xref: /aosp_15_r20/external/mesa3d/src/intel/isl/isl_surface_state.c (revision 6104692788411f58d303aa86923a9ff6ecaded22)
1 /*
2  * Copyright 2016 Intel Corporation
3  *
4  *  Permission is hereby granted, free of charge, to any person obtaining a
5  *  copy of this software and associated documentation files (the "Software"),
6  *  to deal in the Software without restriction, including without limitation
7  *  the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  *  and/or sell copies of the Software, and to permit persons to whom the
9  *  Software is furnished to do so, subject to the following conditions:
10  *
11  *  The above copyright notice and this permission notice (including the next
12  *  paragraph) shall be included in all copies or substantial portions of the
13  *  Software.
14  *
15  *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  *  IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  *  FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  *  THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  *  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  *  FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  *  IN THE SOFTWARE.
22  */
23 
24 #include <stdint.h>
25 #include <inttypes.h>
26 
27 #define __gen_address_type uint64_t
28 #define __gen_user_data void
29 
30 static uint64_t
__gen_combine_address(void * data,void * loc,uint64_t addr,uint32_t delta)31 __gen_combine_address(__attribute__((unused)) void *data,
32                       __attribute__((unused)) void *loc, uint64_t addr,
33                       uint32_t delta)
34 {
35    return addr + delta;
36 }
37 
38 #include "genxml/gen_macros.h"
39 #include "genxml/genX_pack.h"
40 #include "util/log.h"
41 
42 #include "isl_priv.h"
43 #include "isl_genX_helpers.h"
44 
45 #if GFX_VER >= 8
46 static const uint8_t isl_encode_tiling[] = {
47    [ISL_TILING_LINEAR]  = LINEAR,
48    [ISL_TILING_X]       = XMAJOR,
49 #if GFX_VERx10 >= 125
50    [ISL_TILING_4]       = TILE4,
51    [ISL_TILING_64]      = TILE64,
52    [ISL_TILING_64_XE2]  = TILE64,
53 #else
54    [ISL_TILING_Y0]      = YMAJOR,
55    [ISL_TILING_ICL_Yf]  = YMAJOR,
56    [ISL_TILING_ICL_Ys]  = YMAJOR,
57    [ISL_TILING_SKL_Yf]  = YMAJOR,
58    [ISL_TILING_SKL_Ys]  = YMAJOR,
59 #endif /* GFX_VERx10 < 125 */
60 #if GFX_VER <= 11
61    [ISL_TILING_W]       = WMAJOR,
62 #endif
63 };
64 #endif
65 
66 #if GFX_VER >= 9 && GFX_VERx10 <= 120
67 static const uint8_t isl_tiling_encode_trmode[] = {
68    [ISL_TILING_Y0]         = NONE,
69    [ISL_TILING_SKL_Yf]     = TILEYF,
70    [ISL_TILING_SKL_Ys]     = TILEYS,
71    [ISL_TILING_ICL_Yf]     = TILEYF,
72    [ISL_TILING_ICL_Ys]     = TILEYS,
73 };
74 #endif
75 
76 #if GFX_VER >= 7
77 static const uint32_t isl_encode_multisample_layout[] = {
78    [ISL_MSAA_LAYOUT_NONE]           = MSFMT_MSS,
79    [ISL_MSAA_LAYOUT_INTERLEAVED]    = MSFMT_DEPTH_STENCIL,
80    [ISL_MSAA_LAYOUT_ARRAY]          = MSFMT_MSS,
81 };
82 #endif
83 
84 #if GFX_VER >= 20
85 static const uint32_t isl_encode_aux_mode[] = {
86    [ISL_AUX_USAGE_NONE] = AUX_NONE,
87    [ISL_AUX_USAGE_MC] = AUX_NONE,
88    [ISL_AUX_USAGE_MCS] = AUX_MCS,
89    [ISL_AUX_USAGE_MCS_CCS] = AUX_MCS,
90    [ISL_AUX_USAGE_STC_CCS] = AUX_NONE,
91    [ISL_AUX_USAGE_HIZ_CCS_WT] = AUX_NONE,
92 };
93 #elif GFX_VER >= 12
94 static const uint32_t isl_encode_aux_mode[] = {
95    [ISL_AUX_USAGE_NONE] = AUX_NONE,
96    [ISL_AUX_USAGE_MC] = AUX_NONE,
97    [ISL_AUX_USAGE_MCS] = AUX_CCS_E,
98    [ISL_AUX_USAGE_FCV_CCS_E] = AUX_CCS_E,
99    [ISL_AUX_USAGE_CCS_E] = AUX_CCS_E,
100    [ISL_AUX_USAGE_HIZ_CCS_WT] = AUX_CCS_E,
101    [ISL_AUX_USAGE_MCS_CCS] = AUX_MCS_LCE,
102    [ISL_AUX_USAGE_STC_CCS] = AUX_CCS_E,
103 };
104 #elif GFX_VER >= 9
105 static const uint32_t isl_encode_aux_mode[] = {
106    [ISL_AUX_USAGE_NONE] = AUX_NONE,
107    [ISL_AUX_USAGE_HIZ] = AUX_HIZ,
108    [ISL_AUX_USAGE_MCS] = AUX_CCS_D,
109    [ISL_AUX_USAGE_CCS_D] = AUX_CCS_D,
110    [ISL_AUX_USAGE_CCS_E] = AUX_CCS_E,
111 };
112 #elif GFX_VER >= 8
113 static const uint32_t isl_encode_aux_mode[] = {
114    [ISL_AUX_USAGE_NONE] = AUX_NONE,
115    [ISL_AUX_USAGE_HIZ] = AUX_HIZ,
116    [ISL_AUX_USAGE_MCS] = AUX_MCS,
117    [ISL_AUX_USAGE_CCS_D] = AUX_MCS,
118 };
119 #endif
120 
121 static uint8_t
get_surftype(enum isl_surf_dim dim,isl_surf_usage_flags_t usage)122 get_surftype(enum isl_surf_dim dim, isl_surf_usage_flags_t usage)
123 {
124    switch (dim) {
125    default:
126       unreachable("bad isl_surf_dim");
127    case ISL_SURF_DIM_1D:
128       assert(!(usage & ISL_SURF_USAGE_CUBE_BIT));
129       return SURFTYPE_1D;
130    case ISL_SURF_DIM_2D:
131       if ((usage & ISL_SURF_USAGE_CUBE_BIT) &&
132           (usage & ISL_SURF_USAGE_TEXTURE_BIT)) {
133          /* We need SURFTYPE_CUBE to make cube sampling work */
134          return SURFTYPE_CUBE;
135       } else {
136          /* Everything else (render and storage) treat cubes as plain
137           * 2D array textures
138           */
139          return SURFTYPE_2D;
140       }
141    case ISL_SURF_DIM_3D:
142       assert(!(usage & ISL_SURF_USAGE_CUBE_BIT));
143       return SURFTYPE_3D;
144    }
145 }
146 
147 #if GFX_VERx10 >= 125
148 static uint8_t
get_media_compression_format(enum isl_format format,enum isl_format lowered_format)149 get_media_compression_format(enum isl_format format,
150                              enum isl_format lowered_format)
151 {
152    const uint32_t plane_bpb = isl_format_get_layout(lowered_format)->bpb;
153 
154    /* From Bspec 43868, Enumeration_MediaCompressionFormat:
155     *
156     *    Luma P010 has MSB of 0 while chroma P010 has MSB of 1.
157     *    Luma P016 has MSB of 0 while chroma P016 has MSB of 1.
158     *    Luma NV12 has MSB of 0 while chroma NV12 has MSB of 1.
159     */
160    switch (format) {
161    case ISL_FORMAT_PLANAR_420_8: /* NV12 */
162       assert(plane_bpb == 8 || plane_bpb == 16);
163       assert((isl_format_get_aux_map_encoding(format) & 0xf0) == 0);
164 
165       /* drm_fourcc.h defines the chroma plane of NV12 as 16-bpb */
166       return (plane_bpb == 16) << 4 | isl_format_get_aux_map_encoding(format);
167    case ISL_FORMAT_PLANAR_420_10:
168    case ISL_FORMAT_PLANAR_420_12:
169    case ISL_FORMAT_PLANAR_420_16:
170       assert(plane_bpb == 16 || plane_bpb == 32);
171       assert((isl_format_get_aux_map_encoding(format) & 0xf0) == 0);
172 
173       /* drm_fourcc.h defines the chroma plane of P01X as 32-bpb */
174       return (plane_bpb == 32) << 4 | isl_format_get_aux_map_encoding(format);
175    default:
176       return isl_format_get_aux_map_encoding(format);
177    }
178 }
179 #endif
180 
181 static struct isl_swizzle
format_swizzle(enum isl_format format)182 format_swizzle(enum isl_format format)
183 {
184    struct isl_swizzle fmt_swz = {};
185 
186    fmt_swz.r = isl_format_has_color_component(format, 0) ?
187                ISL_CHANNEL_SELECT_RED :
188                ISL_CHANNEL_SELECT_ZERO;
189 
190    fmt_swz.g = isl_format_has_color_component(format, 1) ?
191                ISL_CHANNEL_SELECT_GREEN :
192                ISL_CHANNEL_SELECT_ZERO;
193 
194    fmt_swz.b = isl_format_has_color_component(format, 2) ?
195                ISL_CHANNEL_SELECT_BLUE :
196                ISL_CHANNEL_SELECT_ZERO;
197 
198    fmt_swz.a = isl_format_has_color_component(format, 3) ?
199                ISL_CHANNEL_SELECT_ALPHA :
200                ISL_CHANNEL_SELECT_ONE;
201 
202    return fmt_swz;
203 }
204 
205 static struct isl_swizzle
isl_get_shader_channel_select(enum isl_format format,struct isl_swizzle view_swizzle)206 isl_get_shader_channel_select(enum isl_format format,
207                               struct isl_swizzle view_swizzle)
208 {
209    /* Currently we map both FXT1_RGB/FXT1_RGBA to single FXT1 format
210     * (supporting both 3/4 component) as a result we can use RGBA swizzle as it
211     * is.
212     */
213    if (format == ISL_FORMAT_FXT1)
214       return view_swizzle;
215 
216    /* Bspec 12486: RENDER_SURFACE_STATE::Shader Channel Select Red
217     *
218     *    "The output channel is undefined if the source is to a channel is not
219     *    present for the current surface format. For example, If the surface
220     *    format is R16_float and the shader channel select green specifies
221     *    green as the source the output is undefined. It should instead select
222     *    0 which is the default for a missing color channel."
223     *
224     * Bspec 57023: RENDER_SURFACE_STATE::Shader Channel Select Red
225     *
226     *    "For channels not present in the surface format, the corresponding
227     *    Surface Channel Select is either SCS_ZERO or SCS_ONE."
228     *
229     * This restriction applies to alpha channel as well if associated resource
230     * is not used as render target.
231     */
232    return isl_swizzle_compose(view_swizzle, format_swizzle(format));
233 }
234 
235 void
isl_genX(surf_fill_state_s)236 isl_genX(surf_fill_state_s)(const struct isl_device *dev, void *state,
237                             const struct isl_surf_fill_state_info *restrict info)
238 {
239 #ifndef NDEBUG
240    isl_surf_usage_flags_t _base_usage =
241       info->view->usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
242                            ISL_SURF_USAGE_TEXTURE_BIT |
243                            ISL_SURF_USAGE_STORAGE_BIT);
244    /* They may only specify one of the above bits at a time */
245    assert(__builtin_popcount(_base_usage) == 1);
246    /* The only other allowed bit is ISL_SURF_USAGE_CUBE_BIT */
247    assert((info->view->usage & ~(ISL_SURF_USAGE_CUBE_BIT |
248                                  ISL_SURF_USAGE_PROTECTED_BIT)) ==
249           _base_usage);
250 #endif
251 
252    if (info->surf->dim == ISL_SURF_DIM_3D) {
253       assert(info->view->base_array_layer + info->view->array_len <=
254              info->surf->logical_level0_px.depth);
255    } else {
256       assert(info->view->base_array_layer + info->view->array_len <=
257              info->surf->logical_level0_px.array_len);
258    }
259 
260    struct GENX(RENDER_SURFACE_STATE) s = { 0 };
261 
262    s.SurfaceType = get_surftype(info->surf->dim, info->view->usage);
263 
264    if (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT)
265       assert(isl_format_supports_rendering(dev->info, info->view->format));
266    else if (info->view->usage & ISL_SURF_USAGE_TEXTURE_BIT)
267       assert(isl_format_supports_sampling(dev->info, info->view->format));
268 
269    /* From the Sky Lake PRM Vol. 2d, RENDER_SURFACE_STATE::SurfaceFormat
270     *
271     *    This field cannot be a compressed (BC*, DXT*, FXT*, ETC*, EAC*)
272     *    format if the Surface Type is SURFTYPE_1D
273     */
274    if (info->surf->dim == ISL_SURF_DIM_1D)
275       assert(!isl_format_is_compressed(info->view->format));
276 
277    if (isl_format_is_compressed(info->surf->format)) {
278       /* You're not allowed to make a view of a compressed format with any
279        * format other than the surface format.  None of the userspace APIs
280        * allow for this directly and doing so would mess up a number of
281        * surface parameters such as Width, Height, and alignments.  Ideally,
282        * we'd like to assert that the two formats match.  However, we have an
283        * S3TC workaround that requires us to do reinterpretation.  So assert
284        * that they're at least the same bpb and block size.
285        */
286       ASSERTED const struct isl_format_layout *surf_fmtl =
287          isl_format_get_layout(info->surf->format);
288       ASSERTED const struct isl_format_layout *view_fmtl =
289          isl_format_get_layout(info->view->format);
290 
291       assert(surf_fmtl->bpb == view_fmtl->bpb);
292 
293       /* We could be attempting to upload blocks of compressed data via an
294        * uncompressed view, blocksize will not match there.
295        */
296       if (isl_format_is_compressed(info->view->format)) {
297          assert(surf_fmtl->bw == view_fmtl->bw);
298          assert(surf_fmtl->bh == view_fmtl->bh);
299       }
300    }
301 
302    s.SurfaceFormat = info->view->format;
303 
304 #if GFX_VER >= 12
305    /* The BSpec description of this field says:
306     *
307     *    "This bit field, when set, indicates if the resource is created as
308     *    Depth/Stencil resource."
309     *
310     *    "SW must set this bit for any resource that was created with
311     *    Depth/Stencil resource flag. Setting this bit allows HW to properly
312     *    interpret the data-layout for various cases. For any resource that's
313     *    created without Depth/Stencil resource flag, it must be reset."
314     *
315     * Even though the docs for this bit seem to imply that it's required for
316     * anything which might have been used for depth/stencil, empirical
317     * evidence suggests that it only affects CCS compression usage.  There are
318     * a few things which back this up:
319     *
320     *  1. The docs are also pretty clear that this bit was added as part
321     *     of enabling Gfx12 depth/stencil lossless compression.
322     *
323     *  2. The only new difference between depth/stencil and color images on
324     *     Gfx12 (where the bit was added) is how they treat CCS compression.
325     *     All other differences such as alignment requirements and MSAA layout
326     *     are already covered by other bits.
327     *
328     * Under these assumptions, it makes sense for ISL to model this bit as
329     * being an extension of AuxiliarySurfaceMode where STC_CCS and HIZ_CCS_WT
330     * are indicated by AuxiliarySurfaceMode == CCS_E and DepthStencilResource
331     * == true.
332     */
333    s.DepthStencilResource = info->aux_usage == ISL_AUX_USAGE_HIZ_CCS_WT ||
334                             info->aux_usage == ISL_AUX_USAGE_STC_CCS;
335 #endif
336 
337 #if GFX_VER <= 5
338    s.ColorBufferComponentWriteDisables = info->write_disables;
339    s.ColorBlendEnable = info->blend_enable;
340 #else
341    assert(info->write_disables == 0);
342 #endif
343 
344 #if GFX_VERx10 == 75
345    s.IntegerSurfaceFormat =
346       isl_format_has_int_channel((enum isl_format) s.SurfaceFormat);
347 #endif
348 
349    assert(info->surf->logical_level0_px.width > 0 &&
350           info->surf->logical_level0_px.height > 0);
351 
352    s.Width = info->surf->logical_level0_px.width - 1;
353    s.Height = info->surf->logical_level0_px.height - 1;
354 
355    /* In the gfx6 PRM Volume 1 Part 1: Graphics Core, Section 7.18.3.7.1
356     * (Surface Arrays For all surfaces other than separate stencil buffer):
357     *
358     * "[DevSNB] Errata: Sampler MSAA Qpitch will be 4 greater than the value
359     *  calculated in the equation above , for every other odd Surface Height
360     *  starting from 1 i.e. 1,5,9,13"
361     *
362     * Since this Qpitch errata only impacts the sampler, we have to adjust the
363     * input for the rendering surface to achieve the same qpitch. For the
364     * affected heights, we increment the height by 1 for the rendering
365     * surface.
366     */
367    if (GFX_VER == 6 && (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) &&
368        info->surf->samples > 1 &&
369        (info->surf->logical_level0_px.height % 4) == 1)
370       s.Height++;
371 
372    switch (s.SurfaceType) {
373    case SURFTYPE_1D:
374    case SURFTYPE_2D:
375       /* From the Ivy Bridge PRM >> RENDER_SURFACE_STATE::MinimumArrayElement:
376        *
377        *    "If Number of Multisamples is not MULTISAMPLECOUNT_1, this field
378        *    must be set to zero if this surface is used with sampling engine
379        *    messages."
380        *
381        * This restriction appears to exist only on Ivy Bridge.
382        */
383       if (GFX_VERx10 == 70 && !ISL_DEV_IS_BAYTRAIL(dev) &&
384           (info->view->usage & ISL_SURF_USAGE_TEXTURE_BIT) &&
385           info->surf->samples > 1)
386          assert(info->view->base_array_layer == 0);
387 
388       s.MinimumArrayElement = info->view->base_array_layer;
389 
390       /* From the Broadwell PRM >> RENDER_SURFACE_STATE::Depth:
391        *
392        *    For SURFTYPE_1D, 2D, and CUBE: The range of this field is reduced
393        *    by one for each increase from zero of Minimum Array Element. For
394        *    example, if Minimum Array Element is set to 1024 on a 2D surface,
395        *    the range of this field is reduced to [0,1023].
396        *
397        * In other words, 'Depth' is the number of array layers.
398        */
399       s.Depth = info->view->array_len - 1;
400 
401       /* From the Broadwell PRM >> RENDER_SURFACE_STATE::RenderTargetViewExtent:
402        *
403        *    For Render Target and Typed Dataport 1D and 2D Surfaces:
404        *    This field must be set to the same value as the Depth field.
405        */
406       if (info->view->usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
407                                ISL_SURF_USAGE_STORAGE_BIT))
408          s.RenderTargetViewExtent = s.Depth;
409       break;
410    case SURFTYPE_CUBE:
411       s.MinimumArrayElement = info->view->base_array_layer;
412       /* Same as SURFTYPE_2D, but divided by 6 */
413       s.Depth = info->view->array_len / 6 - 1;
414       if (info->view->usage & (ISL_SURF_USAGE_RENDER_TARGET_BIT |
415                                ISL_SURF_USAGE_STORAGE_BIT))
416          s.RenderTargetViewExtent = s.Depth;
417       break;
418    case SURFTYPE_3D:
419       /* From the Broadwell PRM >> RENDER_SURFACE_STATE::Depth:
420        *
421        *    If the volume texture is MIP-mapped, this field specifies the
422        *    depth of the base MIP level.
423        */
424       s.Depth = info->surf->logical_level0_px.depth - 1;
425 
426       /* From the Broadwell PRM >> RENDER_SURFACE_STATE::RenderTargetViewExtent:
427        *
428        *    For Render Target and Typed Dataport 3D Surfaces: This field
429        *    indicates the extent of the accessible 'R' coordinates minus 1 on
430        *    the LOD currently being rendered to.
431        *
432        * The docs specify that this only matters for render targets and
433        * surfaces used with typed dataport messages.  Prior to Ivy Bridge, the
434        * Depth field has more bits than RenderTargetViewExtent so we can have
435        * textures with more levels than we can render to.  In order to prevent
436        * assert-failures in the packing function below, we only set the field
437        * when it's actually going to be used by the hardware.
438        *
439        * The MinimumArrayElement field is ignored by all hardware
440        * prior to Sky Lake when texturing, and drivers are responsible
441        * for validating the correctness of this parameter.
442        * KHR_gl_texture_3D_image requires this functionality.
443        */
444       s.MinimumArrayElement = info->view->base_array_layer;
445       s.RenderTargetViewExtent = info->view->array_len - 1;
446       break;
447    default:
448       unreachable("bad SurfaceType");
449    }
450 
451 #if GFX_VER >= 7
452    if (INTEL_NEEDS_WA_1806565034) {
453       /* Wa_1806565034:
454        *
455        *    "Only set SurfaceArray if arrayed surface is > 1."
456        *
457        * Since this is a performance workaround, we only enable it when robust
458        * image access is disabled. Otherwise layered robust access is not
459        * specification compliant.
460        */
461       s.SurfaceArray = info->surf->dim != ISL_SURF_DIM_3D &&
462          (info->robust_image_access || info->view->array_len > 1);
463    } else {
464       s.SurfaceArray = info->surf->dim != ISL_SURF_DIM_3D;
465    }
466 #endif
467 
468    if (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) {
469       /* For render target surfaces, the hardware interprets field
470        * MIPCount/LOD as LOD. The Broadwell PRM says:
471        *
472        *    MIPCountLOD defines the LOD that will be rendered into.
473        *    SurfaceMinLOD is ignored.
474        */
475       s.MIPCountLOD = info->view->base_level;
476       s.SurfaceMinLOD = 0;
477    } else {
478       /* For non render target surfaces, the hardware interprets field
479        * MIPCount/LOD as MIPCount.  The range of levels accessible by the
480        * sampler engine is [SurfaceMinLOD, SurfaceMinLOD + MIPCountLOD].
481        */
482       s.SurfaceMinLOD = info->view->base_level;
483       s.MIPCountLOD = MAX(info->view->levels, 1) - 1;
484    }
485 
486 #if GFX_VER >= 9
487    s.MipTailStartLOD = info->surf->miptail_start_level;
488 #endif
489 
490 #if GFX_VERx10 >= 125
491    /* Setting L1 caching policy to Write-back mode. */
492    s.L1CacheControl = L1CC_WB;
493 #endif
494 
495 #if GFX_VER >= 6
496    const struct isl_extent3d image_align =
497       isl_get_image_alignment(info->surf);
498    s.SurfaceVerticalAlignment = isl_encode_valign(image_align.height);
499 #if GFX_VER >= 7
500    s.SurfaceHorizontalAlignment = isl_encode_halign(image_align.width);
501 #endif
502 #endif
503 
504    if (info->surf->dim_layout == ISL_DIM_LAYOUT_GFX9_1D) {
505       /* For gfx9 1-D textures, surface pitch is ignored */
506       s.SurfacePitch = 0;
507    } else {
508       s.SurfacePitch = info->surf->row_pitch_B - 1;
509    }
510 
511 #if GFX_VER >= 8
512    s.SurfaceQPitch = isl_get_qpitch(info->surf) >> 2;
513 #elif GFX_VER == 7
514    s.SurfaceArraySpacing = info->surf->array_pitch_span ==
515                            ISL_ARRAY_PITCH_SPAN_COMPACT;
516 #endif
517 
518 #if GFX_VER >= 9 && GFX_VERx10 <= 120
519    assert(GFX_VER < 12 || info->surf->tiling != ISL_TILING_W);
520 
521    /* From the SKL+ PRMs, RENDER_SURFACE_STATE:TileMode,
522     *
523     *    If Surface Format is ASTC*, this field must be TILEMODE_YMAJOR.
524     */
525    if (isl_format_get_layout(info->view->format)->txc == ISL_TXC_ASTC)
526       assert(isl_tiling_is_any_y(info->surf->tiling));
527 
528    s.TileMode = isl_encode_tiling[info->surf->tiling];
529    if (isl_tiling_is_std_y(info->surf->tiling))
530       s.TiledResourceMode = isl_tiling_encode_trmode[info->surf->tiling];
531 #elif GFX_VER >= 8
532    assert(isl_format_get_layout(info->view->format)->txc != ISL_TXC_ASTC);
533    assert(!isl_tiling_is_std_y(info->surf->tiling));
534    s.TileMode = isl_encode_tiling[info->surf->tiling];
535 #else
536    s.TiledSurface = info->surf->tiling != ISL_TILING_LINEAR,
537    s.TileWalk = info->surf->tiling == ISL_TILING_Y0 ? TILEWALK_YMAJOR :
538                                                       TILEWALK_XMAJOR,
539 #endif
540 
541 #if GFX_VER >= 8
542    s.RenderCacheReadWriteMode = WriteOnlyCache;
543 #else
544    s.RenderCacheReadWriteMode = 0;
545 #endif
546 
547 #if GFX_VER >= 11 && GFX_VERx10 < 125
548    /* From the TGL PRM,
549     *
550     *    This bit should never be programmed to 0
551     */
552    s.EnableUnormPathInColorPipe = true;
553 #endif
554 
555    s.CubeFaceEnablePositiveZ = 1;
556    s.CubeFaceEnableNegativeZ = 1;
557    s.CubeFaceEnablePositiveY = 1;
558    s.CubeFaceEnableNegativeY = 1;
559    s.CubeFaceEnablePositiveX = 1;
560    s.CubeFaceEnableNegativeX = 1;
561 
562 #if GFX_VER >= 6
563    /* From the Broadwell PRM for "Number of Multisamples":
564     *
565     *    "If this field is any value other than MULTISAMPLECOUNT_1, Surface
566     *    Min LOD, Mip Count / LOD, and Resource Min LOD must be set to zero."
567     *
568     * This is fine because no 3D API allows multisampling and mipmapping at
569     * the same time.
570     */
571    if (info->surf->samples > 1) {
572       assert(info->view->min_lod_clamp == 0);
573       assert(info->view->base_level == 0);
574       assert(info->view->levels == 1);
575    }
576    s.NumberofMultisamples = ffs(info->surf->samples) - 1;
577 #if GFX_VER >= 7
578    s.MultisampledSurfaceStorageFormat =
579       isl_encode_multisample_layout[info->surf->msaa_layout];
580 #endif
581 #endif
582 
583 #if GFX_VER >= 7
584    s.ResourceMinLOD = info->view->min_lod_clamp;
585 
586 #if GFX_VERx10 >= 200
587    s.EnableSamplerRoutetoLSC =
588       isl_format_support_sampler_route_to_lsc(info->view->format) &&
589       s.SurfaceType == SURFTYPE_2D && info->view->array_len == 1;
590 
591 /* Wa_14018471104:
592  * For APIs that use ResourceMinLod, do the following: (remains same as before)
593  *    1. If ResourceMinLod == 0.0 then **Enable Sampler Route to LSC**
594  *       in RENDER SURFACE STATE to 1 else to 0
595  */
596 #if INTEL_NEEDS_WA_14018471104
597    s.EnableSamplerRoutetoLSC &= info->view->min_lod_clamp == 0;
598 #endif
599 
600    /* Per application override. */
601    s.EnableSamplerRoutetoLSC &= dev->sampler_route_to_lsc;
602 #endif /* if GFX_VERx10 >= 200 */
603 
604 #else
605    assert(info->view->min_lod_clamp == 0);
606 #endif
607 
608 #if (GFX_VERx10 >= 75)
609    struct isl_swizzle swz = info->view->swizzle;
610    if (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) {
611       assert(isl_swizzle_supports_rendering(dev->info, swz));
612    } else {
613       swz = isl_get_shader_channel_select(info->view->format, swz);
614    }
615 
616    s.ShaderChannelSelectRed = (enum GENX(ShaderChannelSelect)) swz.r;
617    s.ShaderChannelSelectGreen = (enum GENX(ShaderChannelSelect)) swz.g;
618    s.ShaderChannelSelectBlue = (enum GENX(ShaderChannelSelect)) swz.b;
619    s.ShaderChannelSelectAlpha = (enum GENX(ShaderChannelSelect)) swz.a;
620 #else
621    assert(isl_swizzle_is_identity(info->view->swizzle));
622 #endif
623 
624    assert(info->address % info->surf->alignment_B == 0);
625    s.SurfaceBaseAddress = info->address;
626 
627 #if GFX_VER >= 6
628    s.MOCS = info->mocs;
629 #endif
630 
631 #if GFX_VERx10 >= 45
632    if (info->x_offset_sa != 0 || info->y_offset_sa != 0) {
633       /* There are fairly strict rules about when the offsets can be used.
634        * These are mostly taken from the Sky Lake PRM documentation for
635        * RENDER_SURFACE_STATE.
636        */
637       assert(info->surf->tiling != ISL_TILING_LINEAR);
638       assert(info->surf->dim == ISL_SURF_DIM_2D);
639       assert(isl_is_pow2(isl_format_get_layout(info->view->format)->bpb));
640       assert(info->surf->levels == 1);
641       assert(info->surf->logical_level0_px.array_len == 1);
642       assert(info->aux_usage == ISL_AUX_USAGE_NONE);
643 
644       if (GFX_VER >= 8) {
645          /* Broadwell added more rules. */
646          assert(info->surf->samples == 1);
647          if (isl_format_get_layout(info->view->format)->bpb == 8)
648             assert(info->x_offset_sa % 16 == 0);
649          if (isl_format_get_layout(info->view->format)->bpb == 16)
650             assert(info->x_offset_sa % 8 == 0);
651       }
652 
653 #if GFX_VER >= 7
654       s.SurfaceArray = false;
655 #endif
656    }
657 
658    const unsigned x_div = 4;
659    const unsigned y_div = GFX_VER >= 8 ? 4 : 2;
660    assert(info->x_offset_sa % x_div == 0);
661    assert(info->y_offset_sa % y_div == 0);
662    s.XOffset = info->x_offset_sa / x_div;
663    s.YOffset = info->y_offset_sa / y_div;
664 #else
665    assert(info->x_offset_sa == 0);
666    assert(info->y_offset_sa == 0);
667 #endif
668 
669 #if GFX_VER >= 7
670    if (info->aux_usage != ISL_AUX_USAGE_NONE) {
671       /* Check valid aux usages per-gen */
672       if (GFX_VER >= 12) {
673          assert(info->aux_usage == ISL_AUX_USAGE_MCS ||
674                 info->aux_usage == ISL_AUX_USAGE_CCS_E ||
675                 info->aux_usage == ISL_AUX_USAGE_FCV_CCS_E ||
676                 info->aux_usage == ISL_AUX_USAGE_MC ||
677                 info->aux_usage == ISL_AUX_USAGE_HIZ_CCS_WT ||
678                 info->aux_usage == ISL_AUX_USAGE_MCS_CCS ||
679                 info->aux_usage == ISL_AUX_USAGE_STC_CCS);
680       } else if (GFX_VER >= 9) {
681          assert(info->aux_usage == ISL_AUX_USAGE_HIZ ||
682                 info->aux_usage == ISL_AUX_USAGE_MCS ||
683                 info->aux_usage == ISL_AUX_USAGE_CCS_D ||
684                 info->aux_usage == ISL_AUX_USAGE_CCS_E);
685       } else if (GFX_VER >= 8) {
686          assert(info->aux_usage == ISL_AUX_USAGE_HIZ ||
687                 info->aux_usage == ISL_AUX_USAGE_MCS ||
688                 info->aux_usage == ISL_AUX_USAGE_CCS_D);
689       } else if (GFX_VER >= 7) {
690          assert(info->aux_usage == ISL_AUX_USAGE_MCS ||
691                 info->aux_usage == ISL_AUX_USAGE_CCS_D);
692       }
693 
694       /* The docs don't appear to say anything whatsoever about compression
695        * and the data port.  Testing seems to indicate that the data port
696        * completely ignores the AuxiliarySurfaceMode field.
697        *
698        * On gfx12 HDC supports compression.
699        */
700       if (GFX_VER < 12)
701          assert(!(info->view->usage & ISL_SURF_USAGE_STORAGE_BIT));
702 
703       if (isl_surf_usage_is_depth(info->surf->usage))
704          assert(isl_aux_usage_has_hiz(info->aux_usage));
705 
706       if (isl_surf_usage_is_stencil(info->surf->usage))
707          assert(info->aux_usage == ISL_AUX_USAGE_STC_CCS);
708 
709       if (isl_aux_usage_has_hiz(info->aux_usage)) {
710          /* For Gfx8-11, there are some restrictions around sampling from HiZ.
711           * The Skylake PRM docs for RENDER_SURFACE_STATE::AuxiliarySurfaceMode
712           * say:
713           *
714           *    "If this field is set to AUX_HIZ, Number of Multisamples must
715           *    be MULTISAMPLECOUNT_1, and Surface Type cannot be SURFTYPE_3D."
716           *
717           * On Gfx12, the docs are a bit less obvious but the restriction is
718           * the same.  The limitation isn't called out explicitly but the docs
719           * for the CCS_E value of RENDER_SURFACE_STATE::AuxiliarySurfaceMode
720           * say:
721           *
722           *    "If Number of multisamples > 1, programming this value means
723           *    MSAA compression is enabled for that surface. Auxiliary surface
724           *    is MSC with tile y."
725           *
726           * Since this interpretation ignores whether the surface is
727           * depth/stencil or not and since multisampled depth buffers use
728           * ISL_MSAA_LAYOUT_INTERLEAVED which is incompatible with MCS
729           * compression, this means that we can't even specify MSAA depth CCS
730           * in RENDER_SURFACE_STATE::AuxiliarySurfaceMode.
731           *
732           * On Xe2+, the above restriction is not mentioned in the
733           * RENDER_SURFACE_STATE::AuxiliarySurfaceMode.
734           *
735           * Bspec 57023 (r58975)
736           */
737          assert(GFX_VER >= 20 || info->surf->samples == 1);
738 
739          /* Prior to Gfx12, the dimension must not be 3D */
740          if (info->aux_usage == ISL_AUX_USAGE_HIZ)
741             assert(info->surf->dim != ISL_SURF_DIM_3D);
742 
743          /* The format must be one of the following: */
744          switch (info->view->format) {
745          case ISL_FORMAT_R32_FLOAT:
746          case ISL_FORMAT_R24_UNORM_X8_TYPELESS:
747          case ISL_FORMAT_R16_UNORM:
748             break;
749          default:
750             assert(!"Incompatible HiZ Sampling format");
751             break;
752          }
753       }
754 
755 #if GFX_VERx10 >= 125
756       if (info->aux_usage == ISL_AUX_USAGE_MC) {
757          s.CompressionFormat =
758             get_media_compression_format(info->mc_format, info->surf->format);
759       } else {
760          s.CompressionFormat =
761             isl_get_render_compression_format(info->surf->format);
762       }
763 #endif
764 #if GFX_VER == 12
765       s.MemoryCompressionEnable = info->aux_usage == ISL_AUX_USAGE_MC;
766 
767       /* The Tiger Lake PRM for RENDER_SURFACE_STATE::DecompressInL3 says:
768        *
769        *    When this field is set to 1h, the associated compressible surface,
770        *    when accessed by sampler and data-port, will be uncompressed in
771        *    L3. If the surface is not compressible, this bit field is ignored.
772        *
773        * The sampler's decompressor seems to lack support for some types of
774        * format re-interpretation. Use the more capable decompressor for these
775        * cases.
776        */
777       s.DecompressInL3 =
778          !isl_formats_have_same_bits_per_channel(info->surf->format,
779                                                  info->view->format);
780 #endif
781 #if GFX_VER >= 9
782       /* Some CCS aux usages have format restrictions. The Skylake PRM doc for
783        * RENDER_SURFACE_STATE::AuxiliarySurfaceMode says:
784        *
785        *    If Number of Multisamples is MULTISAMPLECOUNT_1, AUX_CCS_E setting
786        *    is only allowed if Surface Format is supported for Render Target
787        *    Compression. This setting enables render target compression.
788        *
789        * If CCS_E is in use, the format must support it.
790        */
791       if (isl_aux_usage_has_ccs_e(info->aux_usage))
792          assert(isl_format_supports_ccs_e(dev->info, info->view->format));
793 
794       /* It also says:
795        *
796        *    If Number of Multisamples is MULTISAMPLECOUNT_1, AUX_CCS_D setting
797        *    is only allowed if Surface Format supported for Fast Clear. In
798        *    addition, if the surface is bound to the sampling engine, Surface
799        *    Format must be supported for Render Target Compression for
800        *    surfaces bound to the sampling engine. For render target surfaces,
801        *    this setting disables render target compression. For sampling
802        *    engine surfaces, this mode behaves the same as AUX_CCS_E.
803        *
804        * If CCS_D is in use while rendering, the format must support it. If
805        * it's in use while sampling, the format must support CCS_E.
806        */
807       if (info->aux_usage == ISL_AUX_USAGE_CCS_D) {
808          if (info->view->usage & ISL_SURF_USAGE_RENDER_TARGET_BIT) {
809             assert(isl_format_supports_ccs_d(dev->info, info->view->format));
810          } else {
811             assert(info->view->usage & ISL_SURF_USAGE_TEXTURE_BIT);
812             assert(isl_format_supports_ccs_e(dev->info, info->view->format));
813          }
814       }
815 #endif
816 #if GFX_VER >= 8
817       s.AuxiliarySurfaceMode = isl_encode_aux_mode[info->aux_usage];
818 #else
819       s.MCSEnable = true;
820 #endif
821    }
822 
823    /* The auxiliary buffer info is filled when it's usable by the HW.
824     *
825     * Starting with Gfx12, the only form of compression that can be used
826     * with RENDER_SURFACE_STATE which requires an aux surface is MCS.
827     * HiZ still requires a surface but the HiZ surface can only be
828     * accessed through 3DSTATE_HIER_DEPTH_BUFFER.
829     *
830     * On all earlier hardware, an aux surface is required for all forms
831     * of compression.
832     */
833    if ((GFX_VER < 12 && info->aux_usage != ISL_AUX_USAGE_NONE) ||
834        (GFX_VER >= 12 && isl_aux_usage_has_mcs(info->aux_usage))) {
835 
836       assert(info->aux_surf != NULL);
837 
838       struct isl_tile_info tile_info;
839       isl_surf_get_tile_info(info->aux_surf, &tile_info);
840       uint32_t pitch_in_tiles =
841          info->aux_surf->row_pitch_B / tile_info.phys_extent_B.width;
842 
843       assert(info->aux_address % info->aux_surf->alignment_B == 0);
844       s.AuxiliarySurfaceBaseAddress = info->aux_address;
845       s.AuxiliarySurfacePitch = pitch_in_tiles - 1;
846 
847 #if GFX_VER >= 8
848       /* Auxiliary surfaces in ISL have compressed formats but the hardware
849        * doesn't expect our definition of the compression, it expects qpitch
850        * in units of samples on the main surface.
851        */
852       s.AuxiliarySurfaceQPitch =
853          isl_surf_get_array_pitch_sa_rows(info->aux_surf) >> 2;
854 #endif
855    }
856 #endif
857 
858 #if GFX_VER >= 8 && GFX_VER < 11
859    /* From the CHV PRM, Volume 2d, page 321 (RENDER_SURFACE_STATE dword 0
860     * bit 9 "Sampler L2 Bypass Mode Disable" Programming Notes):
861     *
862     *    This bit must be set for the following surface types: BC2_UNORM
863     *    BC3_UNORM BC5_UNORM BC5_SNORM BC7_UNORM
864     */
865    if (GFX_VER >= 9 || dev->info->platform == INTEL_PLATFORM_CHV) {
866       switch (info->view->format) {
867       case ISL_FORMAT_BC2_UNORM:
868       case ISL_FORMAT_BC3_UNORM:
869       case ISL_FORMAT_BC5_UNORM:
870       case ISL_FORMAT_BC5_SNORM:
871       case ISL_FORMAT_BC7_UNORM:
872          s.SamplerL2BypassModeDisable = true;
873          break;
874       default:
875          /* From the SKL PRM, Programming Note under Sampler Output Channel
876           * Mapping:
877           *
878           *    If a surface has an associated HiZ Auxiliary surface, the
879           *    Sampler L2 Bypass Mode Disable field in the RENDER_SURFACE_STATE
880           *    must be set.
881           */
882          if (GFX_VER >= 9 && info->aux_usage == ISL_AUX_USAGE_HIZ)
883             s.SamplerL2BypassModeDisable = true;
884          break;
885       }
886    }
887 #endif
888 
889    if (isl_aux_usage_has_fast_clears(info->aux_usage)) {
890       if (info->use_clear_address) {
891 #if GFX_VER > 10 && GFX_VER < 20
892          s.ClearValueAddressEnable = true;
893          s.ClearValueAddress = info->clear_address;
894 #else
895          unreachable("Only Gfx11 and Gfx12 support indirect clear colors");
896 #endif
897       }
898 
899 #if GFX_VER >= 20
900       /* According to Bspec 57023 >> RENDER_SURFACE_STATE, the clear value
901        * address and explicit clear value are removed since Xe2.
902        */
903       assert(!info->use_clear_address);
904 #elif GFX_VER >= 12
905       assert(info->use_clear_address);
906 #elif GFX_VER >= 9
907       if (!info->use_clear_address) {
908          s.RedClearColor = info->clear_color.u32[0];
909          s.GreenClearColor = info->clear_color.u32[1];
910          s.BlueClearColor = info->clear_color.u32[2];
911          s.AlphaClearColor = info->clear_color.u32[3];
912       }
913 #elif GFX_VER >= 7
914       /* Prior to Sky Lake, we only have one bit for the clear color which
915        * gives us 0 or 1 in whatever the surface's format happens to be.
916        */
917       if (isl_format_has_int_channel(info->view->format)) {
918          for (unsigned i = 0; i < 4; i++) {
919             assert(info->clear_color.u32[i] == 0 ||
920                    info->clear_color.u32[i] == 1);
921          }
922          s.RedClearColor = info->clear_color.u32[0] != 0;
923          s.GreenClearColor = info->clear_color.u32[1] != 0;
924          s.BlueClearColor = info->clear_color.u32[2] != 0;
925          s.AlphaClearColor = info->clear_color.u32[3] != 0;
926       } else {
927          for (unsigned i = 0; i < 4; i++) {
928             assert(info->clear_color.f32[i] == 0.0f ||
929                    info->clear_color.f32[i] == 1.0f);
930          }
931          s.RedClearColor = info->clear_color.f32[0] != 0.0f;
932          s.GreenClearColor = info->clear_color.f32[1] != 0.0f;
933          s.BlueClearColor = info->clear_color.f32[2] != 0.0f;
934          s.AlphaClearColor = info->clear_color.f32[3] != 0.0f;
935       }
936 #endif
937    }
938 
939    GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &s);
940 }
941 
942 void
isl_genX(buffer_fill_state_s)943 isl_genX(buffer_fill_state_s)(const struct isl_device *dev, void *state,
944                               const struct isl_buffer_fill_state_info *restrict info)
945 {
946    uint64_t buffer_size = info->size_B;
947 
948    /* Uniform and Storage buffers need to have surface size not less that the
949     * aligned 32-bit size of the buffer. To calculate the array length on
950     * unsized arrays in StorageBuffer the last 2 bits store the padding size
951     * added to the surface, so we can calculate latter the original buffer
952     * size to know the number of elements.
953     *
954     *  surface_size = isl_align(buffer_size, 4) +
955     *                 (isl_align(buffer_size) - buffer_size)
956     *
957     *  buffer_size = (surface_size & ~3) - (surface_size & 3)
958     */
959    if ((info->format == ISL_FORMAT_RAW  ||
960         info->stride_B < isl_format_get_layout(info->format)->bpb / 8) &&
961        !info->is_scratch) {
962       assert(info->stride_B == 1);
963       uint64_t aligned_size = isl_align(buffer_size, 4);
964       buffer_size = aligned_size + (aligned_size - buffer_size);
965    }
966 
967    uint32_t num_elements = buffer_size / info->stride_B;
968 
969    assert(num_elements > 0);
970    if (info->format == ISL_FORMAT_RAW) {
971       assert(num_elements <= dev->max_buffer_size);
972    } else {
973       /* From the IVB PRM, SURFACE_STATE::Height,
974        *
975        *    For typed buffer and structured buffer surfaces, the number
976        *    of entries in the buffer ranges from 1 to 2^27.
977        *
978        * We could assert(num_elements <= (1 << 27)) here, but some DX12 games
979        * misbehave and there's nothing either vkd3d or Anv can do about it.
980        * Therefore we just allow those cases to happen in order to avoid
981        * crashing or further breaking the applications.
982        *
983        * Applications causing this issue generally ignore
984        * PhysicalDevice::maxTexelBufferElements, leading them to disrespect
985        * restrictions such as:
986        *   VUID-VkDescriptorGetInfoEXT-type-09427
987        *   VUID-VkDescriptorGetInfoEXT-type-09428
988        *   VUID-VkBufferViewCreateInfo-range-00930
989        *   VUID-VkBufferViewCreateInfo-range-04059
990        */
991       if (num_elements > (1 << 27)) {
992          mesa_logw("%s: num_elements is too big: %u (buffer size: %"PRIu64")\n",
993                    __func__, num_elements, buffer_size);
994          num_elements = 1 << 27;
995       }
996    }
997 
998    struct GENX(RENDER_SURFACE_STATE) s = { 0, };
999 
1000    s.SurfaceFormat = info->format;
1001 
1002    s.SurfaceType = SURFTYPE_BUFFER;
1003 #if GFX_VERx10 >= 125
1004    if (info->is_scratch) {
1005       /* From the BSpec:
1006        *
1007        *    "For surfaces of type SURFTYPE_SCRATCH, valid range of pitch is:
1008        *    [63,262143] -> [64B, 256KB].  Also, for SURFTYPE_SCRATCH, the
1009        *    pitch must be a multiple of 64bytes."
1010        */
1011       assert(info->format == ISL_FORMAT_RAW);
1012       assert(info->stride_B % 64 == 0);
1013       assert(info->stride_B <= 256 * 1024);
1014       s.SurfaceType = SURFTYPE_SCRATCH;
1015    }
1016 #else
1017    assert(!info->is_scratch);
1018 #endif
1019 
1020    s.SurfacePitch = info->stride_B - 1;
1021 
1022 #if GFX_VER >= 6
1023    s.SurfaceVerticalAlignment = isl_encode_valign(4);
1024 #if GFX_VERx10 >= 125
1025    s.SurfaceHorizontalAlignment = isl_encode_halign(128);
1026 #elif GFX_VER >= 7
1027    s.SurfaceHorizontalAlignment = isl_encode_halign(4);
1028    s.SurfaceArray = false;
1029 #endif
1030 #endif
1031 
1032 #if GFX_VER >= 9
1033    s.Height = ((num_elements - 1) >> 7) & 0x3fff;
1034    s.Width = (num_elements - 1) & 0x7f;
1035    s.Depth = ((num_elements - 1) >> 21) & 0x7ff;
1036 #elif GFX_VER >= 7
1037    s.Height = ((num_elements - 1) >> 7) & 0x3fff;
1038    s.Width = (num_elements - 1) & 0x7f;
1039    s.Depth = ((num_elements - 1) >> 21) & 0x3ff;
1040 #else
1041    s.Height = ((num_elements - 1) >> 7) & 0x1fff;
1042    s.Width = (num_elements - 1) & 0x7f;
1043    s.Depth = ((num_elements - 1) >> 20) & 0x7f;
1044 #endif
1045 
1046 #if GFX_VER >= 6
1047    s.NumberofMultisamples = MULTISAMPLECOUNT_1;
1048 #endif
1049 
1050 #if (GFX_VER >= 8)
1051    s.TileMode = LINEAR;
1052 #else
1053    s.TiledSurface = false;
1054 #endif
1055 
1056 #if (GFX_VER >= 8)
1057    s.RenderCacheReadWriteMode = WriteOnlyCache;
1058 #else
1059    s.RenderCacheReadWriteMode = 0;
1060 #endif
1061 
1062 #if GFX_VERx10 >= 200
1063    s.EnableSamplerRoutetoLSC = isl_format_support_sampler_route_to_lsc(info->format);
1064 #endif /* if GFX_VERx10 >= 200 */
1065 
1066    s.SurfaceBaseAddress = info->address;
1067 #if GFX_VER >= 6
1068    s.MOCS = info->mocs;
1069 #endif
1070 
1071 #if GFX_VER >= 9
1072    /* Store the buffer size in the upper dword of the AUX surface base
1073     * address. Only enabled on Gfx9+ since Gfx8 has an Atom version with only
1074     * 32bits of address space.
1075     */
1076    if (dev->buffer_length_in_aux_addr) {
1077       assert(intel_needs_workaround(dev->info, 14019708328) == false);
1078       s.AuxiliarySurfaceBaseAddress = info->size_B << 32;
1079    } else {
1080       /* Wa_14019708328: all SURFTYPE_BUFFERs has
1081        * AuxiliarySurfaceMode == AUX_NONE so no need to check for it.
1082        * In case workaround is not needed and buffer_length_in_aux_addr is
1083        * false, it will set AuxiliarySurfaceBaseAddress to 0.
1084        */
1085       s.AuxiliarySurfaceBaseAddress = dev->dummy_aux_address;
1086    }
1087 #else
1088    assert(!dev->buffer_length_in_aux_addr);
1089 #endif
1090 
1091 #if GFX_VERx10 >= 125
1092    /* Setting L1 caching policy to Write-back mode. */
1093    s.L1CacheControl = L1CC_WB;
1094 #endif
1095 
1096 #if (GFX_VERx10 >= 75)
1097    struct isl_swizzle swz = isl_get_shader_channel_select(info->format,
1098                                                           info->swizzle);
1099 
1100    s.ShaderChannelSelectRed = (enum GENX(ShaderChannelSelect)) swz.r;
1101    s.ShaderChannelSelectGreen = (enum GENX(ShaderChannelSelect)) swz.g;
1102    s.ShaderChannelSelectBlue = (enum GENX(ShaderChannelSelect)) swz.b;
1103    s.ShaderChannelSelectAlpha = (enum GENX(ShaderChannelSelect)) swz.a;
1104 #endif
1105 
1106    GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &s);
1107 }
1108 
1109 void
isl_genX(null_fill_state_s)1110 isl_genX(null_fill_state_s)(const struct isl_device *dev, void *state,
1111                             const struct isl_null_fill_state_info *restrict info)
1112 {
1113    struct GENX(RENDER_SURFACE_STATE) s = {
1114       .SurfaceType = SURFTYPE_NULL,
1115       /* We previously had this format set to B8G8R8A8_UNORM but ran into
1116        * hangs on IVB. R32_UINT seems to work for everybody.
1117        *
1118        * https://gitlab.freedesktop.org/mesa/mesa/-/issues/1872
1119        */
1120       .SurfaceFormat = ISL_FORMAT_R32_UINT,
1121 #if GFX_VER >= 7
1122       .SurfaceArray = info->size.depth > 1,
1123 #endif
1124 #if GFX_VERx10 >= 125
1125       .TileMode = TILE4,
1126 #elif GFX_VER >= 8
1127       .TileMode = YMAJOR,
1128 #else
1129       .TiledSurface = true,
1130       .TileWalk = TILEWALK_YMAJOR,
1131 #endif
1132 #if GFX_VER >= 6
1133       .MOCS = isl_mocs(dev, 0, false),
1134 #endif
1135 #if GFX_VER == 7
1136       /* According to PRMs: "Volume 4 Part 1: Subsystem and Cores – Shared
1137        * Functions"
1138        *
1139        * RENDER_SURFACE_STATE::Surface Vertical Alignment
1140        *
1141        *    "This field must be set to VALIGN_4 for all tiled Y Render Target
1142        *     surfaces."
1143        *
1144        * Affect IVB, HSW.
1145        */
1146       .SurfaceVerticalAlignment = VALIGN_4,
1147 #endif
1148       .MIPCountLOD = info->levels,
1149       .Width = info->size.width - 1,
1150       .Height = info->size.height - 1,
1151       .Depth = info->size.depth - 1,
1152       .RenderTargetViewExtent = info->size.depth - 1,
1153 #if GFX_VER <= 5
1154       .MinimumArrayElement = info->minimum_array_element,
1155       .ColorBufferComponentWriteDisables = 0xf,
1156 #endif
1157    };
1158    GENX(RENDER_SURFACE_STATE_pack)(NULL, state, &s);
1159 }
1160