4 files changed, 32 insertions, 25 deletions

diff --git a/src/gallium/drivers/radeonsi/si_blit.c b/src/gallium/drivers/radeonsi/si_blit.c
index 4744154c7e2..1f2c4082dbc 100644
--- a/src/gallium/drivers/radeonsi/si_blit.c
+++ b/src/gallium/drivers/radeonsi/si_blit.c
@@ -556,7 +556,7 @@ void si_resource_copy_region(struct pipe_context *ctx,
 
 	/* Fallback for buffers. */
 	if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
-		si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width);
+		si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
 		return;
 	}
 
diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index d46f4e522e6..c9599617ede 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -1054,9 +1054,11 @@ static void si_invalidate_buffer(struct pipe_context *ctx, struct pipe_resource
 #define CP_DMA_MAX_BYTE_COUNT ((1 << 21) - 8)
 
 static void si_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
-			    unsigned offset, unsigned size, unsigned value)
+			    unsigned offset, unsigned size, unsigned value,
+			    bool is_framebuffer)
 {
 	struct si_context *sctx = (struct si_context*)ctx;
+	unsigned flush_flags;
 
 	if (!size)
 		return;
@@ -1081,12 +1083,15 @@ static void si_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 	uint64_t va = r600_resource(dst)->gpu_address + offset;
 
 	/* Flush the caches where the resource is bound. */
-	/* XXX only flush the caches where the buffer is bound. */
-	sctx->b.flags |= SI_CONTEXT_INV_TC_L1 |
-			 SI_CONTEXT_INV_TC_L2 |
-			 SI_CONTEXT_INV_KCACHE |
-			 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
-	sctx->b.flags |= SI_CONTEXT_PS_PARTIAL_FLUSH;
+	if (is_framebuffer)
+		flush_flags = SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+	else
+		flush_flags = SI_CONTEXT_INV_TC_L1 |
+			      SI_CONTEXT_INV_TC_L2 |
+			      SI_CONTEXT_INV_KCACHE;
+
+	sctx->b.flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
+			 flush_flags;
 
 	while (size) {
 		unsigned byte_count = MIN2(size, CP_DMA_MAX_BYTE_COUNT);
@@ -1120,17 +1125,16 @@ static void si_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
 
 	/* Flush the caches again in case the 3D engine has been prefetching
 	 * the resource. */
-	/* XXX only flush the caches where the buffer is bound. */
-	sctx->b.flags |= SI_CONTEXT_INV_TC_L1 |
-			 SI_CONTEXT_INV_TC_L2 |
-			 SI_CONTEXT_INV_KCACHE |
-			 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+	sctx->b.flags |= flush_flags;
 }
 
 void si_copy_buffer(struct si_context *sctx,
 		    struct pipe_resource *dst, struct pipe_resource *src,
-		    uint64_t dst_offset, uint64_t src_offset, unsigned size)
+		    uint64_t dst_offset, uint64_t src_offset, unsigned size,
+		    bool is_framebuffer)
 {
+	unsigned flush_flags;
+
 	if (!size)
 		return;
 
@@ -1144,11 +1148,15 @@ void si_copy_buffer(struct si_context *sctx,
 	src_offset += r600_resource(src)->gpu_address;
 
 	/* Flush the caches where the resource is bound. */
-	sctx->b.flags |= SI_CONTEXT_INV_TC_L1 |
-			 SI_CONTEXT_INV_TC_L2 |
-			 SI_CONTEXT_INV_KCACHE |
-			 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
-			 SI_CONTEXT_PS_PARTIAL_FLUSH;
+	if (is_framebuffer)
+		flush_flags = SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+	else
+		flush_flags = SI_CONTEXT_INV_TC_L1 |
+			      SI_CONTEXT_INV_TC_L2 |
+			      SI_CONTEXT_INV_KCACHE;
+
+	sctx->b.flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
+			 flush_flags;
 
 	while (size) {
 		unsigned sync_flags = 0;
@@ -1180,10 +1188,9 @@ void si_copy_buffer(struct si_context *sctx,
 		dst_offset += byte_count;
 	}
 
-	sctx->b.flags |= SI_CONTEXT_INV_TC_L1 |
-			 SI_CONTEXT_INV_TC_L2 |
-			 SI_CONTEXT_INV_KCACHE |
-			 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER;
+	/* Flush the caches again in case the 3D engine has been prefetching
+	 * the resource. */
+	sctx->b.flags |= flush_flags;
 }
 
 /* INIT/DEINIT */
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 4d8fd65d981..38bff31e005 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -155,7 +155,7 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen, void *
 
 		/* Clear the NULL constant buffer, because loads should return zeros. */
 		sctx->b.clear_buffer(&sctx->b.b, sctx->null_const_buf.buffer, 0,
-				     sctx->null_const_buf.buffer->width0, 0);
+				     sctx->null_const_buf.buffer->width0, 0, false);
 	}
 
 	return &sctx->b.b;
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 8927e50644b..3cd252c0e64 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -238,7 +238,7 @@ void si_release_all_descriptors(struct si_context *sctx);
 void si_all_descriptors_begin_new_cs(struct si_context *sctx);
 void si_copy_buffer(struct si_context *sctx,
 		    struct pipe_resource *dst, struct pipe_resource *src,
-		    uint64_t dst_offset, uint64_t src_offset, unsigned size);
+		    uint64_t dst_offset, uint64_t src_offset, unsigned size, bool is_framebuffer);
 void si_upload_const_buffer(struct si_context *sctx, struct r600_resource **rbuffer,
 			    const uint8_t *ptr, unsigned size, uint32_t *const_offset);