip tail base coord
4372 UINT_32 numMipInTail
, ///< [in] number of mips in tail
4373 Dim3d
* pMetaBlkDim
///< [in] meta block width/height/depth
4376 BOOL_32 isThick
= (pMetaBlkDim
->d
> 1);
4377 UINT_32 mipWidth
= pMetaBlkDim
->w
;
4378 UINT_32 mipHeight
= pMetaBlkDim
->h
>> 1;
4379 UINT_32 mipDepth
= pMetaBlkDim
->d
;
4384 minInc
= (pMetaBlkDim
->h
>= 512) ? 128 : ((pMetaBlkDim
->h
== 256) ? 64 : 32);
4386 else if (pMetaBlkDim
->h
>= 1024)
4390 else if (pMetaBlkDim
->h
== 512)
4399 UINT_32 blk32MipId
= 0xFFFFFFFF;
4401 for (UINT_32 mip
= 0; mip
< numMipInTail
; mip
++)
4403 pInfo
[mip
].inMiptail
= TRUE
;
4404 pInfo
[mip
].startX
= mipCoord
.w
;
4405 pInfo
[mip
].startY
= mipCoord
.h
;
4406 pInfo
[mip
].startZ
= mipCoord
.d
;
4407 pInfo
[mip
].width
= mipWidth
;
4408 pInfo
[mip
].height
= mipHeight
;
4409 pInfo
[mip
].depth
= mipDepth
;
4413 if (blk32MipId
== 0xFFFFFFFF)
4418 mipCoord
.w
= pInfo
[blk32MipId
].startX
;
4419 mipCoord
.h
= pInfo
[blk32MipId
].startY
;
4420 mipCoord
.d
= pInfo
[blk32MipId
].startZ
;
4422 switch (mip
- blk32MipId
)
4425 mipCoord
.w
+= 32; // 16x16
4428 mipCoord
.h
+= 32; // 8x8
4431 mipCoord
.h
+= 32; // 4x4
4435 mipCoord
.h
+= 32; // 2x2
4439 mipCoord
.h
+= 32; // 1x1
4442 // The following are for BC/ASTC formats
4444 mipCoord
.h
+= 48; // 1/2 x 1/2
4447 mipCoord
.h
+= 48; // 1/4 x 1/4
4451 mipCoord
.h
+= 48; // 1/8 x 1/8
4455 mipCoord
.h
+= 48; // 1/16 x 1/16
4459 ADDR_ASSERT_ALWAYS();
4463 mipWidth
= ((mip
- blk32MipId
) == 0) ? 16 : 8;
4464 mipHeight
= mipWidth
;
4468 mipDepth
= mipWidth
;
4473 if (mipWidth
<= minInc
)
4475 // if we're below the minimal increment...
4478 // For 3d, just go in z direction
4479 mipCoord
.d
+= mipDepth
;
4483 // For 2d, first go across, then down
4484 if ((mipWidth
* 2) == minInc
)
4486 // if we're 2 mips below, that's when we go back in x, and down in y
4487 mipCoord
.w
-= minInc
;
4488 mipCoord
.h
+= minInc
;
4492 // otherwise, just go across in x
4493 mipCoord
.w
+= minInc
;
4499 // On even mip, go down, otherwise, go across
4502 mipCoord
.w
+= mipWidth
;
4506 mipCoord
.h
+= mipHeight
;
4509 // Divide the width by 2
4511 // After the first mip in tail, the mip is always a square
4512 mipHeight
= mipWidth
;
4513 // ...or for 3d, a cube
4516 mipDepth
= mipWidth
;
4523 ************************************************************************************************************************
4524 * Gfx9Lib::GetMipStartPos
4527 * Internal function to get out information about mip logical start position
4530 * logical start position in macro block width/heith/depth of one mip level within one slice
4531 ************************************************************************************************************************
4533 Dim3d
Gfx9Lib::GetMipStartPos(
4534 AddrResourceType resourceType
,
4535 AddrSwizzleMode swizzleMode
,
4540 UINT_32 blockHeight
,
4543 UINT_32 log2ElementBytes
,
4544 UINT_32
* pMipTailBytesOffset
) const
4546 Dim3d mipStartPos
= {0};
4547 const Dim3d tailMaxDim
= GetMipTailDim(resourceType
, swizzleMode
, blockWidth
, blockHeight
, blockDepth
);
4549 // Report mip in tail if Mip0 is already in mip tail
4550 BOOL_32 inMipTail
= IsInMipTail(resourceType
, swizzleMode
, tailMaxDim
, width
, height
, depth
);
4551 UINT_32 log2blkSize
= GetBlockSizeLog2(swizzleMode
);
4552 UINT_32 mipIndexInTail
= mipId
;
4554 if (inMipTail
== FALSE
)
4556 // Mip 0 dimension, unit in block
4557 UINT_32 mipWidthInBlk
= width
/ blockWidth
;
4558 UINT_32 mipHeightInBlk
= height
/ blockHeight
;
4559 UINT_32 mipDepthInBlk
= depth
/ blockDepth
;
4560 AddrMajorMode majorMode
= GetMajorMode(resourceType
,
4566 UINT_32 endingMip
= mipId
+ 1;
4568 for (UINT_32 i
= 1; i
<= mipId
; i
++)
4570 if ((i
== 1) || (i
== 3))
4572 if (majorMode
== ADDR_MAJOR_Y
)
4574 mipStartPos
.w
+= mipWidthInBlk
;
4578 mipStartPos
.h
+= mipHeightInBlk
;
4583 if (majorMode
== ADDR_MAJOR_X
)
4585 mipStartPos
.w
+= mipWidthInBlk
;
4587 else if (majorMode
== ADDR_MAJOR_Y
)
4589 mipStartPos
.h
+= mipHeightInBlk
;
4593 mipStartPos
.d
+= mipDepthInBlk
;
4597 BOOL_32 inTail
= FALSE
;
4599 if (IsThick(resourceType
, swizzleMode
))
4601 UINT_32 dim
= log2blkSize
% 3;
4606 (mipWidthInBlk
<= 2) && (mipHeightInBlk
== 1) && (mipDepthInBlk
<= 2);
4611 (mipWidthInBlk
== 1) && (mipHeightInBlk
<= 2) && (mipDepthInBlk
<= 2);
4616 (mipWidthInBlk
<= 2) && (mipHeightInBlk
<= 2) && (mipDepthInBlk
== 1);
4621 if (log2blkSize
& 1)
4623 inTail
= (mipWidthInBlk
<= 2) && (mipHeightInBlk
== 1);
4627 inTail
= (mipWidthInBlk
== 1) && (mipHeightInBlk
<= 2);
4637 mipWidthInBlk
= RoundHalf(mipWidthInBlk
);
4638 mipHeightInBlk
= RoundHalf(mipHeightInBlk
);
4639 mipDepthInBlk
= RoundHalf(mipDepthInBlk
);
4642 if (mipId
>= endingMip
)
4645 mipIndexInTail
= mipId
- endingMip
;
4651 UINT_32 index
= mipIndexInTail
+ MaxMacroBits
- log2blkSize
;
4652 ADDR_ASSERT(index
< sizeof(MipTailOffset256B
) / sizeof(UINT_32
));
4653 *pMipTailBytesOffset
= MipTailOffset256B
[index
] << 8;
4660 ************************************************************************************************************************
4661 * Gfx9Lib::HwlComputeSurfaceAddrFromCoordTiled
4664 * Internal function to calculate address from coord for tiled swizzle surface
4668 ************************************************************************************************************************
4670 ADDR_E_RETURNCODE
Gfx9Lib::HwlComputeSurfaceAddrFromCoordTiled(
4671 const ADDR2_COMPUTE_SURFACE_ADDRFROMCOORD_INPUT
* pIn
, ///< [in] input structure
4672 ADDR2_COMPUTE_SURFACE_ADDRFROMCOORD_OUTPUT
* pOut
///< [out] output structure
4675 ADDR2_COMPUTE_SURFACE_INFO_INPUT localIn
= {0};
4676 localIn
.swizzleMode
= pIn
->swizzleMode
;
4677 localIn
.flags
= pIn
->flags
;
4678 localIn
.resourceType
= pIn
->resourceType
;
4679 localIn
.bpp
= pIn
->bpp
;
4680 localIn
.width
= Max(pIn
->unalignedWidth
, 1u);
4681 localIn
.height
= Max(pIn
->unalignedHeight
, 1u);
4682 localIn
.numSlices
= Max(pIn
->numSlices
, 1u);
4683 localIn
.numMipLevels
= Max(pIn
->numMipLevels
, 1u);
4684 localIn
.numSamples
= Max(pIn
->numSamples
, 1u);
4685 localIn
.numFrags
= Max(pIn
->numFrags
, 1u);
4686 if (localIn
.numMipLevels
<= 1)
4688 localIn
.pitchInElement
= pIn
->pitchInElement
;
4691 ADDR2_COMPUTE_SURFACE_INFO_OUTPUT localOut
= {0};
4692 ADDR_E_RETURNCODE returnCode
= ComputeSurfaceInfoTiled(&localIn
, &localOut
);
4694 BOOL_32 valid
= (returnCode
== ADDR_OK
) &&
4695 (IsThin(pIn
->resourceType
, pIn
->swizzleMode
) ||
4696 IsThick(pIn
->resourceType
, pIn
->swizzleMode
)) &&
4697 ((pIn
->pipeBankXor
== 0) || (IsXor(pIn
->swizzleMode
)));
4701 UINT_32 log2ElementBytes
= Log2(pIn
->bpp
>> 3);
4702 Dim3d mipStartPos
= {0};
4703 UINT_32 mipTailBytesOffset
= 0;
4705 if (pIn
->numMipLevels
> 1)
4707 // Mip-map chain cannot be MSAA surface
4708 ADDR_ASSERT((pIn
->numSamples
<= 1) && (pIn
->numFrags
<= 1));
4710 mipStartPos
= GetMipStartPos(pIn
->resourceType
,
4715 localOut
.blockWidth
,
4716 localOut
.blockHeight
,
4717 localOut
.blockSlices
,
4720 &mipTailBytesOffset
);
4723 UINT_32 interleaveOffset
= 0;
4724 UINT_32 pipeBits
= 0;
4725 UINT_32 pipeXor
= 0;
4726 UINT_32 bankBits
= 0;
4727 UINT_32 bankXor
= 0;
4729 if (IsThin(pIn
->resourceType
, pIn
->swizzleMode
))
4731 UINT_32 blockOffset
= 0;
4732 UINT_32 log2blkSize
= GetBlockSizeLog2(pIn
->swizzleMode
);
4734 if (IsZOrderSwizzle(pIn
->swizzleMode
))
4736 // Morton generation
4737 if ((log2ElementBytes
== 0) || (log2ElementBytes
== 2))
4739 UINT_32 totalLowBits
= 6 - log2ElementBytes
;
4740 UINT_32 mortBits
= totalLowBits
/ 2;
4741 UINT_32 lowBitsValue
= MortonGen2d(pIn
->y
, pIn
->x
, mortBits
);
4742 // Are 9 bits enough?
4743 UINT_32 highBitsValue
=
4744 MortonGen2d(pIn
->x
>> mortBits
, pIn
->y
>> mortBits
, 9) << totalLowBits
;
4745 blockOffset
= lowBitsValue
| highBitsValue
;
4746 ADDR_ASSERT(blockOffset
== lowBitsValue
+ highBitsValue
);
4750 blockOffset
= MortonGen2d(pIn
->y
, pIn
->x
, 13);
4753 // Fill LSBs with sample bits
4754 if (pIn
->numSamples
> 1)
4756 blockOffset
*= pIn
->numSamples
;
4757 blockOffset
|= pIn
->sample
;
4760 // Shift according to BytesPP
4761 blockOffset
<<= log2ElementBytes
;
4765 // Micro block offset
4766 UINT_32 microBlockOffset
= ComputeSurface2DMicroBlockOffset(pIn
);
4767 blockOffset
= microBlockOffset
;
4769 // Micro block dimension
4770 ADDR_ASSERT(log2ElementBytes
< MaxNumOfBpp
);
4771 Dim2d microBlockDim
= Block256_2d
[log2ElementBytes
];
4772 // Morton generation, does 12 bit enough?
4774 MortonGen2d((pIn
->x
/ microBlockDim
.w
), (pIn
->y
/ microBlockDim
.h
), 12) << 8;
4776 // Sample bits start location
4777 UINT_32 sampleStart
= log2blkSize
- Log2(pIn
->numSamples
);
4778 // Join sample bits information to the highest Macro block bits
4779 if (IsNonPrtXor(pIn
->swizzleMode
))
4781 // Non-prt-Xor : xor highest Macro block bits with sample bits
4782 blockOffset
= blockOffset
^ (pIn
->sample
<< sampleStart
);
4786 // Non-Xor or prt-Xor: replace highest Macro block bits with sample bits
4787 // after this op, the blockOffset only contains log2 Macro block size bits
4788 blockOffset
%= (1 << sampleStart
);
4789 blockOffset
|= (pIn
->sample
<< sampleStart
);
4790 ADDR_ASSERT((blockOffset
>> log2blkSize
) == 0);
4794 if (IsXor(pIn
->swizzleMode
))
4796 // Mask off bits above Macro block bits to keep page synonyms working for prt
4797 if (IsPrt(pIn
->swizzleMode
))
4799 blockOffset
&= ((1 << log2blkSize
) - 1);
4802 // Preserve offset inside pipe interleave
4803 interleaveOffset
= blockOffset
& ((1 << m_pipeInterleaveLog2
) - 1);
4804 blockOffset
>>= m_pipeInterleaveLog2
;
4807 pipeBits
= GetPipeXorBits(log2blkSize
);
4809 pipeXor
= FoldXor2d(blockOffset
, pipeBits
);
4810 blockOffset
>>= pipeBits
;
4813 bankBits
= GetBankXorBits(log2blkSize
);
4815 bankXor
= FoldXor2d(blockOffset
, bankBits
);
4816 blockOffset
>>= bankBits
;
4818 // Put all the part back together
4819 blockOffset
<<= bankBits
;
4820 blockOffset
|= bankXor
;
4821 blockOffset
<<= pipeBits
;
4822 blockOffset
|= pipeXor
;
4823 blockOffset
<<= m_pipeInterleaveLog2
;
4824 blockOffset
|= interleaveOffset
;
4827 ADDR_ASSERT((blockOffset
| mipTailBytesOffset
) == (blockOffset
+ mipTailBytesOffset
));
4828 ADDR_ASSERT((mipTailBytesOffset
== 0u) || (blockOffset
< (1u << log2blkSize
)));
4830 blockOffset
|= mipTailBytesOffset
;
4832 if (IsNonPrtXor(pIn
->swizzleMode
) && (pIn
->numSamples
<= 1))
4834 // Apply slice xor if not MSAA/PRT
4835 blockOffset
^= (ReverseBitVector(pIn
->slice
, pipeBits
) << m_pipeInterleaveLog2
);
4836 blockOffset
^= (ReverseBitVector(pIn
->slice
>> pipeBits
, bankBits
) <<
4837 (m_pipeInterleaveLog2
+ pipeBits
));
4840 returnCode
= ApplyCustomerPipeBankXor(pIn
->swizzleMode
, pIn
->pipeBankXor
,
4841 bankBits
, pipeBits
, &blockOffset
);
4843 blockOffset
%= (1 << log2blkSize
);
4845 UINT_32 pitchInMacroBlock
= localOut
.mipChainPitch
/ localOut
.blockWidth
;
4846 UINT_32 paddedHeightInMacroBlock
= localOut
.mipChainHeight
/ localOut
.blockHeight
;
4847 UINT_32 sliceSizeInMacroBlock
= pitchInMacroBlock
* paddedHeightInMacroBlock
;
4848 UINT_64 macroBlockIndex
=
4849 (pIn
->slice
+ mipStartPos
.d
) * sliceSizeInMacroBlock
+
4850 ((pIn
->y
/ localOut
.blockHeight
) + mipStartPos
.h
) * pitchInMacroBlock
+
4851 ((pIn
->x
/ localOut
.blockWidth
) + mipStartPos
.w
);
4853 pOut
->addr
= blockOffset
| (macroBlockIndex
<< log2blkSize
);
4857 UINT_32 log2blkSize
= GetBlockSizeLog2(pIn
->swizzleMode
);
4859 Dim3d microBlockDim
= Block1K_3d
[log2ElementBytes
];
4861 UINT_32 blockOffset
= MortonGen3d((pIn
->x
/ microBlockDim
.w
),
4862 (pIn
->y
/ microBlockDim
.h
),
4863 (pIn
->slice
/ microBlockDim
.d
),
4867 blockOffset
|= ComputeSurface3DMicroBlockOffset(pIn
);
4869 if (IsXor(pIn
->swizzleMode
))
4871 // Mask off bits above Macro block bits to keep page synonyms working for prt
4872 if (IsPrt(pIn
->swizzleMode
))
4874 blockOffset
&= ((1 << log2blkSize
) - 1);
4877 // Preserve offset inside pipe interleave
4878 interleaveOffset
= blockOffset
& ((1 << m_pipeInterleaveLog2
) - 1);
4879 blockOffset
>>= m_pipeInterleaveLog2
;
4882 pipeBits
= GetPipeXorBits(log2blkSize
);
4884 pipeXor
= FoldXor3d(blockOffset
, pipeBits
);
4885 blockOffset
>>= pipeBits
;
4888 bankBits
= GetBankXorBits(log2blkSize
);
4890 bankXor
= FoldXor3d(blockOffset
, bankBits
);
4891 blockOffset
>>= bankBits
;
4893 // Put all the part back together
4894 blockOffset
<<= bankBits
;
4895 blockOffset
|= bankXor
;
4896 blockOffset
<<= pipeBits
;
4897 blockOffset
|= pipeXor
;
4898 blockOffset
<<= m_pipeInterleaveLog2
;
4899 blockOffset
|= interleaveOffset
;
4902 ADDR_ASSERT((blockOffset
| mipTailBytesOffset
) == (blockOffset
+ mipTailBytesOffset
));
4903 ADDR_ASSERT((mipTailBytesOffset
== 0u) || (blockOffset
< (1u << log2blkSize
)));
4904 blockOffset
|= mipTailBytesOffset
;
4906 returnCode
= ApplyCustomerPipeBankXor(pIn
->swizzleMode
, pIn
->pipeBankXor
,
4907 bankBits
, pipeBits
, &blockOffset
);
4909 blockOffset
%= (1 << log2blkSize
);
4911 UINT_32 xb
= pIn
->x
/ localOut
.blockWidth
+ mipStartPos
.w
;
4912 UINT_32 yb
= pIn
->y
/ localOut
.blockHeight
+ mipStartPos
.h
;
4913 UINT_32 zb
= pIn
->slice
/ localOut
.blockSlices
+ + mipStartPos
.d
;
4915 UINT_32 pitchInBlock
= localOut
.mipChainPitch
/ localOut
.blockWidth
;
4916 UINT_32 sliceSizeInBlock
=
4917 (localOut
.mipChainHeight
/ localOut
.blockHeight
) * pitchInBlock
;
4918 UINT_64 blockIndex
= zb
* sliceSizeInBlock
+ yb
* pitchInBlock
+ xb
;
4920 pOut
->addr
= blockOffset
| (blockIndex
<< log2blkSize
);
4925 returnCode
= ADDR_INVALIDPARAMS
;
4932 ************************************************************************************************************************
4933 * Gfx9Lib::ComputeSurfaceInfoLinear
4936 * Internal function to calculate padding for linear swizzle 2D/3D surface
4940 ************************************************************************************************************************
4942 ADDR_E_RETURNCODE
Gfx9Lib::ComputeSurfaceLinearPadding(
4943 const ADDR2_COMPUTE_SURFACE_INFO_INPUT
* pIn
, ///< [in] input srtucture
4944 UINT_32
* pMipmap0PaddedWidth
, ///< [out] padded width in element
4945 UINT_32
* pSlice0PaddedHeight
, ///< [out] padded height for HW
4946 ADDR2_MIP_INFO
* pMipInfo
///< [out] per mip information
4949 ADDR_E_RETURNCODE returnCode
= ADDR_OK
;
4951 UINT_32 elementBytes
= pIn
->bpp
>> 3;
4952 UINT_32 pitchAlignInElement
= 0;
4954 if (pIn
->swizzleMode
== ADDR_SW_LINEAR_GENERAL
)
4956 ADDR_ASSERT(pIn
->numMipLevels
<= 1);
4957 ADDR_ASSERT(pIn
->numSlices
<= 1);
4958 pitchAlignInElement
= 1;
4962 pitchAlignInElement
= (256 / elementBytes
);
4965 UINT_32 mipChainWidth
= PowTwoAlign(pIn
->width
, pitchAlignInElement
);
4966 UINT_32 slice0PaddedHeight
= pIn
->height
;
4968 returnCode
= ApplyCustomizedPitchHeight(pIn
, elementBytes
, pitchAlignInElement
,
4969 &mipChainWidth
, &slice0PaddedHeight
);
4971 if (returnCode
== ADDR_OK
)
4973 UINT_32 mipChainHeight
= 0;
4974 UINT_32 mipHeight
= pIn
->height
;
4975 UINT_32 mipDepth
= (pIn
->resourceType
== ADDR_RSRC_TEX_3D
) ? pIn
->numSlices
: 1;
4977 for (UINT_32 i
= 0; i
< pIn
->numMipLevels
; i
++)
4979 if (pMipInfo
!= NULL
)
4981 pMipInfo
[i
].offset
= mipChainWidth
* mipChainHeight
* elementBytes
;
4982 pMipInfo
[i
].pitch
= mipChainWidth
;
4983 pMipInfo
[i
].height
= mipHeight
;
4984 pMipInfo
[i
].depth
= mipDepth
;
4987 mipChainHeight
+= mipHeight
;
4988 mipHeight
= RoundHalf(mipHeight
);
4989 mipHeight
= Max(mipHeight
, 1u);
4992 *pMipmap0PaddedWidth
= mipChainWidth
;
4993 *pSlice0PaddedHeight
= (pIn
->numMipLevels
> 1) ? mipChainHeight
: slice0PaddedHeight
;