radeonsi: remove si_vid_join_surfaces and use combined planar allocations

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>

src/gallium/auxiliary/vl/vl_video_buffer.c

@@ -455,3 +455,43 @@ vl_video_buffer_create_ex2(struct pipe_context *pipe,
 
    return &buffer->base;
 }
+
+/* Create pipe_video_buffer by using resource_create with planar formats. */
+struct pipe_video_buffer *
+vl_video_buffer_create_as_resource(struct pipe_context *pipe,
+                                   const struct pipe_video_buffer *tmpl)
+{
+   struct pipe_resource templ, *resources[VL_NUM_COMPONENTS] = {};
+   unsigned array_size =  tmpl->interlaced ? 2 : 1;
+
+   memset(&templ, 0, sizeof(templ));
+   templ.target = array_size > 1 ? PIPE_TEXTURE_2D_ARRAY : PIPE_TEXTURE_2D;
+   templ.width0 = align(tmpl->width, VL_MACROBLOCK_WIDTH);
+   templ.height0 = align(tmpl->height / array_size, VL_MACROBLOCK_HEIGHT);
+   templ.depth0 = 1;
+   templ.array_size = array_size;
+   templ.bind = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET | tmpl->bind;
+   templ.usage = PIPE_USAGE_DEFAULT;
+
+   if (tmpl->buffer_format == PIPE_FORMAT_YUYV)
+      templ.format = PIPE_FORMAT_R8G8_R8B8_UNORM;
+   else if (tmpl->buffer_format == PIPE_FORMAT_UYVY)
+      templ.format = PIPE_FORMAT_G8R8_B8R8_UNORM;
+   else
+      templ.format = tmpl->buffer_format;
+
+   resources[0] = pipe->screen->resource_create(pipe->screen, &templ);
+   if (!resources[0])
+      return NULL;
+
+   if (resources[0]->next) {
+      pipe_resource_reference(&resources[1], resources[0]->next);
+      if (resources[1]->next)
+         pipe_resource_reference(&resources[2], resources[1]->next);
+   }
+
+   struct pipe_video_buffer vidtemplate = *tmpl;
+   vidtemplate.width = templ.width0;
+   vidtemplate.height = templ.height0 * array_size;
+   return vl_video_buffer_create_ex2(pipe, &vidtemplate, resources);
+}

src/gallium/auxiliary/vl/vl_video_buffer.h

@@ -145,4 +145,9 @@ vl_video_buffer_create_ex2(struct pipe_context *pipe,
                            const struct pipe_video_buffer *templat,
                            struct pipe_resource *resources[VL_NUM_COMPONENTS]);
 
+/* Create pipe_video_buffer by using resource_create with planar formats. */
+struct pipe_video_buffer *
+vl_video_buffer_create_as_resource(struct pipe_context *pipe,
+                                   const struct pipe_video_buffer *tmpl);
+
 #endif /* vl_video_buffer_h */

src/gallium/drivers/radeon/radeon_video.c

@@ -125,89 +125,3 @@ void si_vid_clear_buffer(struct pipe_context *context, struct rvid_buffer* buffe
 	si_sdma_clear_buffer(sctx, &buffer->res->b.b, 0, buffer->res->b.b.width0, 0);
 	context->flush(context, NULL, 0);
 }
-
-/**
- * join surfaces into the same buffer with identical tiling params
- * sumup their sizes and replace the backend buffers with a single bo
- */
-void si_vid_join_surfaces(struct si_context *sctx,
-			  struct pb_buffer** buffers[VL_NUM_COMPONENTS],
-			  struct radeon_surf *surfaces[VL_NUM_COMPONENTS])
-{
-	struct radeon_winsys *ws = sctx->ws;;
-	unsigned best_tiling, best_wh, off;
-	unsigned size, alignment;
-	struct pb_buffer *pb;
-	unsigned i, j;
-
-	for (i = 0, best_tiling = 0, best_wh = ~0; i < VL_NUM_COMPONENTS; ++i) {
-		unsigned wh;
-
-		if (!surfaces[i])
-			continue;
-
-		if (sctx->chip_class < GFX9) {
-			/* choose the smallest bank w/h for now */
-			wh = surfaces[i]->u.legacy.bankw * surfaces[i]->u.legacy.bankh;
-			if (wh < best_wh) {
-				best_wh = wh;
-				best_tiling = i;
-			}
-		}
-	}
-
-	for (i = 0, off = 0; i < VL_NUM_COMPONENTS; ++i) {
-		if (!surfaces[i])
-			continue;
-
-		/* adjust the texture layer offsets */
-		off = align(off, surfaces[i]->surf_alignment);
-
-		if (sctx->chip_class < GFX9) {
-			/* copy the tiling parameters */
-			surfaces[i]->u.legacy.bankw = surfaces[best_tiling]->u.legacy.bankw;
-			surfaces[i]->u.legacy.bankh = surfaces[best_tiling]->u.legacy.bankh;
-			surfaces[i]->u.legacy.mtilea = surfaces[best_tiling]->u.legacy.mtilea;
-			surfaces[i]->u.legacy.tile_split = surfaces[best_tiling]->u.legacy.tile_split;
-
-			for (j = 0; j < ARRAY_SIZE(surfaces[i]->u.legacy.level); ++j)
-				surfaces[i]->u.legacy.level[j].offset += off;
-		} else {
-			surfaces[i]->u.gfx9.surf_offset += off;
-			for (j = 0; j < ARRAY_SIZE(surfaces[i]->u.gfx9.offset); ++j)
-				surfaces[i]->u.gfx9.offset[j] += off;
-		}
-
-		surfaces[i]->flags |= RADEON_SURF_IMPORTED;
-		off += surfaces[i]->surf_size;
-	}
-
-	for (i = 0, size = 0, alignment = 0; i < VL_NUM_COMPONENTS; ++i) {
-		if (!buffers[i] || !*buffers[i])
-			continue;
-
-		size = align(size, (*buffers[i])->alignment);
-		size += (*buffers[i])->size;
-		alignment = MAX2(alignment, (*buffers[i])->alignment * 1);
-	}
-
-	if (!size)
-		return;
-
-	/* TODO: 2D tiling workaround */
-	alignment *= 2;
-
-	pb = ws->buffer_create(ws, size, alignment, RADEON_DOMAIN_VRAM,
-			       RADEON_FLAG_GTT_WC);
-	if (!pb)
-		return;
-
-	for (i = 0; i < VL_NUM_COMPONENTS; ++i) {
-		if (!buffers[i] || !*buffers[i])
-			continue;
-
-		pb_reference(buffers[i], pb);
-	}
-
-	pb_reference(&pb, NULL);
-}

src/gallium/drivers/radeon/radeon_video.h

@@ -60,10 +60,4 @@ bool si_vid_resize_buffer(struct pipe_screen *screen, struct radeon_cmdbuf *cs,
 /* clear the buffer with zeros */
 void si_vid_clear_buffer(struct pipe_context *context, struct rvid_buffer* buffer);
 
-/* join surfaces into the same buffer with identical tiling params
-   sumup their sizes and replace the backend buffers with a single bo */
-void si_vid_join_surfaces(struct si_context *sctx,
-			  struct pb_buffer** buffers[VL_NUM_COMPONENTS],
-			  struct radeon_surf *surfaces[VL_NUM_COMPONENTS]);
-
 #endif // RADEON_VIDEO_H

src/gallium/drivers/radeonsi/meson.build

@@ -129,7 +129,7 @@ driver_radeonsi = declare_dependency(
   compile_args : '-DGALLIUM_RADEONSI',
   sources : si_driinfo_h,
   link_with : [
-    libradeonsi, libradeonwinsys, libamdgpuwinsys, libamd_common, libamd_common_llvm
+    libradeonsi, libradeonwinsys, libamdgpuwinsys, libamd_common, libamd_common_llvm, libgalliumvl
   ],
   dependencies : idep_nir,
 )

src/gallium/drivers/radeonsi/si_uvd.c

@@ -40,71 +40,11 @@
 struct pipe_video_buffer *si_video_buffer_create(struct pipe_context *pipe,
 						 const struct pipe_video_buffer *tmpl)
 {
-	struct si_context *ctx = (struct si_context *)pipe;
+	struct pipe_video_buffer vidbuf = *tmpl;
-	struct si_texture *resources[VL_NUM_COMPONENTS] = {};
-	struct radeon_surf *surfaces[VL_NUM_COMPONENTS] = {};
-	struct pb_buffer **pbs[VL_NUM_COMPONENTS] = {};
-	enum pipe_format resource_formats[VL_NUM_COMPONENTS];
-	struct pipe_video_buffer vidtemplate;
-	struct pipe_resource templ;
-	unsigned i, array_size;
-
-	assert(pipe);
-
-	/* first create the needed resources as "normal" textures */
-	vl_get_video_buffer_formats(pipe->screen, tmpl->buffer_format, resource_formats);
-
-	array_size = tmpl->interlaced ? 2 : 1;
-	vidtemplate = *tmpl;
-	vidtemplate.width = align(tmpl->width, VL_MACROBLOCK_WIDTH);
-	vidtemplate.height = align(tmpl->height / array_size, VL_MACROBLOCK_HEIGHT);
-
-	assert(resource_formats[0] != PIPE_FORMAT_NONE);
-
-	for (i = 0; i < VL_NUM_COMPONENTS; ++i) {
-		if (resource_formats[i] != PIPE_FORMAT_NONE) {
-			vl_video_buffer_template(&templ, &vidtemplate,
-			                         resource_formats[i], 1,
-			                         array_size, PIPE_USAGE_DEFAULT, i);
-			/* Set PIPE_BIND_SHARED to avoid reallocation in si_texture_get_handle,
-			 * which can't handle joined surfaces. */
 	/* TODO: get tiling working */
-			templ.bind = PIPE_BIND_LINEAR | PIPE_BIND_SHARED;
+	vidbuf.bind |= PIPE_BIND_LINEAR;
-			resources[i] = (struct si_texture *)
-			                pipe->screen->resource_create(pipe->screen, &templ);
-			if (!resources[i])
-				goto error;
-		}
-	}
 
-	for (i = 0; i < VL_NUM_COMPONENTS; ++i) {
+	return vl_video_buffer_create_as_resource(pipe, &vidbuf);
-		if (!resources[i])
-			continue;
-
-		surfaces[i] = & resources[i]->surface;
-		pbs[i] = &resources[i]->buffer.buf;
-	}
-
-	si_vid_join_surfaces(ctx, pbs, surfaces);
-
-	for (i = 0; i < VL_NUM_COMPONENTS; ++i) {
-		if (!resources[i])
-			continue;
-
-		/* reset the address */
-		resources[i]->buffer.gpu_address = ctx->ws->buffer_get_virtual_address(
-			resources[i]->buffer.buf);
-		resources[i]->buffer.bo_size = resources[i]->buffer.buf->size;
-	}
-
-	vidtemplate.height *= array_size;
-	return vl_video_buffer_create_ex2(pipe, &vidtemplate, (struct pipe_resource **)resources);
-
-error:
-	for (i = 0; i < VL_NUM_COMPONENTS; ++i)
-		si_texture_reference(&resources[i], NULL);
-
-	return NULL;
 }
 
 /* set the decoding target buffer offsets */