radv: create decompress pipelines for separate depth/stencil layouts

No functional changes as the driver still uses the depth+stencil
pipeline.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

diff --git a/src/amd/vulkan/radv_constants.h b/src/amd/vulkan/radv_constants.h
index 7e33d0c584f9143cc7f602d938aec459c717d3e6..6abcf08520655941e90dff8d36c4cf22ab49e108 100644 (file)
--- a/src/amd/vulkan/radv_constants.h
+++ b/src/amd/vulkan/radv_constants.h
@@ -53,6 +53,7 @@
 #define MAX_INLINE_UNIFORM_BLOCK_COUNT 64
 
 #define NUM_DEPTH_CLEAR_PIPELINES 3
+#define NUM_DEPTH_DECOMPRESS_PIPELINES 3
 
 /*
  * This is the point we switch from using CP to compute shader

diff --git a/src/amd/vulkan/radv_meta_decompress.c b/src/amd/vulkan/radv_meta_decompress.c
index db1579be8aad8beb11589cd0be468b33c5730979..39340c7a778cddddc6769d97273c4f0f65d565f1 100644 (file)
--- a/src/amd/vulkan/radv_meta_decompress.c
+++ b/src/amd/vulkan/radv_meta_decompress.c
@@ -33,6 +33,12 @@ enum radv_depth_op {
        DEPTH_RESUMMARIZE,
 };
 
+enum radv_depth_decompress {
+       DECOMPRESS_DEPTH_STENCIL,
+       DECOMPRESS_DEPTH,
+       DECOMPRESS_STENCIL,
+};
+
 static VkResult
 create_pass(struct radv_device *device,
            uint32_t samples,
@@ -104,6 +110,7 @@ create_pipeline(struct radv_device *device,
                VkRenderPass pass,
                VkPipelineLayout layout,
                enum radv_depth_op op,
+               enum radv_depth_decompress decompress,
                VkPipeline *pipeline)
 {
        VkResult result;
@@ -214,8 +221,10 @@ create_pipeline(struct radv_device *device,
 
        struct radv_graphics_pipeline_create_info extra = {
                .use_rectlist = true,
-               .db_flush_depth_inplace = true,
-               .db_flush_stencil_inplace = true,
+               .db_flush_depth_inplace = decompress == DECOMPRESS_DEPTH_STENCIL ||
+                                         decompress == DECOMPRESS_DEPTH,
+               .db_flush_stencil_inplace = decompress == DECOMPRESS_DEPTH_STENCIL ||
+                                           decompress == DECOMPRESS_STENCIL,
                .db_resummarize = op == DEPTH_RESUMMARIZE,
        };
 
@@ -245,9 +254,12 @@ radv_device_finish_meta_depth_decomp_state(struct radv_device *device)
                radv_DestroyPipelineLayout(radv_device_to_handle(device),
                                           state->depth_decomp[i].p_layout,
                                           &state->alloc);
-               radv_DestroyPipeline(radv_device_to_handle(device),
-                                    state->depth_decomp[i].decompress_pipeline,
-                                    &state->alloc);
+
+               for (uint32_t j = 0; j < NUM_DEPTH_DECOMPRESS_PIPELINES; j++) {
+                       radv_DestroyPipeline(radv_device_to_handle(device),
+                                            state->depth_decomp[i].decompress_pipeline[j],
+                                            &state->alloc);
+               }
                radv_DestroyPipeline(radv_device_to_handle(device),
                                     state->depth_decomp[i].resummarize_pipeline,
                                     &state->alloc);
@@ -284,18 +296,22 @@ radv_device_init_meta_depth_decomp_state(struct radv_device *device, bool on_dem
                if (on_demand)
                        continue;
 
-               res = create_pipeline(device, vs_module_h, samples,
-                                     state->depth_decomp[i].pass,
-                                     state->depth_decomp[i].p_layout,
-                                     DEPTH_DECOMPRESS,
-                                     &state->depth_decomp[i].decompress_pipeline);
-               if (res != VK_SUCCESS)
-                       goto fail;
+               for (uint32_t j = 0; j < NUM_DEPTH_DECOMPRESS_PIPELINES; j++) {
+                       res = create_pipeline(device, vs_module_h, samples,
+                                             state->depth_decomp[i].pass,
+                                             state->depth_decomp[i].p_layout,
+                                             DEPTH_DECOMPRESS,
+                                             j,
+                                             &state->depth_decomp[i].decompress_pipeline[j]);
+                       if (res != VK_SUCCESS)
+                               goto fail;
+               }
 
                res = create_pipeline(device, vs_module_h, samples,
                                      state->depth_decomp[i].pass,
                                      state->depth_decomp[i].p_layout,
                                      DEPTH_RESUMMARIZE,
+                                     0, /* unused */
                                      &state->depth_decomp[i].resummarize_pipeline);
                if (res != VK_SUCCESS)
                        goto fail;
@@ -321,23 +337,27 @@ radv_get_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
        uint32_t samples_log2 = ffs(samples) - 1;
        VkPipeline *pipeline;
 
-       if (!state->depth_decomp[samples_log2].decompress_pipeline) {
+       if (!state->depth_decomp[samples_log2].decompress_pipeline[DECOMPRESS_DEPTH_STENCIL]) {
                VkResult ret;
 
-               ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
-                                     state->depth_decomp[samples_log2].pass,
-                                     state->depth_decomp[samples_log2].p_layout,
-                                     DEPTH_DECOMPRESS,
-                                     &state->depth_decomp[samples_log2].decompress_pipeline);
-               if (ret != VK_SUCCESS) {
-                       cmd_buffer->record_result = ret;
-                       return NULL;
+               for (uint32_t i = 0; i < NUM_DEPTH_DECOMPRESS_PIPELINES; i++) {
+                       ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
+                                             state->depth_decomp[samples_log2].pass,
+                                             state->depth_decomp[samples_log2].p_layout,
+                                             DEPTH_DECOMPRESS,
+                                             i,
+                                             &state->depth_decomp[samples_log2].decompress_pipeline[i]);
+                       if (ret != VK_SUCCESS) {
+                               cmd_buffer->record_result = ret;
+                               return NULL;
+                       }
                }
 
                ret = create_pipeline(cmd_buffer->device, VK_NULL_HANDLE, samples,
                                      state->depth_decomp[samples_log2].pass,
                                      state->depth_decomp[samples_log2].p_layout,
                                      DEPTH_RESUMMARIZE,
+                                     0, /* unused */
                                      &state->depth_decomp[samples_log2].resummarize_pipeline);
                if (ret != VK_SUCCESS) {
                        cmd_buffer->record_result = ret;
@@ -347,7 +367,7 @@ radv_get_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
 
        switch (op) {
        case DEPTH_DECOMPRESS:
-               pipeline = &state->depth_decomp[samples_log2].decompress_pipeline;
+               pipeline = &state->depth_decomp[samples_log2].decompress_pipeline[DECOMPRESS_DEPTH_STENCIL];
                break;
        case DEPTH_RESUMMARIZE:
                pipeline = &state->depth_decomp[samples_log2].resummarize_pipeline;

diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 5d63f4145f2528512bedceba054a74456824ee7c..1e15de10924ebf84a3edd5afb4c48d9c21bdadfb 100644 (file)
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -653,7 +653,7 @@ struct radv_meta_state {
 
        struct {
                VkPipelineLayout                          p_layout;
-               VkPipeline                                decompress_pipeline;
+               VkPipeline                                decompress_pipeline[NUM_DEPTH_DECOMPRESS_PIPELINES];
                VkPipeline                                resummarize_pipeline;
                VkRenderPass                              pass;
        } depth_decomp[MAX_SAMPLES_LOG2];