src/intel/vulkan/genX_state.c

   1 /*
   2  * Copyright © 2015 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 <assert.h>
  25 #include <stdbool.h>
  26 #include <string.h>
  27 #include <unistd.h>
  28 #include <fcntl.h>
  29
  30 #include "anv_private.h"
  31
  32 #include "common/gen_sample_positions.h"
  33 #include "genxml/gen_macros.h"
  34 #include "genxml/genX_pack.h"
  35
  36 #include "vk_util.h"
  37
  38 #if GEN_GEN == 10
  39 /**
  40  * From Gen10 Workarounds page in h/w specs:
  41  * WaSampleOffsetIZ:
  42  *    "Prior to the 3DSTATE_SAMPLE_PATTERN driver must ensure there are no
  43  *     markers in the pipeline by programming a PIPE_CONTROL with stall."
  44  */
  45 static void
  46 gen10_emit_wa_cs_stall_flush(struct anv_batch *batch)
  47 {
  48
  49    anv_batch_emit(batch, GENX(PIPE_CONTROL), pc) {
  50       pc.CommandStreamerStallEnable = true;
  51       pc.StallAtPixelScoreboard = true;
  52    }
  53 }
  54
  55 /**
  56  * From Gen10 Workarounds page in h/w specs:
  57  * WaSampleOffsetIZ:_cs_stall_flush
  58  *    "When 3DSTATE_SAMPLE_PATTERN is programmed, driver must then issue an
  59  *     MI_LOAD_REGISTER_IMM command to an offset between 0x7000 and 0x7FFF(SVL)
  60  *     after the command to ensure the state has been delivered prior to any
  61  *     command causing a marker in the pipeline."
  62  */
  63 static void
  64 gen10_emit_wa_lri_to_cache_mode_zero(struct anv_batch *batch)
  65 {
  66    /* Before changing the value of CACHE_MODE_0 register, GFX pipeline must
  67     * be idle; i.e., full flush is required.
  68     */
  69    anv_batch_emit(batch, GENX(PIPE_CONTROL), pc) {
  70       pc.DepthCacheFlushEnable = true;
  71       pc.DCFlushEnable = true;
  72       pc.RenderTargetCacheFlushEnable = true;
  73       pc.InstructionCacheInvalidateEnable = true;
  74       pc.StateCacheInvalidationEnable = true;
  75       pc.TextureCacheInvalidationEnable = true;
  76       pc.VFCacheInvalidationEnable = true;
  77       pc.ConstantCacheInvalidationEnable =true;
  78    }
  79
  80    /* Write to CACHE_MODE_0 (0x7000) */
  81    uint32_t cache_mode_0 = 0;
  82    anv_pack_struct(&cache_mode_0, GENX(CACHE_MODE_0));
  83
  84    anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
  85       lri.RegisterOffset = GENX(CACHE_MODE_0_num);
  86       lri.DataDWord      = cache_mode_0;
  87    }
  88 }
  89 #endif
  90
  91 VkResult
  92 genX(init_device_state)(struct anv_device *device)
  93 {
  94    GENX(MEMORY_OBJECT_CONTROL_STATE_pack)(NULL, &device->default_mocs,
  95                                           &GENX(MOCS));
  96
  97    struct anv_batch batch;
  98
  99    uint32_t cmds[64];
 100    batch.start = batch.next = cmds;
 101    batch.end = (void *) cmds + sizeof(cmds);
 102
 103    anv_batch_emit(&batch, GENX(PIPELINE_SELECT), ps) {
 104 #if GEN_GEN >= 9
 105       ps.MaskBits = 3;
 106 #endif
 107       ps.PipelineSelection = _3D;
 108    }
 109
 110 #if GEN_GEN == 9
 111    uint32_t cache_mode_1;
 112    anv_pack_struct(&cache_mode_1, GENX(CACHE_MODE_1),
 113                    .FloatBlendOptimizationEnable = true,
 114                    .FloatBlendOptimizationEnableMask = true,
 115                    .PartialResolveDisableInVC = true,
 116                    .PartialResolveDisableInVCMask = true);
 117
 118    anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
 119       lri.RegisterOffset = GENX(CACHE_MODE_1_num);
 120       lri.DataDWord      = cache_mode_1;
 121    }
 122 #endif
 123
 124 #if GEN_GEN == 10 || GEN_GEN == 11
 125    uint32_t cache_mode_ss;
 126    anv_pack_struct(&cache_mode_ss, GENX(CACHE_MODE_SS),
 127                    .FloatBlendOptimizationEnable = true,
 128                    .FloatBlendOptimizationEnableMask = true);
 129
 130    anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
 131       lri.RegisterOffset = GENX(CACHE_MODE_SS_num);
 132       lri.DataDWord      = cache_mode_ss;
 133    }
 134 #endif
 135
 136    anv_batch_emit(&batch, GENX(3DSTATE_AA_LINE_PARAMETERS), aa);
 137
 138    anv_batch_emit(&batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
 139       rect.ClippedDrawingRectangleYMin = 0;
 140       rect.ClippedDrawingRectangleXMin = 0;
 141       rect.ClippedDrawingRectangleYMax = UINT16_MAX;
 142       rect.ClippedDrawingRectangleXMax = UINT16_MAX;
 143       rect.DrawingRectangleOriginY = 0;
 144       rect.DrawingRectangleOriginX = 0;
 145    }
 146
 147 #if GEN_GEN >= 8
 148    anv_batch_emit(&batch, GENX(3DSTATE_WM_CHROMAKEY), ck);
 149
 150 #if GEN_GEN == 10
 151    gen10_emit_wa_cs_stall_flush(&batch);
 152 #endif
 153
 154    /* See the Vulkan 1.0 spec Table 24.1 "Standard sample locations" and
 155     * VkPhysicalDeviceFeatures::standardSampleLocations.
 156     */
 157    anv_batch_emit(&batch, GENX(3DSTATE_SAMPLE_PATTERN), sp) {
 158       GEN_SAMPLE_POS_1X(sp._1xSample);
 159       GEN_SAMPLE_POS_2X(sp._2xSample);
 160       GEN_SAMPLE_POS_4X(sp._4xSample);
 161       GEN_SAMPLE_POS_8X(sp._8xSample);
 162 #if GEN_GEN >= 9
 163       GEN_SAMPLE_POS_16X(sp._16xSample);
 164 #endif
 165    }
 166 #endif
 167
 168 #if GEN_GEN == 10
 169    gen10_emit_wa_lri_to_cache_mode_zero(&batch);
 170 #endif
 171
 172    /* Set the "CONSTANT_BUFFER Address Offset Disable" bit, so
 173     * 3DSTATE_CONSTANT_XS buffer 0 is an absolute address.
 174     *
 175     * This is only safe on kernels with context isolation support.
 176     */
 177    if (GEN_GEN >= 8 &&
 178        device->instance->physicalDevice.has_context_isolation) {
 179       UNUSED uint32_t tmp_reg;
 180 #if GEN_GEN >= 9
 181       anv_pack_struct(&tmp_reg, GENX(CS_DEBUG_MODE2),
 182                       .CONSTANT_BUFFERAddressOffsetDisable = true,
 183                       .CONSTANT_BUFFERAddressOffsetDisableMask = true);
 184       anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
 185          lri.RegisterOffset = GENX(CS_DEBUG_MODE2_num);
 186          lri.DataDWord      = tmp_reg;
 187       }
 188 #elif GEN_GEN == 8
 189       anv_pack_struct(&tmp_reg, GENX(INSTPM),
 190                       .CONSTANT_BUFFERAddressOffsetDisable = true,
 191                       .CONSTANT_BUFFERAddressOffsetDisableMask = true);
 192       anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
 193          lri.RegisterOffset = GENX(INSTPM_num);
 194          lri.DataDWord      = tmp_reg;
 195       }
 196 #endif
 197    }
 198
 199    anv_batch_emit(&batch, GENX(MI_BATCH_BUFFER_END), bbe);
 200
 201    assert(batch.next <= batch.end);
 202
 203    return anv_device_submit_simple_batch(device, &batch);
 204 }
 205
 206 static uint32_t
 207 vk_to_gen_tex_filter(VkFilter filter, bool anisotropyEnable)
 208 {
 209    switch (filter) {
 210    default:
 211       assert(!"Invalid filter");
 212    case VK_FILTER_NEAREST:
 213       return anisotropyEnable ? MAPFILTER_ANISOTROPIC : MAPFILTER_NEAREST;
 214    case VK_FILTER_LINEAR:
 215       return anisotropyEnable ? MAPFILTER_ANISOTROPIC : MAPFILTER_LINEAR;
 216    }
 217 }
 218
 219 static uint32_t
 220 vk_to_gen_max_anisotropy(float ratio)
 221 {
 222    return (anv_clamp_f(ratio, 2, 16) - 2) / 2;
 223 }
 224
 225 static const uint32_t vk_to_gen_mipmap_mode[] = {
 226    [VK_SAMPLER_MIPMAP_MODE_NEAREST]          = MIPFILTER_NEAREST,
 227    [VK_SAMPLER_MIPMAP_MODE_LINEAR]           = MIPFILTER_LINEAR
 228 };
 229
 230 static const uint32_t vk_to_gen_tex_address[] = {
 231    [VK_SAMPLER_ADDRESS_MODE_REPEAT]          = TCM_WRAP,
 232    [VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT] = TCM_MIRROR,
 233    [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE]   = TCM_CLAMP,
 234    [VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE] = TCM_MIRROR_ONCE,
 235    [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER] = TCM_CLAMP_BORDER,
 236 };
 237
 238 /* Vulkan specifies the result of shadow comparisons as:
 239  *     1     if   ref <op> texel,
 240  *     0     otherwise.
 241  *
 242  * The hardware does:
 243  *     0     if texel <op> ref,
 244  *     1     otherwise.
 245  *
 246  * So, these look a bit strange because there's both a negation
 247  * and swapping of the arguments involved.
 248  */
 249 static const uint32_t vk_to_gen_shadow_compare_op[] = {
 250    [VK_COMPARE_OP_NEVER]                        = PREFILTEROPALWAYS,
 251    [VK_COMPARE_OP_LESS]                         = PREFILTEROPLEQUAL,
 252    [VK_COMPARE_OP_EQUAL]                        = PREFILTEROPNOTEQUAL,
 253    [VK_COMPARE_OP_LESS_OR_EQUAL]                = PREFILTEROPLESS,
 254    [VK_COMPARE_OP_GREATER]                      = PREFILTEROPGEQUAL,
 255    [VK_COMPARE_OP_NOT_EQUAL]                    = PREFILTEROPEQUAL,
 256    [VK_COMPARE_OP_GREATER_OR_EQUAL]             = PREFILTEROPGREATER,
 257    [VK_COMPARE_OP_ALWAYS]                       = PREFILTEROPNEVER,
 258 };
 259
 260 VkResult genX(CreateSampler)(
 261     VkDevice                                    _device,
 262     const VkSamplerCreateInfo*                  pCreateInfo,
 263     const VkAllocationCallbacks*                pAllocator,
 264     VkSampler*                                  pSampler)
 265 {
 266    ANV_FROM_HANDLE(anv_device, device, _device);
 267    struct anv_sampler *sampler;
 268
 269    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);
 270
 271    sampler = vk_zalloc2(&device->alloc, pAllocator, sizeof(*sampler), 8,
 272                         VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 273    if (!sampler)
 274       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 275
 276    sampler->n_planes = 1;
 277
 278    uint32_t border_color_offset = device->border_colors.offset +
 279                                   pCreateInfo->borderColor * 64;
 280
 281    vk_foreach_struct(ext, pCreateInfo->pNext) {
 282       switch (ext->sType) {
 283       case VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO: {
 284          VkSamplerYcbcrConversionInfo *pSamplerConversion =
 285             (VkSamplerYcbcrConversionInfo *) ext;
 286          ANV_FROM_HANDLE(anv_ycbcr_conversion, conversion,
 287                          pSamplerConversion->conversion);
 288
 289          if (conversion == NULL)
 290             break;
 291
 292          sampler->n_planes = conversion->format->n_planes;
 293          sampler->conversion = conversion;
 294          break;
 295       }
 296       default:
 297          anv_debug_ignored_stype(ext->sType);
 298          break;
 299       }
 300    }
 301
 302    for (unsigned p = 0; p < sampler->n_planes; p++) {
 303       const bool plane_has_chroma =
 304          sampler->conversion && sampler->conversion->format->planes[p].has_chroma;
 305       const VkFilter min_filter =
 306          plane_has_chroma ? sampler->conversion->chroma_filter : pCreateInfo->minFilter;
 307       const VkFilter mag_filter =
 308          plane_has_chroma ? sampler->conversion->chroma_filter : pCreateInfo->magFilter;
 309       const bool enable_min_filter_addr_rounding = min_filter != VK_FILTER_NEAREST;
 310       const bool enable_mag_filter_addr_rounding = mag_filter != VK_FILTER_NEAREST;
 311       /* From Broadwell PRM, SAMPLER_STATE:
 312        *   "Mip Mode Filter must be set to MIPFILTER_NONE for Planar YUV surfaces."
 313        */
 314       const uint32_t mip_filter_mode =
 315          (sampler->conversion &&
 316           isl_format_is_yuv(sampler->conversion->format->planes[0].isl_format)) ?
 317          MIPFILTER_NONE : vk_to_gen_mipmap_mode[pCreateInfo->mipmapMode];
 318
 319       struct GENX(SAMPLER_STATE) sampler_state = {
 320          .SamplerDisable = false,
 321          .TextureBorderColorMode = DX10OGL,
 322
 323 #if GEN_GEN >= 8
 324          .LODPreClampMode = CLAMP_MODE_OGL,
 325 #else
 326          .LODPreClampEnable = CLAMP_ENABLE_OGL,
 327 #endif
 328
 329 #if GEN_GEN == 8
 330          .BaseMipLevel = 0.0,
 331 #endif
 332          .MipModeFilter = mip_filter_mode,
 333          .MagModeFilter = vk_to_gen_tex_filter(mag_filter, pCreateInfo->anisotropyEnable),
 334          .MinModeFilter = vk_to_gen_tex_filter(min_filter, pCreateInfo->anisotropyEnable),
 335          .TextureLODBias = anv_clamp_f(pCreateInfo->mipLodBias, -16, 15.996),
 336          .AnisotropicAlgorithm = EWAApproximation,
 337          .MinLOD = anv_clamp_f(pCreateInfo->minLod, 0, 14),
 338          .MaxLOD = anv_clamp_f(pCreateInfo->maxLod, 0, 14),
 339          .ChromaKeyEnable = 0,
 340          .ChromaKeyIndex = 0,
 341          .ChromaKeyMode = 0,
 342          .ShadowFunction = vk_to_gen_shadow_compare_op[pCreateInfo->compareOp],
 343          .CubeSurfaceControlMode = OVERRIDE,
 344
 345          .BorderColorPointer = border_color_offset,
 346
 347 #if GEN_GEN >= 8
 348          .LODClampMagnificationMode = MIPNONE,
 349 #endif
 350
 351          .MaximumAnisotropy = vk_to_gen_max_anisotropy(pCreateInfo->maxAnisotropy),
 352          .RAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
 353          .RAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
 354          .VAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
 355          .VAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
 356          .UAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
 357          .UAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
 358          .TrilinearFilterQuality = 0,
 359          .NonnormalizedCoordinateEnable = pCreateInfo->unnormalizedCoordinates,
 360          .TCXAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressModeU],
 361          .TCYAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressModeV],
 362          .TCZAddressControlMode = vk_to_gen_tex_address[pCreateInfo->addressModeW],
 363       };
 364
 365       GENX(SAMPLER_STATE_pack)(NULL, sampler->state[p], &sampler_state);
 366    }
 367
 368    *pSampler = anv_sampler_to_handle(sampler);
 369
 370    return VK_SUCCESS;
 371 }