From fb5abb956d63691b56806bf534222ae204f892be Mon Sep 17 00:00:00 2001
From: Jesse Natalie <jenatali@microsoft.com>
Date: Thu, 30 Mar 2023 14:44:11 -0700
Subject: [PATCH] dzn: Add a driconf option for enabling subgroup ops in VS/GS

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22225>
---
 src/microsoft/vulkan/dzn_device.c | 10 +++++++---
 src/util/00-mesa-defaults.conf    |  1 +
 src/util/driconf.h                |  3 +++
 3 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/src/microsoft/vulkan/dzn_device.c b/src/microsoft/vulkan/dzn_device.c
index 1ad89cd..efbea8d 100644
--- a/src/microsoft/vulkan/dzn_device.c
+++ b/src/microsoft/vulkan/dzn_device.c
@@ -1136,6 +1136,9 @@ static const driOptionDescription dzn_dri_options[] = {
    DRI_CONF_SECTION_DEBUG
       DRI_CONF_DZN_CLAIM_WIDE_LINES(false)
       DRI_CONF_DZN_ENABLE_8BIT_LOADS_STORES(false)
+      /* Default-disabled because the CTS doesn't check subgroupQuadOperationsInAllStages
+       * and tries to do quad ops in VS/GS which is unsupported. */
+      DRI_CONF_DZN_ENABLE_SUBGROUP_OPS_IN_VTX_PIPELINE(false)
    DRI_CONF_SECTION_END
 };
 
@@ -1586,6 +1589,7 @@ dzn_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice,
                                  VkPhysicalDeviceProperties2 *pProperties)
 {
    VK_FROM_HANDLE(dzn_physical_device, pdevice, physicalDevice);
+   struct dzn_instance *instance = container_of(pdevice->vk.instance, struct dzn_instance, vk);
 
    (void)dzn_physical_device_get_d3d12_dev(pdevice);
 
@@ -1777,9 +1781,9 @@ dzn_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice,
                                      VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT |
                                      VK_SUBGROUP_FEATURE_QUAD_BIT |
                                      VK_SUBGROUP_FEATURE_ARITHMETIC_BIT,
-      /* Note: The CTS doesn't seem to respect the subgroupQuadOperationsInAllStages bit, and it
-       * seems more useful to support quad ops in FS/CS than subgroup ops at all in VS/GS. */
-      .subgroupSupportedStages = VK_SHADER_STAGE_FRAGMENT_BIT | VK_SHADER_STAGE_COMPUTE_BIT,
+      .subgroupSupportedStages = VK_SHADER_STAGE_FRAGMENT_BIT | VK_SHADER_STAGE_COMPUTE_BIT |
+                                 (driQueryOptionb(&instance->dri_options, "dzn_enable_subgroup_ops_in_vtx_pipeline") ?
+                                    (VK_SHADER_STAGE_GEOMETRY_BIT | VK_SHADER_STAGE_VERTEX_BIT) : 0),
       .subgroupQuadOperationsInAllStages = false,
       .subgroupSize = pdevice->options1.WaveOps ? pdevice->options1.WaveLaneCountMin : 1,
    };
diff --git a/src/util/00-mesa-defaults.conf b/src/util/00-mesa-defaults.conf
index 3452f16..f39825c 100644
--- a/src/util/00-mesa-defaults.conf
+++ b/src/util/00-mesa-defaults.conf
@@ -1043,6 +1043,7 @@ TODO: document the other workarounds.
     <device driver="dzn">
         <application name="DOOMEternal" executable="DOOMEternalx64vk.exe">
             <option name="dzn_enable_8bit_loads_stores" value="true" />
+            <option name="dzn_enable_subgroup_ops_in_vtx_pipeline" value="true" />
             <option name="dzn_claim_wide_lines" value="true" />
         </application>
         <application name="No Man's Sky" executable="NMS.exe">
diff --git a/src/util/driconf.h b/src/util/driconf.h
index 83b2533..32be1d7 100644
--- a/src/util/driconf.h
+++ b/src/util/driconf.h
@@ -664,4 +664,7 @@
 #define DRI_CONF_DZN_ENABLE_8BIT_LOADS_STORES(def) \
    DRI_CONF_OPT_B(dzn_enable_8bit_loads_stores, def, "Enable VK_KHR_8bit_loads_stores")
 
+#define DRI_CONF_DZN_ENABLE_SUBGROUP_OPS_IN_VTX_PIPELINE(def) \
+   DRI_CONF_OPT_B(dzn_enable_subgroup_ops_in_vtx_pipeline, def, "Enable subgroup ops in pre-rasterizer stages (VS/GS)")
+
 #endif
-- 
2.7.4