AlexIndustrial/mesa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvk: Take CRS into account when allocating the SLM area

Browse Source 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30402>
This commit is contained in:
Faith Ekstrand
2024-07-26 17:18:34 -05:00
committed by Marge Bot
parent b78f7d208b
commit d19ea8524b
3 changed files with 16 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/nouveau/vulkan/nvk_cmd_buffer.c
+4 -2
View File
@@ -585,8 +585,10 @@ nvk_cmd_bind_shaders(struct vk_command_buffer *vk_cmd,
struct nvk_shader *shader =
container_of(shaders[i], struct nvk_shader, vk);
if (shader != NULL && shader->info.slm_size > 0)
nvk_device_ensure_slm(dev, shader->info.slm_size);
if (shader != NULL) {
nvk_device_ensure_slm(dev, shader->info.slm_size,
shader->info.crs_size);
}
if (stages[i] == MESA_SHADER_COMPUTE ||
stages[i] == MESA_SHADER_KERNEL)
src/nouveau/vulkan/nvk_device.c
+10 -9
View File
@@ -52,17 +52,15 @@ nvk_slm_area_get_mem_ref(struct nvk_slm_area *area,
static VkResult
nvk_slm_area_ensure(struct nvk_device *dev,
struct nvk_slm_area *area,
uint32_t bytes_per_thread)
uint32_t slm_bytes_per_lane,
uint32_t crs_bytes_per_warp)
{
struct nvk_physical_device *pdev = nvk_device_physical(dev);
VkResult result;
assert(bytes_per_thread < (1 << 24));
/* TODO: Volta+doesn't use CRC */
const uint32_t crs_size = 0;
uint64_t bytes_per_warp = bytes_per_thread * 32 + crs_size;
assert(slm_bytes_per_lane < (1 << 24));
assert(crs_bytes_per_warp <= (1 << 20));
uint64_t bytes_per_warp = slm_bytes_per_lane * 32 + crs_bytes_per_warp;
/* The hardware seems to require this alignment for
* NV9097_SET_SHADER_LOCAL_MEMORY_E_DEFAULT_SIZE_PER_WARP
@@ -359,7 +357,10 @@ nvk_GetCalibratedTimestampsKHR(VkDevice _device,
VkResult
nvk_device_ensure_slm(struct nvk_device *dev,
uint32_t bytes_per_thread)
uint32_t slm_bytes_per_lane,
uint32_t crs_bytes_per_warp)
{
return nvk_slm_area_ensure(dev, &dev->slm, bytes_per_thread);
return nvk_slm_area_ensure(dev, &dev->slm,
slm_bytes_per_lane,
crs_bytes_per_warp);
}
src/nouveau/vulkan/nvk_device.h
+2 -1
View File
@@ -55,7 +55,8 @@ struct nvk_device {
VK_DEFINE_HANDLE_CASTS(nvk_device, vk.base, VkDevice, VK_OBJECT_TYPE_DEVICE)
VkResult nvk_device_ensure_slm(struct nvk_device *dev,
uint32_t bytes_per_thread);
uint32_t slm_bytes_per_lane,
uint32_t crs_bytes_per_warp);
static inline struct nvk_physical_device *
nvk_device_physical(struct nvk_device *dev)