panvk: Upload render state in panvk_shader

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29161>

src/panfrost/vulkan/bifrost/panvk_pipeline.h

@@ -28,7 +28,6 @@
 
 struct panvk_pipeline_shader {
    struct panvk_shader *base;
-   struct panvk_priv_mem rsd;
 
    struct {
       struct panvk_priv_mem attribs;

src/panfrost/vulkan/bifrost/panvk_vX_pipeline.c

@@ -75,15 +75,6 @@ init_pipeline_shader(struct panvk_pipeline *pipeline,
       compute_pipeline->local_size = shader->local_size;
    }
 
-   if (stage_info->stage != VK_SHADER_STAGE_FRAGMENT_BIT) {
-      pshader->rsd = panvk_pool_alloc_desc(&dev->mempools.rw, RENDERER_STATE);
-
-      pan_pack(panvk_priv_mem_host_addr(pshader->rsd), RENDERER_STATE, cfg) {
-         pan_shader_prepare_rsd(&pshader->info,
-                                panvk_shader_get_dev_addr(pshader->base), &cfg);
-      }
-   }
-
    return VK_SUCCESS;
 }
 
@@ -94,8 +85,6 @@ cleanup_pipeline_shader(struct panvk_pipeline *pipeline,
 {
    struct panvk_device *dev = to_panvk_device(pipeline->base.device);
 
-   panvk_pool_free_mem(&dev->mempools.rw, pshader->rsd);
-
    if (pshader->base != NULL)
       panvk_per_arch(shader_destroy)(dev, pshader->base, alloc);
 }

src/panfrost/vulkan/jm/panvk_vX_cmd_dispatch.c

@@ -36,6 +36,13 @@ panvk_per_arch(CmdDispatch)(VkCommandBuffer commandBuffer, uint32_t x,
                             uint32_t y, uint32_t z)
 {
    VK_FROM_HANDLE(panvk_cmd_buffer, cmdbuf, commandBuffer);
+   const struct panvk_compute_pipeline *pipeline =
+      cmdbuf->state.compute.pipeline;
+
+   /* If there's no compute shader, we can skip the dispatch. */
+   if (!panvk_priv_mem_dev_addr(pipeline->cs.base->rsd))
+      return;
+
    struct panvk_device *dev = to_panvk_device(cmdbuf->vk.base.device);
    struct panvk_physical_device *phys_dev =
       to_panvk_physical_device(dev->vk.physical);
@@ -50,8 +57,6 @@ panvk_per_arch(CmdDispatch)(VkCommandBuffer commandBuffer, uint32_t x,
       &cmdbuf->state.compute.desc_state;
    struct panvk_shader_desc_state *cs_desc_state =
       &cmdbuf->state.compute.cs.desc;
-   const struct panvk_compute_pipeline *pipeline =
-      cmdbuf->state.compute.pipeline;
 
    panvk_per_arch(cmd_alloc_tls_desc)(cmdbuf, false);
    dispatch.tsd = batch->tls.gpu;
@@ -111,7 +116,7 @@ panvk_per_arch(CmdDispatch)(VkCommandBuffer commandBuffer, uint32_t x,
    }
 
    pan_section_pack(job.cpu, COMPUTE_JOB, DRAW, cfg) {
-      cfg.state = panvk_priv_mem_dev_addr(pipeline->cs.rsd);
+      cfg.state = panvk_priv_mem_dev_addr(pipeline->cs.base->rsd);
       cfg.attributes = cs_desc_state->img_attrib_table;
       cfg.attribute_buffers =
          cs_desc_state->tables[PANVK_BIFROST_DESC_TABLE_IMG];

src/panfrost/vulkan/jm/panvk_vX_cmd_draw.c

@@ -806,7 +806,7 @@ panvk_draw_prepare_vertex_job(struct panvk_cmd_buffer *cmdbuf,
    }
 
    pan_section_pack(ptr.cpu, COMPUTE_JOB, DRAW, cfg) {
-      cfg.state = panvk_priv_mem_dev_addr(pipeline->vs.rsd);
+      cfg.state = panvk_priv_mem_dev_addr(pipeline->vs.base->rsd);
       cfg.attributes = draw->vs.attributes;
       cfg.attribute_buffers = draw->vs.attribute_bufs;
       cfg.varyings = draw->vs.varyings;
@@ -1043,6 +1043,10 @@ panvk_cmd_draw(struct panvk_cmd_buffer *cmdbuf, struct panvk_draw_info *draw)
    const struct vk_rasterization_state *rs =
       &cmdbuf->vk.dynamic_graphics_state.rs;
 
+   /* If there's no vertex shader, we can skip the draw. */
+   if (!panvk_priv_mem_dev_addr(pipeline->vs.base->rsd))
+      return;
+
    /* There are only 16 bits in the descriptor for the job ID. Each job has a
     * pilot shader dealing with descriptor copies, and we need one
     * <vertex,tiler> pair per draw.

src/panfrost/vulkan/panvk_shader.h

@@ -123,6 +123,7 @@ struct panvk_shader {
    uint32_t bin_size;
 
    struct panvk_priv_mem code_mem;
+   struct panvk_priv_mem rsd;
 };
 
 static inline mali_ptr

src/panfrost/vulkan/panvk_vX_shader.c

@@ -140,11 +140,23 @@ static VkResult
 panvk_shader_upload(struct panvk_device *dev, struct panvk_shader *shader,
                     const VkAllocationCallbacks *pAllocator)
 {
-   if (shader->bin_size > 0) {
-      shader->code_mem = panvk_pool_upload_aligned(
-         &dev->mempools.exec, shader->bin_ptr, shader->bin_size, 128);
-   } else {
-      shader->code_mem = (struct panvk_priv_mem){0};
+   shader->code_mem = (struct panvk_priv_mem){0};
+   shader->rsd = (struct panvk_priv_mem){0};
+
+   if (!shader->bin_size)
+      return VK_SUCCESS;
+
+   shader->code_mem = panvk_pool_upload_aligned(
+      &dev->mempools.exec, shader->bin_ptr, shader->bin_size, 128);
+
+   if (shader->info.stage == MESA_SHADER_FRAGMENT)
+      return VK_SUCCESS;
+
+   shader->rsd = panvk_pool_alloc_desc(&dev->mempools.rw, RENDERER_STATE);
+
+   pan_pack(panvk_priv_mem_host_addr(shader->rsd), RENDERER_STATE, cfg) {
+      pan_shader_prepare_rsd(&shader->info, panvk_shader_get_dev_addr(shader),
+                             &cfg);
    }
 
    return VK_SUCCESS;
@@ -394,6 +406,7 @@ panvk_per_arch(shader_destroy)(struct panvk_device *dev,
                                struct panvk_shader *shader,
                                const VkAllocationCallbacks *alloc)
 {
+   panvk_pool_free_mem(&dev->mempools.rw, shader->rsd);
    panvk_pool_free_mem(&dev->mempools.rw, shader->desc_info.others.map);
    panvk_pool_free_mem(&dev->mempools.exec, shader->code_mem);