anv/cmd_buffer: Move num_workgroups to compute state
[mesa.git] / src / intel / vulkan / genX_state.c
index 91da05cddbfd4fe2a436955588bcfeaa24a4d2ef..54fb8634fdc84ffae839f15fe23c6d0da0330264 100644 (file)
 #include "genxml/gen_macros.h"
 #include "genxml/genX_pack.h"
 
+#include "vk_util.h"
+
+#if GEN_GEN == 10
+/**
+ * From Gen10 Workarounds page in h/w specs:
+ * WaSampleOffsetIZ:
+ *    "Prior to the 3DSTATE_SAMPLE_PATTERN driver must ensure there are no
+ *     markers in the pipeline by programming a PIPE_CONTROL with stall."
+ */
+static void
+gen10_emit_wa_cs_stall_flush(struct anv_batch *batch)
+{
+
+   anv_batch_emit(batch, GENX(PIPE_CONTROL), pc) {
+      pc.CommandStreamerStallEnable = true;
+      pc.StallAtPixelScoreboard = true;
+   }
+}
+
+/**
+ * From Gen10 Workarounds page in h/w specs:
+ * WaSampleOffsetIZ:_cs_stall_flush
+ *    "When 3DSTATE_SAMPLE_PATTERN is programmed, driver must then issue an
+ *     MI_LOAD_REGISTER_IMM command to an offset between 0x7000 and 0x7FFF(SVL)
+ *     after the command to ensure the state has been delivered prior to any
+ *     command causing a marker in the pipeline."
+ */
+static void
+gen10_emit_wa_lri_to_cache_mode_zero(struct anv_batch *batch)
+{
+   /* Before changing the value of CACHE_MODE_0 register, GFX pipeline must
+    * be idle; i.e., full flush is required.
+    */
+   anv_batch_emit(batch, GENX(PIPE_CONTROL), pc) {
+      pc.DepthCacheFlushEnable = true;
+      pc.DCFlushEnable = true;
+      pc.RenderTargetCacheFlushEnable = true;
+      pc.InstructionCacheInvalidateEnable = true;
+      pc.StateCacheInvalidationEnable = true;
+      pc.TextureCacheInvalidationEnable = true;
+      pc.VFCacheInvalidationEnable = true;
+      pc.ConstantCacheInvalidationEnable =true;
+   }
+
+   /* Write to CACHE_MODE_0 (0x7000) */
+   uint32_t cache_mode_0 = 0;
+   anv_pack_struct(&cache_mode_0, GENX(CACHE_MODE_0));
+
+   anv_batch_emit(batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
+      lri.RegisterOffset = GENX(CACHE_MODE_0_num);
+      lri.DataDWord      = cache_mode_0;
+   }
+}
+#endif
+
 VkResult
 genX(init_device_state)(struct anv_device *device)
 {
@@ -66,6 +121,18 @@ genX(init_device_state)(struct anv_device *device)
    }
 #endif
 
+#if GEN_GEN == 10
+   uint32_t cache_mode_ss;
+   anv_pack_struct(&cache_mode_ss, GENX(CACHE_MODE_SS),
+                   .FloatBlendOptimizationEnable = true,
+                   .FloatBlendOptimizationEnableMask = true);
+
+   anv_batch_emit(&batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
+      lri.RegisterOffset = GENX(CACHE_MODE_SS_num);
+      lri.DataDWord      = cache_mode_ss;
+   }
+#endif
+
    anv_batch_emit(&batch, GENX(3DSTATE_AA_LINE_PARAMETERS), aa);
 
    anv_batch_emit(&batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
@@ -80,6 +147,10 @@ genX(init_device_state)(struct anv_device *device)
 #if GEN_GEN >= 8
    anv_batch_emit(&batch, GENX(3DSTATE_WM_CHROMAKEY), ck);
 
+#if GEN_GEN == 10
+   gen10_emit_wa_cs_stall_flush(&batch);
+#endif
+
    /* See the Vulkan 1.0 spec Table 24.1 "Standard sample locations" and
     * VkPhysicalDeviceFeatures::standardSampleLocations.
     */
@@ -94,6 +165,10 @@ genX(init_device_state)(struct anv_device *device)
    }
 #endif
 
+#if GEN_GEN == 10
+   gen10_emit_wa_lri_to_cache_mode_zero(&batch);
+#endif
+
    anv_batch_emit(&batch, GENX(MI_BATCH_BUFFER_END), bbe);
 
    assert(batch.next <= batch.end);
@@ -176,12 +251,44 @@ VkResult genX(CreateSampler)(
    uint32_t border_color_offset = device->border_colors.offset +
                                   pCreateInfo->borderColor * 64;
 
-   bool enable_min_filter_addr_rounding =
-      pCreateInfo->minFilter != VK_FILTER_NEAREST;
-   bool enable_mag_filter_addr_rounding =
-      pCreateInfo->magFilter != VK_FILTER_NEAREST;
+   vk_foreach_struct(ext, pCreateInfo->pNext) {
+      switch (ext->sType) {
+      case VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO_KHR: {
+         VkSamplerYcbcrConversionInfoKHR *pSamplerConversion =
+            (VkSamplerYcbcrConversionInfoKHR *) ext;
+         ANV_FROM_HANDLE(anv_ycbcr_conversion, conversion,
+                         pSamplerConversion->conversion);
+
+         if (conversion == NULL)
+            break;
+
+         sampler->n_planes = conversion->format->n_planes;
+         sampler->conversion = conversion;
+         break;
+      }
+      default:
+         anv_debug_ignored_stype(ext->sType);
+         break;
+      }
+   }
 
    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 uint32_t mip_filter_mode =
+         (sampler->conversion &&
+          isl_format_is_yuv(sampler->conversion->format->planes[0].isl_format)) ?
+         MIPFILTER_NONE : vk_to_gen_mipmap_mode[pCreateInfo->mipmapMode];
+
       struct GENX(SAMPLER_STATE) sampler_state = {
          .SamplerDisable = false,
          .TextureBorderColorMode = DX10OGL,
@@ -195,11 +302,9 @@ VkResult genX(CreateSampler)(
 #if GEN_GEN == 8
          .BaseMipLevel = 0.0,
 #endif
-         .MipModeFilter = vk_to_gen_mipmap_mode[pCreateInfo->mipmapMode],
-         .MagModeFilter = vk_to_gen_tex_filter(pCreateInfo->magFilter,
-                                               pCreateInfo->anisotropyEnable),
-         .MinModeFilter = vk_to_gen_tex_filter(pCreateInfo->minFilter,
-                                               pCreateInfo->anisotropyEnable),
+         .MipModeFilter = mip_filter_mode,
+         .MagModeFilter = vk_to_gen_tex_filter(mag_filter, pCreateInfo->anisotropyEnable),
+         .MinModeFilter = vk_to_gen_tex_filter(min_filter, pCreateInfo->anisotropyEnable),
          .TextureLODBias = anv_clamp_f(pCreateInfo->mipLodBias, -16, 15.996),
          .AnisotropicAlgorithm = EWAApproximation,
          .MinLOD = anv_clamp_f(pCreateInfo->minLod, 0, 14),