[Mesa-dev] [PATCH] radeonsi: use unsynchronized transfers for shader binary uploads

[Samuel Pitoiset]

Because the buffer is new, it can't be referenced by any CS.

This can save few CPU cycles by skipping the whole
PIPE_TRANSFER_UNSYNCHRONIZED if in amdgpu_bo_map().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
---
 src/gallium/drivers/radeonsi/si_shader.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index bfccf813a6..2a5f21447c 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -6205,7 +6205,8 @@ int si_shader_binary_upload(struct si_screen *sscreen, struct si_shader *shader)
 
 	/* Upload. */
 	ptr = sscreen->b.ws->buffer_map(shader->bo->buf, NULL,
-					PIPE_TRANSFER_READ_WRITE);
+					PIPE_TRANSFER_READ_WRITE |
+					PIPE_TRANSFER_UNSYNCHRONIZED);
 
 	if (prolog) {
 		util_memcpy_cpu_to_le32(ptr, prolog->code, prolog->code_size);
-- 
2.12.2