genX_state.c

/*
 * Copyright © 2015 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
 * IN THE SOFTWARE.
 */

#include <assert.h>
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>

#include "anv_private.h"

#include "common/intel_aux_map.h"
#include "common/intel_sample_positions.h"
#include "genxml/gen_macros.h"
#include "genxml/genX_pack.h"

#include "vk_util.h"

/**
 * Compute an \p n x \p m pixel hashing table usable as slice, subslice or
 * pixel pipe hashing table.  The resulting table is the cyclic repetition of
 * a fixed pattern with periodicity equal to \p period.
 *
 * If \p index is specified to be equal to \p period, a 2-way hashing table
 * will be generated such that indices 0 and 1 are returned for the following
 * fractions of entries respectively:
 *
 *   p_0 = ceil(period / 2) / period
 *   p_1 = floor(period / 2) / period
 *
 * If \p index is even and less than \p period, a 3-way hashing table will be
 * generated such that indices 0, 1 and 2 are returned for the following
 * fractions of entries:
 *
 *   p_0 = (ceil(period / 2) - 1) / period
 *   p_1 = floor(period / 2) / period
 *   p_2 = 1 / period
 *
 * The equations above apply if \p flip is equal to 0, if it is equal to 1 p_0
 * and p_1 will be swapped for the result.  Note that in the context of pixel
 * pipe hashing this can be always 0 on Gfx12 platforms, since the hardware
 * transparently remaps logical indices found on the table to physical pixel
 * pipe indices from the highest to lowest EU count.
 */
UNUSED static void
calculate_pixel_hashing_table(unsigned n, unsigned m,
                              unsigned period, unsigned index, bool flip,
                              uint32_t *p)
{
   for (unsigned i = 0; i < n; i++) {
      for (unsigned j = 0; j < m; j++) {
         const unsigned k = (i + j) % period;
         p[j + m * i] = (k == index ? 2 : (k & 1) ^ flip);
      }
   }
}

static void
genX(emit_slice_hashing_state)(struct anv_device *device,
                               struct anv_batch *batch)
{
   device->slice_hash = (struct anv_state) { 0 };

#if GFX_VER == 11
   assert(device->info.ppipe_subslices[2] == 0);

   if (device->info.ppipe_subslices[0] == device->info.ppipe_subslices[1])
     return;

   unsigned size = GENX(SLICE_HASH_TABLE_length) * 4;
   device->slice_hash =
      anv_state_pool_alloc(&device->dynamic_state_pool, size, 64);

   const bool flip = device->info.ppipe_subslices[0] <
                     device->info.ppipe_subslices[1];
   struct GENX(SLICE_HASH_TABLE) table;
   calculate_pixel_hashing_table(16, 16, 3, 3, flip, table.Entry[0]);

   GENX(SLICE_HASH_TABLE_pack)(NULL, device->slice_hash.map, &table);

   anv_batch_emit(batch, GENX(3DSTATE_SLICE_TABLE_STATE_POINTERS), ptr) {
      ptr.SliceHashStatePointerValid = true;
      ptr.SliceHashTableStatePointer = device->slice_hash.offset;
   }

   anv_batch_emit(batch, GENX(3DSTATE_3D_MODE), mode) {
      mode.SliceHashingTableEnable = true;
   }
#elif GFX_VERx10 == 120
   /* For each n calculate ppipes_of[n], equal to the number of pixel pipes
    * present with n active dual subslices.
    */
   unsigned ppipes_of[3] = {};

   for (unsigned n = 0; n < ARRAY_SIZE(ppipes_of); n++) {
      for (unsigned p = 0; p < ARRAY_SIZE(device->info.ppipe_subslices); p++)
         ppipes_of[n] += (device->info.ppipe_subslices[p] == n);
   }

   /* Gfx12 has three pixel pipes. */
   assert(ppipes_of[0] + ppipes_of[1] + ppipes_of[2] == 3);

   if (ppipes_of[2] == 3 || ppipes_of[0] == 2) {
      /* All three pixel pipes have the maximum number of active dual
       * subslices, or there is only one active pixel pipe: Nothing to do.
       */
      return;
   }

   anv_batch_emit(batch, GENX(3DSTATE_SUBSLICE_HASH_TABLE), p) {
      p.SliceHashControl[0] = TABLE_0;

      if (ppipes_of[2] == 2 && ppipes_of[0] == 1)
         calculate_pixel_hashing_table(8, 16, 2, 2, 0, p.TwoWayTableEntry[0]);
      else if (ppipes_of[2] == 1 && ppipes_of[1] == 1 && ppipes_of[0] == 1)
         calculate_pixel_hashing_table(8, 16, 3, 3, 0, p.TwoWayTableEntry[0]);

      if (ppipes_of[2] == 2 && ppipes_of[1] == 1)
         calculate_pixel_hashing_table(8, 16, 5, 4, 0, p.ThreeWayTableEntry[0]);
      else if (ppipes_of[2] == 2 && ppipes_of[0] == 1)
         calculate_pixel_hashing_table(8, 16, 2, 2, 0, p.ThreeWayTableEntry[0]);
      else if (ppipes_of[2] == 1 && ppipes_of[1] == 1 && ppipes_of[0] == 1)
         calculate_pixel_hashing_table(8, 16, 3, 3, 0, p.ThreeWayTableEntry[0]);
      else
         unreachable("Illegal fusing.");
   }

   anv_batch_emit(batch, GENX(3DSTATE_3D_MODE), p) {
      p.SubsliceHashingTableEnable = true;
      p.SubsliceHashingTableEnableMask = true;
   }
#endif
}

static VkResult
init_render_queue_state(struct anv_queue *queue)
{
   struct anv_device *device = queue->device;
   struct anv_batch batch;

   uint32_t cmds[64];
   batch.start = batch.next = cmds;
   batch.end = (void *) cmds + sizeof(cmds);

   anv_batch_emit(&batch, GENX(PIPELINE_SELECT), ps) {
#if GFX_VER >= 9
      ps.MaskBits = GFX_VER >= 12 ? 0x13 : 3;
      ps.MediaSamplerDOPClockGateEnable = GFX_VER >= 12;
#endif
      ps.PipelineSelection = _3D;
   }

#if GFX_VER == 9
   anv_batch_write_reg(&batch, GENX(CACHE_MODE_1), cm1) {
      cm1.FloatBlendOptimizationEnable = true;
      cm1.FloatBlendOptimizationEnableMask = true;
      cm1.MSCRAWHazardAvoidanceBit = true;
      cm1.MSCRAWHazardAvoidanceBitMask = true;
      cm1.PartialResolveDisableInVC = true;
      cm1.PartialResolveDisableInVCMask = true;
   }
#endif

   anv_batch_emit(&batch, GENX(3DSTATE_AA_LINE_PARAMETERS), aa);

   anv_batch_emit(&batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
      rect.ClippedDrawingRectangleYMin = 0;
      rect.ClippedDrawingRectangleXMin = 0;
      rect.ClippedDrawingRectangleYMax = UINT16_MAX;
      rect.ClippedDrawingRectangleXMax = UINT16_MAX;
      rect.DrawingRectangleOriginY = 0;
      rect.DrawingRectangleOriginX = 0;
   }

#if GFX_VER >= 8
   anv_batch_emit(&batch, GENX(3DSTATE_WM_CHROMAKEY), ck);

   genX(emit_sample_pattern)(&batch, 0, NULL);

   /* The BDW+ docs describe how to use the 3DSTATE_WM_HZ_OP instruction in the
    * section titled, "Optimized Depth Buffer Clear and/or Stencil Buffer
    * Clear." It mentions that the packet overrides GPU state for the clear
    * operation and needs to be reset to 0s to clear the overrides. Depending
    * on the kernel, we may not get a context with the state for this packet
    * zeroed. Do it ourselves just in case. We've observed this to prevent a
    * number of GPU hangs on ICL.
    */
   anv_batch_emit(&batch, GENX(3DSTATE_WM_HZ_OP), hzp);
#endif

#if GFX_VER == 11
   /* The default behavior of bit 5 "Headerless Message for Pre-emptable
    * Contexts" in SAMPLER MODE register is set to 0, which means
    * headerless sampler messages are not allowed for pre-emptable
    * contexts. Set the bit 5 to 1 to allow them.
    */
   anv_batch_write_reg(&batch, GENX(SAMPLER_MODE), sm) {
      sm.HeaderlessMessageforPreemptableContexts = true;
      sm.HeaderlessMessageforPreemptableContextsMask = true;
   }

   /* Bit 1 "Enabled Texel Offset Precision Fix" must be set in
    * HALF_SLICE_CHICKEN7 register.
    */
   anv_batch_write_reg(&batch, GENX(HALF_SLICE_CHICKEN7), hsc7) {
      hsc7.EnabledTexelOffsetPrecisionFix = true;
      hsc7.EnabledTexelOffsetPrecisionFixMask = true;
   }

   anv_batch_write_reg(&batch, GENX(TCCNTLREG), tcc) {
      tcc.L3DataPartialWriteMergingEnable = true;
      tcc.ColorZPartialWriteMergingEnable = true;
      tcc.URBPartialWriteMergingEnable = true;
      tcc.TCDisable = true;
   }
#endif
   genX(emit_slice_hashing_state)(device, &batch);

#if GFX_VER >= 11
   /* hardware specification recommends disabling repacking for
    * the compatibility with decompression mechanism in display controller.
    */
   if (device->info.disable_ccs_repack) {
      anv_batch_write_reg(&batch, GENX(CACHE_MODE_0), cm0) {
         cm0.DisableRepackingforCompression = true;
         cm0.DisableRepackingforCompressionMask = true;
      }
   }

   /* an unknown issue is causing vs push constants to become
    * corrupted during object-level preemption. For now, restrict
    * to command buffer level preemption to avoid rendering
    * corruption.
    */
   anv_batch_write_reg(&batch, GENX(CS_CHICKEN1), cc1) {
      cc1.ReplayMode = MidcmdbufferPreemption;
      cc1.ReplayModeMask = true;
   }
#endif

#if GFX_VER == 12
   if (device->info.has_aux_map) {
      uint64_t aux_base_addr = intel_aux_map_get_base(device->aux_map_ctx);
      assert(aux_base_addr % (32 * 1024) == 0);
      anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
         lri.RegisterOffset = GENX(GFX_AUX_TABLE_BASE_ADDR_num);
         lri.DataDWord = aux_base_addr & 0xffffffff;
      }
      anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
         lri.RegisterOffset = GENX(GFX_AUX_TABLE_BASE_ADDR_num) + 4;
         lri.DataDWord = aux_base_addr >> 32;
      }
   }
#endif

   /* Set the "CONSTANT_BUFFER Address Offset Disable" bit, so
    * 3DSTATE_CONSTANT_XS buffer 0 is an absolute address.
    *
    * This is only safe on kernels with context isolation support.
    */
   if (GFX_VER >= 8 && device->physical->has_context_isolation) {
#if GFX_VER >= 9
      anv_batch_write_reg(&batch, GENX(CS_DEBUG_MODE2), csdm2) {
         csdm2.CONSTANT_BUFFERAddressOffsetDisable = true;
         csdm2.CONSTANT_BUFFERAddressOffsetDisableMask = true;
      }
#elif GFX_VER == 8
      anv_batch_write_reg(&batch, GENX(INSTPM), instpm) {
         instpm.CONSTANT_BUFFERAddressOffsetDisable = true;
         instpm.CONSTANT_BUFFERAddressOffsetDisableMask = true;
      }
#endif
   }

#if GFX_VER >= 11
   /* Starting with GFX version 11, SLM is no longer part of the L3$ config
    * so it never changes throughout the lifetime of the VkDevice.
    */
   const struct intel_l3_config *cfg = intel_get_default_l3_config(&device->info);
   genX(emit_l3_config)(&batch, device, cfg);
   device->l3_config = cfg;
#endif

   anv_batch_emit(&batch, GENX(MI_BATCH_BUFFER_END), bbe);

   assert(batch.next <= batch.end);

   return anv_queue_submit_simple_batch(queue, &batch);
}

void
genX(init_physical_device_state)(ASSERTED struct anv_physical_device *device)
{
   assert(device->info.verx10 == GFX_VERx10);
}

VkResult
genX(init_device_state)(struct anv_device *device)
{
   VkResult res;

   for (uint32_t i = 0; i < device->queue_count; i++) {
      struct anv_queue *queue = &device->queues[i];
      switch (queue->family->engine_class) {
      case I915_ENGINE_CLASS_RENDER:
         res = init_render_queue_state(queue);
         break;
      default:
         res = vk_error(VK_ERROR_INITIALIZATION_FAILED);
         break;
      }
      if (res != VK_SUCCESS)
         return res;
   }

   return res;
}

void
genX(emit_l3_config)(struct anv_batch *batch,
                     const struct anv_device *device,
                     const struct intel_l3_config *cfg)
{
   UNUSED const struct intel_device_info *devinfo = &device->info;

#if GFX_VER >= 8

#if GFX_VER >= 12
#define L3_ALLOCATION_REG GENX(L3ALLOC)
#define L3_ALLOCATION_REG_num GENX(L3ALLOC_num)
#else
#define L3_ALLOCATION_REG GENX(L3CNTLREG)
#define L3_ALLOCATION_REG_num GENX(L3CNTLREG_num)
#endif

   anv_batch_write_reg(batch, L3_ALLOCATION_REG, l3cr) {
      if (cfg == NULL) {
#if GFX_VER >= 12
         l3cr.L3FullWayAllocationEnable = true;
#else
         unreachable("Invalid L3$ config");
#endif
      } else {
#if GFX_VER < 11
         l3cr.SLMEnable = cfg->n[INTEL_L3P_SLM];
#endif
#if GFX_VER == 11
         /* Wa_1406697149: Bit 9 "Error Detection Behavior Control" must be
          * set in L3CNTLREG register. The default setting of the bit is not
          * the desirable behavior.
          */
         l3cr.ErrorDetectionBehaviorControl = true;
         l3cr.UseFullWays = true;
#endif /* GFX_VER == 11 */
         assert(cfg->n[INTEL_L3P_IS] == 0);
         assert(cfg->n[INTEL_L3P_C] == 0);
         assert(cfg->n[INTEL_L3P_T] == 0);
         l3cr.URBAllocation = cfg->n[INTEL_L3P_URB];
         l3cr.ROAllocation = cfg->n[INTEL_L3P_RO];
         l3cr.DCAllocation = cfg->n[INTEL_L3P_DC];
         l3cr.AllAllocation = cfg->n[INTEL_L3P_ALL];
      }
   }

#else /* GFX_VER < 8 */

   const bool has_dc = cfg->n[INTEL_L3P_DC] || cfg->n[INTEL_L3P_ALL];
   const bool has_is = cfg->n[INTEL_L3P_IS] || cfg->n[INTEL_L3P_RO] ||
                       cfg->n[INTEL_L3P_ALL];
   const bool has_c = cfg->n[INTEL_L3P_C] || cfg->n[INTEL_L3P_RO] ||
                      cfg->n[INTEL_L3P_ALL];
   const bool has_t = cfg->n[INTEL_L3P_T] || cfg->n[INTEL_L3P_RO] ||
                      cfg->n[INTEL_L3P_ALL];

   assert(!cfg->n[INTEL_L3P_ALL]);

   /* When enabled SLM only uses a portion of the L3 on half of the banks,
    * the matching space on the remaining banks has to be allocated to a
    * client (URB for all validated configurations) set to the
    * lower-bandwidth 2-bank address hashing mode.
    */
   const bool urb_low_bw = cfg->n[INTEL_L3P_SLM] && !devinfo->is_baytrail;
   assert(!urb_low_bw || cfg->n[INTEL_L3P_URB] == cfg->n[INTEL_L3P_SLM]);

   /* Minimum number of ways that can be allocated to the URB. */
   const unsigned n0_urb = devinfo->is_baytrail ? 32 : 0;
   assert(cfg->n[INTEL_L3P_URB] >= n0_urb);

   anv_batch_write_reg(batch, GENX(L3SQCREG1), l3sqc) {
      l3sqc.ConvertDC_UC = !has_dc;
      l3sqc.ConvertIS_UC = !has_is;
      l3sqc.ConvertC_UC = !has_c;
      l3sqc.ConvertT_UC = !has_t;
#if GFX_VERx10 == 75
      l3sqc.L3SQGeneralPriorityCreditInitialization = SQGPCI_DEFAULT;
#else
      l3sqc.L3SQGeneralPriorityCreditInitialization =
         devinfo->is_baytrail ? BYT_SQGPCI_DEFAULT : SQGPCI_DEFAULT;
#endif
      l3sqc.L3SQHighPriorityCreditInitialization = SQHPCI_DEFAULT;
   }

   anv_batch_write_reg(batch, GENX(L3CNTLREG2), l3cr2) {
      l3cr2.SLMEnable = cfg->n[INTEL_L3P_SLM];
      l3cr2.URBLowBandwidth = urb_low_bw;
      l3cr2.URBAllocation = cfg->n[INTEL_L3P_URB] - n0_urb;
#if !GFX_VERx10 == 75
      l3cr2.ALLAllocation = cfg->n[INTEL_L3P_ALL];
#endif
      l3cr2.ROAllocation = cfg->n[INTEL_L3P_RO];
      l3cr2.DCAllocation = cfg->n[INTEL_L3P_DC];
   }

   anv_batch_write_reg(batch, GENX(L3CNTLREG3), l3cr3) {
      l3cr3.ISAllocation = cfg->n[INTEL_L3P_IS];
      l3cr3.ISLowBandwidth = 0;
      l3cr3.CAllocation = cfg->n[INTEL_L3P_C];
      l3cr3.CLowBandwidth = 0;
      l3cr3.TAllocation = cfg->n[INTEL_L3P_T];
      l3cr3.TLowBandwidth = 0;
   }

#if GFX_VERx10 == 75
   if (device->physical->cmd_parser_version >= 4) {
      /* Enable L3 atomics on HSW if we have a DC partition, otherwise keep
       * them disabled to avoid crashing the system hard.
       */
      anv_batch_write_reg(batch, GENX(SCRATCH1), s1) {
         s1.L3AtomicDisable = !has_dc;
      }
      anv_batch_write_reg(batch, GENX(CHICKEN3), c3) {
         c3.L3AtomicDisableMask = true;
         c3.L3AtomicDisable = !has_dc;
      }
   }
#endif /* GFX_VERx10 == 75 */

#endif /* GFX_VER < 8 */
}

void
genX(emit_multisample)(struct anv_batch *batch, uint32_t samples,
                       const VkSampleLocationEXT *locations)
{
   anv_batch_emit(batch, GENX(3DSTATE_MULTISAMPLE), ms) {
      ms.NumberofMultisamples       = __builtin_ffs(samples) - 1;

      ms.PixelLocation              = CENTER;
#if GFX_VER >= 8
      /* The PRM says that this bit is valid only for DX9:
       *
       *    SW can choose to set this bit only for DX9 API. DX10/OGL API's
       *    should not have any effect by setting or not setting this bit.
       */
      ms.PixelPositionOffsetEnable  = false;
#else

      if (locations) {
         switch (samples) {
         case 1:
            INTEL_SAMPLE_POS_1X_ARRAY(ms.Sample, locations);
            break;
         case 2:
            INTEL_SAMPLE_POS_2X_ARRAY(ms.Sample, locations);
            break;
         case 4:
            INTEL_SAMPLE_POS_4X_ARRAY(ms.Sample, locations);
            break;
         case 8:
            INTEL_SAMPLE_POS_8X_ARRAY(ms.Sample, locations);
            break;
         default:
            break;
         }
      } else {
         switch (samples) {
         case 1:
            INTEL_SAMPLE_POS_1X(ms.Sample);
            break;
         case 2:
            INTEL_SAMPLE_POS_2X(ms.Sample);
            break;
         case 4:
            INTEL_SAMPLE_POS_4X(ms.Sample);
            break;
         case 8:
            INTEL_SAMPLE_POS_8X(ms.Sample);
            break;
         default:
            break;
         }
      }
#endif
   }
}

#if GFX_VER >= 8
void
genX(emit_sample_pattern)(struct anv_batch *batch, uint32_t samples,
                          const VkSampleLocationEXT *locations)
{
   /* See the Vulkan 1.0 spec Table 24.1 "Standard sample locations" and
    * VkPhysicalDeviceFeatures::standardSampleLocations.
    */
   anv_batch_emit(batch, GENX(3DSTATE_SAMPLE_PATTERN), sp) {
      if (locations) {
         /* The Skylake PRM Vol. 2a "3DSTATE_SAMPLE_PATTERN" says:
          *
          *    "When programming the sample offsets (for NUMSAMPLES_4 or _8
          *    and MSRASTMODE_xxx_PATTERN), the order of the samples 0 to 3
          *    (or 7 for 8X, or 15 for 16X) must have monotonically increasing
          *    distance from the pixel center. This is required to get the
          *    correct centroid computation in the device."
          *
          * However, the Vulkan spec seems to require that the the samples
          * occur in the order provided through the API. The standard sample
          * patterns have the above property that they have monotonically
          * increasing distances from the center but client-provided ones do
          * not. As long as this only affects centroid calculations as the
          * docs say, we should be ok because OpenGL and Vulkan only require
          * that the centroid be some lit sample and that it's the same for
          * all samples in a pixel; they have no requirement that it be the
          * one closest to center.
          */
         switch (samples) {
         case 1:
            INTEL_SAMPLE_POS_1X_ARRAY(sp._1xSample, locations);
            break;
         case 2:
            INTEL_SAMPLE_POS_2X_ARRAY(sp._2xSample, locations);
            break;
         case 4:
            INTEL_SAMPLE_POS_4X_ARRAY(sp._4xSample, locations);
            break;
         case 8:
            INTEL_SAMPLE_POS_8X_ARRAY(sp._8xSample, locations);
            break;
#if GFX_VER >= 9
         case 16:
            INTEL_SAMPLE_POS_16X_ARRAY(sp._16xSample, locations);
            break;
#endif
         default:
            break;
         }
      } else {
         INTEL_SAMPLE_POS_1X(sp._1xSample);
         INTEL_SAMPLE_POS_2X(sp._2xSample);
         INTEL_SAMPLE_POS_4X(sp._4xSample);
         INTEL_SAMPLE_POS_8X(sp._8xSample);
#if GFX_VER >= 9
         INTEL_SAMPLE_POS_16X(sp._16xSample);
#endif
      }
   }
}
#endif

#if GFX_VER >= 11
void
genX(emit_shading_rate)(struct anv_batch *batch,
                        const struct anv_graphics_pipeline *pipeline,
                        struct anv_state cps_states,
                        struct anv_dynamic_state *dynamic_state)
{
   const struct brw_wm_prog_data *wm_prog_data = get_wm_prog_data(pipeline);
   const bool cps_enable = wm_prog_data && wm_prog_data->per_coarse_pixel_dispatch;

#if GFX_VER == 11
   anv_batch_emit(batch, GENX(3DSTATE_CPS), cps) {
      cps.CoarsePixelShadingMode = cps_enable ? CPS_MODE_CONSTANT : CPS_MODE_NONE;
      if (cps_enable) {
         cps.MinCPSizeX = dynamic_state->fragment_shading_rate.width;
         cps.MinCPSizeY = dynamic_state->fragment_shading_rate.height;
      }
   }
#elif GFX_VER == 12
   for (uint32_t i = 0; i < dynamic_state->viewport.count; i++) {
      uint32_t *cps_state_dwords =
         cps_states.map + GENX(CPS_STATE_length) * 4 * i;
      struct GENX(CPS_STATE) cps_state = {
         .CoarsePixelShadingMode = cps_enable ? CPS_MODE_CONSTANT : CPS_MODE_NONE,
      };

      if (cps_enable) {
         cps_state.MinCPSizeX = dynamic_state->fragment_shading_rate.width;
         cps_state.MinCPSizeY = dynamic_state->fragment_shading_rate.height;
      }

      GENX(CPS_STATE_pack)(NULL, cps_state_dwords, &cps_state);
   }

   anv_batch_emit(batch, GENX(3DSTATE_CPS_POINTERS), cps) {
      cps.CoarsePixelShadingStateArrayPointer = cps_states.offset;
   }
#endif
}
#endif /* GFX_VER >= 11 */

static uint32_t
vk_to_intel_tex_filter(VkFilter filter, bool anisotropyEnable)
{
   switch (filter) {
   default:
      assert(!"Invalid filter");
   case VK_FILTER_NEAREST:
      return anisotropyEnable ? MAPFILTER_ANISOTROPIC : MAPFILTER_NEAREST;
   case VK_FILTER_LINEAR:
      return anisotropyEnable ? MAPFILTER_ANISOTROPIC : MAPFILTER_LINEAR;
   }
}

static uint32_t
vk_to_intel_max_anisotropy(float ratio)
{
   return (anv_clamp_f(ratio, 2, 16) - 2) / 2;
}

static const uint32_t vk_to_intel_mipmap_mode[] = {
   [VK_SAMPLER_MIPMAP_MODE_NEAREST]          = MIPFILTER_NEAREST,
   [VK_SAMPLER_MIPMAP_MODE_LINEAR]           = MIPFILTER_LINEAR
};

static const uint32_t vk_to_intel_tex_address[] = {
   [VK_SAMPLER_ADDRESS_MODE_REPEAT]          = TCM_WRAP,
   [VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT] = TCM_MIRROR,
   [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE]   = TCM_CLAMP,
   [VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE] = TCM_MIRROR_ONCE,
   [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER] = TCM_CLAMP_BORDER,
};

/* Vulkan specifies the result of shadow comparisons as:
 *     1     if   ref <op> texel,
 *     0     otherwise.
 *
 * The hardware does:
 *     0     if texel <op> ref,
 *     1     otherwise.
 *
 * So, these look a bit strange because there's both a negation
 * and swapping of the arguments involved.
 */
static const uint32_t vk_to_intel_shadow_compare_op[] = {
   [VK_COMPARE_OP_NEVER]                        = PREFILTEROP_ALWAYS,
   [VK_COMPARE_OP_LESS]                         = PREFILTEROP_LEQUAL,
   [VK_COMPARE_OP_EQUAL]                        = PREFILTEROP_NOTEQUAL,
   [VK_COMPARE_OP_LESS_OR_EQUAL]                = PREFILTEROP_LESS,
   [VK_COMPARE_OP_GREATER]                      = PREFILTEROP_GEQUAL,
   [VK_COMPARE_OP_NOT_EQUAL]                    = PREFILTEROP_EQUAL,
   [VK_COMPARE_OP_GREATER_OR_EQUAL]             = PREFILTEROP_GREATER,
   [VK_COMPARE_OP_ALWAYS]                       = PREFILTEROP_NEVER,
};

#if GFX_VER >= 9
static const uint32_t vk_to_intel_sampler_reduction_mode[] = {
   [VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT] = STD_FILTER,
   [VK_SAMPLER_REDUCTION_MODE_MIN_EXT]              = MINIMUM,
   [VK_SAMPLER_REDUCTION_MODE_MAX_EXT]              = MAXIMUM,
};
#endif

VkResult genX(CreateSampler)(
    VkDevice                                    _device,
    const VkSamplerCreateInfo*                  pCreateInfo,
    const VkAllocationCallbacks*                pAllocator,
    VkSampler*                                  pSampler)
{
   ANV_FROM_HANDLE(anv_device, device, _device);
   struct anv_sampler *sampler;

   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO);

   sampler = vk_object_zalloc(&device->vk, pAllocator, sizeof(*sampler),
                              VK_OBJECT_TYPE_SAMPLER);
   if (!sampler)
      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);

   sampler->n_planes = 1;

   uint32_t border_color_stride = GFX_VERx10 == 75 ? 512 : 64;
   uint32_t border_color_offset;
   ASSERTED bool has_custom_color = false;
   if (pCreateInfo->borderColor <= VK_BORDER_COLOR_INT_OPAQUE_WHITE) {
      border_color_offset = device->border_colors.offset +
                            pCreateInfo->borderColor *
                            border_color_stride;
   } else {
      assert(GFX_VER >= 8);
      sampler->custom_border_color =
         anv_state_reserved_pool_alloc(&device->custom_border_colors);
      border_color_offset = sampler->custom_border_color.offset;
   }

#if GFX_VER >= 9
   unsigned sampler_reduction_mode = STD_FILTER;
   bool enable_sampler_reduction = false;
#endif

   vk_foreach_struct(ext, pCreateInfo->pNext) {
      switch (ext->sType) {
      case VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO: {
         VkSamplerYcbcrConversionInfo *pSamplerConversion =
            (VkSamplerYcbcrConversionInfo *) ext;
         ANV_FROM_HANDLE(anv_ycbcr_conversion, conversion,
                         pSamplerConversion->conversion);

         /* Ignore conversion for non-YUV formats. This fulfills a requirement
          * for clients that want to utilize same code path for images with
          * external formats (VK_FORMAT_UNDEFINED) and "regular" RGBA images
          * where format is known.
          */
         if (conversion == NULL || !conversion->format->can_ycbcr)
            break;

         sampler->n_planes = conversion->format->n_planes;
         sampler->conversion = conversion;
         break;
      }
#if GFX_VER >= 9
      case VK_STRUCTURE_TYPE_SAMPLER_REDUCTION_MODE_CREATE_INFO: {
         VkSamplerReductionModeCreateInfo *sampler_reduction =
            (VkSamplerReductionModeCreateInfo *) ext;
         sampler_reduction_mode =
            vk_to_intel_sampler_reduction_mode[sampler_reduction->reductionMode];
         enable_sampler_reduction = true;
         break;
      }
#endif
      case VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT: {
         VkSamplerCustomBorderColorCreateInfoEXT *custom_border_color =
            (VkSamplerCustomBorderColorCreateInfoEXT *) ext;
         if (sampler->custom_border_color.map == NULL)
            break;
         struct gfx8_border_color *cbc = sampler->custom_border_color.map;
         if (custom_border_color->format == VK_FORMAT_B4G4R4A4_UNORM_PACK16) {
            /* B4G4R4A4_UNORM_PACK16 is treated as R4G4B4A4_UNORM_PACK16 with
             * a swizzle, but this does not carry over to the sampler for
             * border colors, so we need to do the swizzle ourselves here.
             */
            cbc->uint32[0] = custom_border_color->customBorderColor.uint32[2];
            cbc->uint32[1] = custom_border_color->customBorderColor.uint32[1];
            cbc->uint32[2] = custom_border_color->customBorderColor.uint32[0];
            cbc->uint32[3] = custom_border_color->customBorderColor.uint32[3];
         } else {
            /* Both structs share the same layout, so just copy them over. */
            memcpy(cbc, &custom_border_color->customBorderColor,
                   sizeof(VkClearColorValue));
         }
         has_custom_color = true;
         break;
      }
      default:
         anv_debug_ignored_stype(ext->sType);
         break;
      }
   }

   assert((sampler->custom_border_color.map == NULL) || has_custom_color);

   if (device->physical->has_bindless_samplers) {
      /* If we have bindless, allocate enough samplers.  We allocate 32 bytes
       * for each sampler instead of 16 bytes because we want all bindless
       * samplers to be 32-byte aligned so we don't have to use indirect
       * sampler messages on them.
       */
      sampler->bindless_state =
         anv_state_pool_alloc(&device->dynamic_state_pool,
                              sampler->n_planes * 32, 32);
   }

   for (unsigned p = 0; p < sampler->n_planes; p++) {
      const bool plane_has_chroma =
         sampler->conversion && sampler->conversion->format->planes[p].has_chroma;
      const VkFilter min_filter =
         plane_has_chroma ? sampler->conversion->chroma_filter : pCreateInfo->minFilter;
      const VkFilter mag_filter =
         plane_has_chroma ? sampler->conversion->chroma_filter : pCreateInfo->magFilter;
      const bool enable_min_filter_addr_rounding = min_filter != VK_FILTER_NEAREST;
      const bool enable_mag_filter_addr_rounding = mag_filter != VK_FILTER_NEAREST;
      /* From Broadwell PRM, SAMPLER_STATE:
       *   "Mip Mode Filter must be set to MIPFILTER_NONE for Planar YUV surfaces."
       */
      const bool isl_format_is_planar_yuv = sampler->conversion &&
         isl_format_is_yuv(sampler->conversion->format->planes[0].isl_format) &&
         isl_format_is_planar(sampler->conversion->format->planes[0].isl_format);

      const uint32_t mip_filter_mode =
         isl_format_is_planar_yuv ?
         MIPFILTER_NONE : vk_to_intel_mipmap_mode[pCreateInfo->mipmapMode];

      struct GENX(SAMPLER_STATE) sampler_state = {
         .SamplerDisable = false,
         .TextureBorderColorMode = DX10OGL,

#if GFX_VER >= 11
         .CPSLODCompensationEnable = true,
#endif

#if GFX_VER >= 8
         .LODPreClampMode = CLAMP_MODE_OGL,
#else
         .LODPreClampEnable = CLAMP_ENABLE_OGL,
#endif

#if GFX_VER == 8
         .BaseMipLevel = 0.0,
#endif
         .MipModeFilter = mip_filter_mode,
         .MagModeFilter = vk_to_intel_tex_filter(mag_filter, pCreateInfo->anisotropyEnable),
         .MinModeFilter = vk_to_intel_tex_filter(min_filter, pCreateInfo->anisotropyEnable),
         .TextureLODBias = anv_clamp_f(pCreateInfo->mipLodBias, -16, 15.996),
         .AnisotropicAlgorithm =
            pCreateInfo->anisotropyEnable ? EWAApproximation : LEGACY,
         .MinLOD = anv_clamp_f(pCreateInfo->minLod, 0, 14),
         .MaxLOD = anv_clamp_f(pCreateInfo->maxLod, 0, 14),
         .ChromaKeyEnable = 0,
         .ChromaKeyIndex = 0,
         .ChromaKeyMode = 0,
         .ShadowFunction =
            vk_to_intel_shadow_compare_op[pCreateInfo->compareEnable ?
                                        pCreateInfo->compareOp : VK_COMPARE_OP_NEVER],
         .CubeSurfaceControlMode = OVERRIDE,

         .BorderColorPointer = border_color_offset,

#if GFX_VER >= 8
         .LODClampMagnificationMode = MIPNONE,
#endif

         .MaximumAnisotropy = vk_to_intel_max_anisotropy(pCreateInfo->maxAnisotropy),
         .RAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
         .RAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
         .VAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
         .VAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
         .UAddressMinFilterRoundingEnable = enable_min_filter_addr_rounding,
         .UAddressMagFilterRoundingEnable = enable_mag_filter_addr_rounding,
         .TrilinearFilterQuality = 0,
         .NonnormalizedCoordinateEnable = pCreateInfo->unnormalizedCoordinates,
         .TCXAddressControlMode = vk_to_intel_tex_address[pCreateInfo->addressModeU],
         .TCYAddressControlMode = vk_to_intel_tex_address[pCreateInfo->addressModeV],
         .TCZAddressControlMode = vk_to_intel_tex_address[pCreateInfo->addressModeW],

#if GFX_VER >= 9
         .ReductionType = sampler_reduction_mode,
         .ReductionTypeEnable = enable_sampler_reduction,
#endif
      };

      GENX(SAMPLER_STATE_pack)(NULL, sampler->state[p], &sampler_state);

      if (sampler->bindless_state.map) {
         memcpy(sampler->bindless_state.map + p * 32,
                sampler->state[p], GENX(SAMPLER_STATE_length) * 4);
      }
   }

   *pSampler = anv_sampler_to_handle(sampler);

   return VK_SUCCESS;
}