winsys/amdgpu: explicitly declare whether buffer_map is permanent or not

Introduce a new driver-private transfer flag RADEON_TRANSFER_TEMPORARY
that specifies whether the caller will use buffer_unmap or not. The
default behavior is set to permanent maps, because that's what drivers
do for Gallium buffer maps.

This should eliminate the need for hacks in libdrm. Assertions are added
to catch when the buffer_unmap calls don't match the (temporary)
buffer_map calls.

I did my best to update r600 for consistency (r300 needs no changes
because it never calls buffer_unmap), even though the radeon winsys
ignores the new flag.

As an added bonus, this should actually improve the performance of
the normal fast path, because we no longer call into libdrm at all
after the first map, and there's one less atomic in the winsys itself
(there are now no atomics left in the UNSYNCHRONIZED fast path).

Cc: Leo Liu <[email protected]>
v2:
- remove comment about visible VRAM (Marek)
- don't rely on amdgpu_bo_cpu_map doing an atomic write
Reviewed-by: Marek Olšák <[email protected]>

6 files changed, 21 insertions, 9 deletions

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index a77f58242e3..9085be4e2f3 100644
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -438,7 +438,9 @@ static void *evergreen_create_compute_state(struct pipe_context *ctx,
 	/* Upload code + ROdata */
 	shader->code_bo = r600_compute_buffer_alloc_vram(rctx->screen,
 							shader->bc.ndw * 4);
-	p = r600_buffer_map_sync_with_rings(&rctx->b, shader->code_bo, PIPE_TRANSFER_WRITE);
+	p = r600_buffer_map_sync_with_rings(
+		&rctx->b, shader->code_bo,
+		PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
 	//TODO: use util_memcpy_cpu_to_le32 ?
 	memcpy(p, shader->bc.bytecode, shader->bc.ndw * 4);
 	rctx->b.ws->buffer_unmap(shader->code_bo->buf);
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 7029be24f4b..4ba77c535f9 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -2772,7 +2772,9 @@ void *r600_create_vertex_fetch_shader(struct pipe_context *ctx,
 		return NULL;
 	}
 
-	bytecode = r600_buffer_map_sync_with_rings(&rctx->b, shader->buffer, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED);
+	bytecode = r600_buffer_map_sync_with_rings
+		(&rctx->b, shader->buffer,
+		PIPE_TRANSFER_WRITE | PIPE_TRANSFER_UNSYNCHRONIZED | RADEON_TRANSFER_TEMPORARY);
 	bytecode += shader->offset / 4;
 
 	if (R600_BIG_ENDIAN) {
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 408939d1105..fc826470d69 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -141,7 +141,9 @@ static int store_shader(struct pipe_context *ctx,
 		if (shader->bo == NULL) {
 			return -ENOMEM;
 		}
-		ptr = r600_buffer_map_sync_with_rings(&rctx->b, shader->bo, PIPE_TRANSFER_WRITE);
+		ptr = r600_buffer_map_sync_with_rings(
+			&rctx->b, shader->bo,
+			PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
 		if (R600_BIG_ENDIAN) {
 			for (i = 0; i < shader->shader.bc.ndw; ++i) {
 				ptr[i] = util_cpu_to_le32(shader->shader.bc.bytecode[i]);
diff --git a/src/gallium/drivers/r600/radeon_uvd.c b/src/gallium/drivers/r600/radeon_uvd.c
index 495a93dc55a..5568f2138e4 100644
--- a/src/gallium/drivers/r600/radeon_uvd.c
+++ b/src/gallium/drivers/r600/radeon_uvd.c
@@ -152,7 +152,8 @@ static void map_msg_fb_it_buf(struct ruvd_decoder *dec)
 	buf = &dec->msg_fb_it_buffers[dec->cur_buffer];
 
 	/* and map it for CPU access */
-	ptr = dec->ws->buffer_map(buf->res->buf, dec->cs, PIPE_TRANSFER_WRITE);
+	ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
+                                  PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
 
 	/* calc buffer offsets */
 	dec->msg = (struct ruvd_msg *)ptr;
@@ -1068,7 +1069,7 @@ static void ruvd_begin_frame(struct pipe_video_codec *decoder,
 	dec->bs_size = 0;
 	dec->bs_ptr = dec->ws->buffer_map(
 		dec->bs_buffers[dec->cur_buffer].res->buf,
-		dec->cs, PIPE_TRANSFER_WRITE);
+		dec->cs, PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
 }
 
 /**
@@ -1121,7 +1122,8 @@ static void ruvd_decode_bitstream(struct pipe_video_codec *decoder,
 			}
 
 			dec->bs_ptr = dec->ws->buffer_map(buf->res->buf, dec->cs,
-							  PIPE_TRANSFER_WRITE);
+							  PIPE_TRANSFER_WRITE |
+							  RADEON_TRANSFER_TEMPORARY);
 			if (!dec->bs_ptr)
 				return;
 
diff --git a/src/gallium/drivers/r600/radeon_vce.c b/src/gallium/drivers/r600/radeon_vce.c
index 60ba12a593a..e38b927b1d4 100644
--- a/src/gallium/drivers/r600/radeon_vce.c
+++ b/src/gallium/drivers/r600/radeon_vce.c
@@ -353,7 +353,9 @@ static void rvce_get_feedback(struct pipe_video_codec *encoder,
 	struct rvid_buffer *fb = feedback;
 
 	if (size) {
-		uint32_t *ptr = enc->ws->buffer_map(fb->res->buf, enc->cs, PIPE_TRANSFER_READ_WRITE);
+		uint32_t *ptr = enc->ws->buffer_map(
+			fb->res->buf, enc->cs,
+			PIPE_TRANSFER_READ_WRITE | RADEON_TRANSFER_TEMPORARY);
 
 		if (ptr[1]) {
 			*size = ptr[4] - ptr[9];
diff --git a/src/gallium/drivers/r600/radeon_video.c b/src/gallium/drivers/r600/radeon_video.c
index 02fcf77d4ff..8e0af448be5 100644
--- a/src/gallium/drivers/r600/radeon_video.c
+++ b/src/gallium/drivers/r600/radeon_video.c
@@ -97,11 +97,13 @@ bool rvid_resize_buffer(struct pipe_screen *screen, struct radeon_cmdbuf *cs,
 	if (!rvid_create_buffer(screen, new_buf, new_size, new_buf->usage))
 		goto error;
 
-	src = ws->buffer_map(old_buf.res->buf, cs, PIPE_TRANSFER_READ);
+	src = ws->buffer_map(old_buf.res->buf, cs,
+			     PIPE_TRANSFER_READ | RADEON_TRANSFER_TEMPORARY);
 	if (!src)
 		goto error;
 
-	dst = ws->buffer_map(new_buf->res->buf, cs, PIPE_TRANSFER_WRITE);
+	dst = ws->buffer_map(new_buf->res->buf, cs,
+			     PIPE_TRANSFER_WRITE | RADEON_TRANSFER_TEMPORARY);
 	if (!dst)
 		goto error;