panfrost: Use preuploaded shader descriptors

For non-fragment shaders, we can just use the preuploaded BO as-is and
do no packing/copying at draw-time. Fragment shaders are still a bit of
an edge case, in having rasterizer/zsa/blend state in the same
descriptor, but now we can specialize that path further for
fragment-only.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6440>

diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c
index 7dd43f4e3e916635251af5588cb0db35bac3c5a9..801a39216f224462e8117ee1971deba345893289 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_cmdstream.c
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.c
@@ -306,22 +306,6 @@ panfrost_vt_set_draw_info(struct panfrost_context *ctx,
         }
 }
 
-static void
-panfrost_emit_compute_shader(struct panfrost_context *ctx,
-                          enum pipe_shader_type st,
-                          struct mali_shader_meta *meta)
-{
-        const struct panfrost_device *dev = pan_device(ctx->base.screen);
-        struct panfrost_shader_state *ss = panfrost_get_shader_state(ctx, st);
-
-        memset(meta, 0, sizeof(*meta));
-        memcpy(&meta->shader, &ss->shader, sizeof(ss->shader));
-        memcpy(&meta->midgard_props, &ss->properties, sizeof(ss->properties));
-
-        if (dev->quirks & IS_BIFROST)
-                memcpy(&meta->bifrost_preload, &ss->preload, sizeof(ss->preload));
-}
-
 static unsigned
 translate_tex_wrap(enum pipe_tex_wrap w)
 {
@@ -715,78 +699,77 @@ panfrost_emit_frag_shader(struct panfrost_context *ctx,
         }
 }
 
-void
-panfrost_emit_shader_meta(struct panfrost_batch *batch,
-                          enum pipe_shader_type st,
-                          struct mali_vertex_tiler_postfix *postfix)
+mali_ptr
+panfrost_emit_compute_shader_meta(struct panfrost_batch *batch, enum pipe_shader_type stage)
 {
-        struct panfrost_context *ctx = batch->ctx;
-        struct panfrost_shader_state *ss = panfrost_get_shader_state(ctx, st);
+        struct panfrost_shader_state *ss = panfrost_get_shader_state(batch->ctx, stage);
 
-        if (!ss) {
-                postfix->shader = 0;
-                return;
-        }
-
-        struct mali_shader_meta meta;
-
-        /* Add the shader BO to the batch. */
         panfrost_batch_add_bo(batch, ss->bo,
                               PAN_BO_ACCESS_PRIVATE |
                               PAN_BO_ACCESS_READ |
-                              panfrost_bo_access_for_stage(st));
+                              PAN_BO_ACCESS_VERTEX_TILER);
 
-        mali_ptr shader_ptr;
+        panfrost_batch_add_bo(batch, pan_resource(ss->upload.rsrc)->bo,
+                              PAN_BO_ACCESS_PRIVATE |
+                              PAN_BO_ACCESS_READ |
+                              PAN_BO_ACCESS_VERTEX_TILER);
 
-        if (st == PIPE_SHADER_FRAGMENT) {
-                struct panfrost_device *dev = pan_device(ctx->base.screen);
-                unsigned rt_count = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
-                size_t desc_size = sizeof(meta);
-                void *rts = NULL;
-                struct panfrost_transfer xfer;
-                unsigned rt_size;
+        return pan_resource(ss->upload.rsrc)->bo->gpu + ss->upload.offset;
+}
 
-                if (dev->quirks & MIDGARD_SFBD)
-                        rt_size = 0;
-                else if (dev->quirks & IS_BIFROST)
-                        rt_size = sizeof(struct bifrost_blend_rt);
-                else
-                        rt_size = sizeof(struct midgard_blend_rt);
+mali_ptr
+panfrost_emit_frag_shader_meta(struct panfrost_batch *batch)
+{
+        struct panfrost_context *ctx = batch->ctx;
+        struct panfrost_shader_state *ss = panfrost_get_shader_state(ctx, PIPE_SHADER_FRAGMENT);
+        struct mali_shader_meta meta;
 
-                desc_size += rt_size * rt_count;
+        /* Add the shader BO to the batch. */
+        panfrost_batch_add_bo(batch, ss->bo,
+                              PAN_BO_ACCESS_PRIVATE |
+                              PAN_BO_ACCESS_READ |
+                              PAN_BO_ACCESS_FRAGMENT);
 
-                if (rt_size)
-                        rts = rzalloc_size(ctx, rt_size * rt_count);
+        struct panfrost_device *dev = pan_device(ctx->base.screen);
+        unsigned rt_count = MAX2(ctx->pipe_framebuffer.nr_cbufs, 1);
+        size_t desc_size = sizeof(meta);
+        void *rts = NULL;
+        struct panfrost_transfer xfer;
+        unsigned rt_size;
+
+        if (dev->quirks & MIDGARD_SFBD)
+                rt_size = 0;
+        else if (dev->quirks & IS_BIFROST)
+                rt_size = sizeof(struct bifrost_blend_rt);
+        else
+                rt_size = sizeof(struct midgard_blend_rt);
 
-                struct panfrost_blend_final blend[PIPE_MAX_COLOR_BUFS];
+        desc_size += rt_size * rt_count;
 
-                for (unsigned c = 0; c < ctx->pipe_framebuffer.nr_cbufs; ++c)
-                        blend[c] = panfrost_get_blend_for_context(ctx, c);
+        if (rt_size)
+                rts = rzalloc_size(ctx, rt_size * rt_count);
 
-                panfrost_emit_frag_shader(ctx, &meta, blend);
+        struct panfrost_blend_final blend[PIPE_MAX_COLOR_BUFS];
 
-                if (!(dev->quirks & MIDGARD_SFBD))
-                        panfrost_emit_blend(batch, rts, blend);
-                else
-                        batch->draws |= PIPE_CLEAR_COLOR0;
+        for (unsigned c = 0; c < ctx->pipe_framebuffer.nr_cbufs; ++c)
+                blend[c] = panfrost_get_blend_for_context(ctx, c);
 
-                xfer = panfrost_pool_alloc_aligned(&batch->pool, desc_size, sizeof(meta));
+        panfrost_emit_frag_shader(ctx, &meta, blend);
 
-                memcpy(xfer.cpu, &meta, sizeof(meta));
-                memcpy(xfer.cpu + sizeof(meta), rts, rt_size * rt_count);
+        if (!(dev->quirks & MIDGARD_SFBD))
+                panfrost_emit_blend(batch, rts, blend);
+        else
+                batch->draws |= PIPE_CLEAR_COLOR0;
 
-                if (rt_size)
-                        ralloc_free(rts);
+        xfer = panfrost_pool_alloc_aligned(&batch->pool, desc_size, sizeof(meta));
 
-                shader_ptr = xfer.gpu;
-        } else {
-                panfrost_emit_compute_shader(ctx, st, &meta);
+        memcpy(xfer.cpu, &meta, sizeof(meta));
+        memcpy(xfer.cpu + sizeof(meta), rts, rt_size * rt_count);
 
-                shader_ptr = panfrost_pool_upload(&batch->pool, &meta,
-                                                       sizeof(meta));
-        }
+        if (rt_size)
+                ralloc_free(rts);
 
-        postfix->shader = shader_ptr;
+        return xfer.gpu;
 }
 
 void

diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.h b/src/gallium/drivers/panfrost/pan_cmdstream.h
index c6916033cc161f62d780fa4904597c69770b920f..f03d1a617616c0561d78be5bc6753efc4b23dd90 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_cmdstream.h
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.h
@@ -51,10 +51,11 @@ panfrost_vt_set_draw_info(struct panfrost_context *ctx,
                           unsigned *vertex_count,
                           unsigned *padded_count);
 
-void
-panfrost_emit_shader_meta(struct panfrost_batch *batch,
-                          enum pipe_shader_type st,
-                          struct mali_vertex_tiler_postfix *postfix);
+mali_ptr
+panfrost_emit_compute_shader_meta(struct panfrost_batch *batch, enum pipe_shader_type stage);
+
+mali_ptr
+panfrost_emit_frag_shader_meta(struct panfrost_batch *batch);
 
 void
 panfrost_emit_viewport(struct panfrost_batch *batch,

diff --git a/src/gallium/drivers/panfrost/pan_compute.c b/src/gallium/drivers/panfrost/pan_compute.c
index 64d868448448d41461881361dbb850413956ad90..8ad1485a566c30b8d3757939e03e4409dcfcdcde 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_compute.c
+++ b/src/gallium/drivers/panfrost/pan_compute.c
@@ -119,7 +119,7 @@ panfrost_launch_grid(struct pipe_context *pipe,
 
         panfrost_vt_init(ctx, PIPE_SHADER_COMPUTE, &payload.prefix, &payload.postfix);
 
-        panfrost_emit_shader_meta(batch, PIPE_SHADER_COMPUTE, &payload.postfix);
+        payload.postfix.shader = panfrost_emit_compute_shader_meta(batch, PIPE_SHADER_COMPUTE);
         panfrost_emit_const_buf(batch, PIPE_SHADER_COMPUTE, &payload.postfix);
         panfrost_emit_shared_memory(batch, info, &payload);
 

diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index bcf0d199bd45f1ba6f768977d3a3463d467225a9..7b0edd6853d8cfc3a62ab50d66868dfa638829a4 100644 (file)
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -325,8 +325,6 @@ panfrost_draw_vbo(
                                          ctx->instance_count,
                                          &vertex_postfix, &tiler_postfix,
                                          &primitive_size);
-        panfrost_emit_shader_meta(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
-        panfrost_emit_shader_meta(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
         panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
         panfrost_emit_sampler_descriptors(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
         panfrost_emit_texture_descriptors(batch, PIPE_SHADER_VERTEX, &vertex_postfix);
@@ -335,6 +333,9 @@ panfrost_draw_vbo(
         panfrost_emit_const_buf(batch, PIPE_SHADER_FRAGMENT, &tiler_postfix);
         panfrost_emit_viewport(batch, &tiler_postfix);
 
+        vertex_postfix.shader = panfrost_emit_compute_shader_meta(batch, PIPE_SHADER_VERTEX);
+        tiler_postfix.shader = panfrost_emit_frag_shader_meta(batch);
+
         panfrost_vt_update_primitive_size(ctx, &tiler_prefix, &primitive_size);
 
         /* Fire off the draw itself */