From b3c9c3545d6ec97e3cad2b889ebfe932c13bcf67 Mon Sep 17 00:00:00 2001
From: Lynne <dev@lynne.ee>
Date: Sat, 12 Aug 2023 10:46:45 +0000
Subject: [PATCH] vulkan: enable VK_KHR_cooperative_matrix
It's of interest to API users, and of interest to us,
as a DCT/DST can be implemented via matrix multiplies.
---
configure | 4 ++--
libavutil/hwcontext_vulkan.c | 14 ++++++++++++--
libavutil/vulkan.c | 25 ++++++++++++++++++++++++-
libavutil/vulkan.h | 4 ++++
libavutil/vulkan_functions.h | 2 ++
libavutil/vulkan_loader.h | 1 +
6 files changed, 45 insertions(+), 5 deletions(-)
@@ -7144,8 +7144,8 @@ enabled crystalhd && check_lib crystalhd "stdint.h libcrystalhd/libcrystalhd_if.
"in maintaining it."
if enabled vulkan; then
- check_pkg_config_header_only vulkan "vulkan >= 1.3.238" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
- check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 238)"
+ check_pkg_config_header_only vulkan "vulkan >= 1.3.255" "vulkan/vulkan.h" "defined VK_VERSION_1_3" ||
+ check_cpp_condition vulkan "vulkan/vulkan.h" "defined(VK_VERSION_1_4) || (defined(VK_VERSION_1_3) && VK_HEADER_VERSION >= 255)"
fi
if enabled x86; then
@@ -99,6 +99,7 @@ typedef struct VulkanDevicePriv {
VkPhysicalDeviceVulkan13Features device_features_1_3;
VkPhysicalDeviceDescriptorBufferFeaturesEXT desc_buf_features;
VkPhysicalDeviceShaderAtomicFloatFeaturesEXT atomic_float_features;
+ VkPhysicalDeviceCooperativeMatrixFeaturesKHR coop_matrix_features;
/* Queues */
pthread_mutex_t **qf_mutex;
@@ -405,6 +406,7 @@ static const VulkanOptExtension optional_device_exts[] = {
{ VK_EXT_DESCRIPTOR_BUFFER_EXTENSION_NAME, FF_VK_EXT_DESCRIPTOR_BUFFER, },
{ VK_EXT_PHYSICAL_DEVICE_DRM_EXTENSION_NAME, FF_VK_EXT_DEVICE_DRM },
{ VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME, FF_VK_EXT_ATOMIC_FLOAT },
+ { VK_KHR_COOPERATIVE_MATRIX_EXTENSION_NAME, FF_VK_EXT_COOP_MATRIX },
/* Imports/exports */
{ VK_KHR_EXTERNAL_MEMORY_FD_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_FD_MEMORY },
@@ -1202,9 +1204,13 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx,
VkPhysicalDeviceTimelineSemaphoreFeatures timeline_features = {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES,
};
+ VkPhysicalDeviceCooperativeMatrixFeaturesKHR coop_matrix_features = {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_FEATURES_KHR,
+ .pNext = &timeline_features,
+ };
VkPhysicalDeviceShaderAtomicFloatFeaturesEXT atomic_float_features = {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_FLOAT_FEATURES_EXT,
- .pNext = &timeline_features,
+ .pNext = &coop_matrix_features,
};
VkPhysicalDeviceDescriptorBufferFeaturesEXT desc_buf_features = {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_BUFFER_FEATURES_EXT,
@@ -1242,7 +1248,9 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx,
p->desc_buf_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_BUFFER_FEATURES_EXT;
p->desc_buf_features.pNext = &p->atomic_float_features;
p->atomic_float_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_FLOAT_FEATURES_EXT;
- p->atomic_float_features.pNext = NULL;
+ p->atomic_float_features.pNext = &p->coop_matrix_features;
+ p->coop_matrix_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_FEATURES_KHR;
+ p->coop_matrix_features.pNext = NULL;
ctx->free = vulkan_device_free;
@@ -1304,6 +1312,8 @@ static int vulkan_device_create_internal(AVHWDeviceContext *ctx,
p->atomic_float_features.shaderBufferFloat32Atomics = atomic_float_features.shaderBufferFloat32Atomics;
p->atomic_float_features.shaderBufferFloat32AtomicAdd = atomic_float_features.shaderBufferFloat32AtomicAdd;
+ p->coop_matrix_features.cooperativeMatrix = coop_matrix_features.cooperativeMatrix;
+
dev_info.pNext = &hwctx->device_features;
/* Setup queue family */
@@ -90,9 +90,13 @@ int ff_vk_load_props(FFVulkanContext *s)
s->hprops = (VkPhysicalDeviceExternalMemoryHostPropertiesEXT) {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT,
};
+ s->coop_matrix_props = (VkPhysicalDeviceCooperativeMatrixPropertiesKHR) {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COOPERATIVE_MATRIX_PROPERTIES_KHR,
+ .pNext = &s->hprops,
+ };
s->subgroup_props = (VkPhysicalDeviceSubgroupSizeControlProperties) {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES,
- .pNext = &s->hprops,
+ .pNext = &s->coop_matrix_props,
};
s->desc_buf_props = (VkPhysicalDeviceDescriptorBufferPropertiesEXT) {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_BUFFER_PROPERTIES_EXT,
@@ -161,6 +165,25 @@ int ff_vk_load_props(FFVulkanContext *s)
vk->GetPhysicalDeviceQueueFamilyProperties2(s->hwctx->phys_dev, &s->tot_nb_qfs, s->qf_props);
+ if (vk->GetPhysicalDeviceCooperativeMatrixPropertiesKHR) {
+ vk->GetPhysicalDeviceCooperativeMatrixPropertiesKHR(s->hwctx->phys_dev,
+ &s->coop_mat_props_nb, NULL);
+
+ if (s->coop_mat_props_nb) {
+ s->coop_mat_props = av_malloc_array(s->coop_mat_props_nb,
+ sizeof(VkCooperativeMatrixPropertiesKHR));
+ for (int i = 0; i < s->coop_mat_props_nb; i++) {
+ s->coop_mat_props[i] = (VkCooperativeMatrixPropertiesKHR) {
+ .sType = VK_STRUCTURE_TYPE_COOPERATIVE_MATRIX_PROPERTIES_KHR,
+ };
+ }
+
+ vk->GetPhysicalDeviceCooperativeMatrixPropertiesKHR(s->hwctx->phys_dev,
+ &s->coop_mat_props_nb,
+ s->coop_mat_props);
+ }
+ }
+
return 0;
}
@@ -236,11 +236,15 @@ typedef struct FFVulkanContext {
VkPhysicalDeviceExternalMemoryHostPropertiesEXT hprops;
VkPhysicalDeviceDescriptorBufferPropertiesEXT desc_buf_props;
VkPhysicalDeviceSubgroupSizeControlProperties subgroup_props;
+ VkPhysicalDeviceCooperativeMatrixPropertiesKHR coop_matrix_props;
VkQueueFamilyQueryResultStatusPropertiesKHR *query_props;
VkQueueFamilyVideoPropertiesKHR *video_props;
VkQueueFamilyProperties2 *qf_props;
int tot_nb_qfs;
+ VkCooperativeMatrixPropertiesKHR *coop_mat_props;
+ uint32_t coop_mat_props_nb;
+
VkPhysicalDeviceShaderAtomicFloatFeaturesEXT atomic_float_feats;
VkPhysicalDeviceVulkan12Features feats_12;
VkPhysicalDeviceFeatures2 feats;
@@ -45,6 +45,7 @@ typedef enum FFVulkanExtensions {
FF_VK_EXT_VIDEO_DECODE_H265 = 1ULL << 13, /* VK_EXT_video_decode_h265 */
FF_VK_EXT_VIDEO_DECODE_AV1 = 1ULL << 14, /* VK_MESA_video_decode_av1 */
FF_VK_EXT_ATOMIC_FLOAT = 1ULL << 15, /* VK_EXT_shader_atomic_float */
+ FF_VK_EXT_COOP_MATRIX = 1ULL << 16, /* VK_KHR_cooperative_matrix */
FF_VK_EXT_NO_FLAG = 1ULL << 31,
} FFVulkanExtensions;
@@ -80,6 +81,7 @@ typedef enum FFVulkanExtensions {
MACRO(1, 0, FF_VK_EXT_NO_FLAG, GetPhysicalDeviceImageFormatProperties2) \
MACRO(1, 0, FF_VK_EXT_NO_FLAG, GetPhysicalDeviceQueueFamilyProperties) \
MACRO(1, 0, FF_VK_EXT_NO_FLAG, GetPhysicalDeviceQueueFamilyProperties2) \
+ MACRO(1, 0, FF_VK_EXT_COOP_MATRIX, GetPhysicalDeviceCooperativeMatrixPropertiesKHR) \
\
/* Command pool */ \
MACRO(1, 1, FF_VK_EXT_NO_FLAG, CreateCommandPool) \
@@ -46,6 +46,7 @@ static inline uint64_t ff_vk_extensions_to_mask(const char * const *extensions,
{ VK_EXT_DEBUG_UTILS_EXTENSION_NAME, FF_VK_EXT_DEBUG_UTILS },
{ VK_EXT_PHYSICAL_DEVICE_DRM_EXTENSION_NAME, FF_VK_EXT_DEVICE_DRM },
{ VK_EXT_SHADER_ATOMIC_FLOAT_EXTENSION_NAME, FF_VK_EXT_ATOMIC_FLOAT },
+ { VK_KHR_COOPERATIVE_MATRIX_EXTENSION_NAME, FF_VK_EXT_COOP_MATRIX },
#ifdef _WIN32
{ VK_KHR_EXTERNAL_MEMORY_WIN32_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_WIN32_MEMORY },
{ VK_KHR_EXTERNAL_SEMAPHORE_WIN32_EXTENSION_NAME, FF_VK_EXT_EXTERNAL_WIN32_SEM },
--
2.40.1