Mesa (master): radeonsi: add si_descriptors:: gpu_address and remove buffer_offset

Marek Olšák mareko at kemper.freedesktop.org
Tue Oct 17 20:03:29 UTC 2017


Module: Mesa
Branch: master
Commit: ee0e1a47ce11c9199bb4a132edd5f622051f989c
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=ee0e1a47ce11c9199bb4a132edd5f622051f989c

Author: Marek Olšák <marek.olsak at amd.com>
Date:   Sun Oct  8 17:52:57 2017 +0200

radeonsi: add si_descriptors::gpu_address and remove buffer_offset

This allows us to change the pointer arbitrarily.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle at amd.com>

---

 src/gallium/drivers/radeonsi/si_cp_dma.c      |  3 ++-
 src/gallium/drivers/radeonsi/si_descriptors.c | 27 +++++++++++++++------------
 src/gallium/drivers/radeonsi/si_state.h       |  2 +-
 3 files changed, 18 insertions(+), 14 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c b/src/gallium/drivers/radeonsi/si_cp_dma.c
index 97adc27e7d..9a3971ab17 100644
--- a/src/gallium/drivers/radeonsi/si_cp_dma.c
+++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
@@ -460,7 +460,8 @@ static void cik_prefetch_VBO_descriptors(struct si_context *sctx)
 		return;
 
 	cik_prefetch_TC_L2_async(sctx, &sctx->vertex_buffers.buffer->b.b,
-				 sctx->vertex_buffers.buffer_offset,
+				 sctx->vertex_buffers.gpu_address -
+				 sctx->vertex_buffers.buffer->gpu_address,
 				 sctx->vertex_elements->desc_list_byte_size);
 }
 
diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index 3c33e85b0a..0c1fca8718 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -149,13 +149,16 @@ static bool si_upload_descriptors(struct si_context *sctx,
 		return true;
 
 	uint32_t *ptr;
+	int buffer_offset;
 	u_upload_alloc(sctx->b.b.const_uploader, 0, upload_size,
 		       si_optimal_tcc_alignment(sctx, upload_size),
-		       (unsigned*)&desc->buffer_offset,
+		       (unsigned*)&buffer_offset,
 		       (struct pipe_resource**)&desc->buffer,
 		       (void**)&ptr);
-	if (!desc->buffer)
+	if (!desc->buffer) {
+		desc->gpu_address = 0;
 		return false; /* skip the draw call */
+	}
 
 	util_memcpy_cpu_to_le32(ptr, (char*)desc->list + first_slot_offset,
 				upload_size);
@@ -165,7 +168,8 @@ static bool si_upload_descriptors(struct si_context *sctx,
                             RADEON_USAGE_READ, RADEON_PRIO_DESCRIPTORS);
 
 	/* The shader pointer should point to slot 0. */
-	desc->buffer_offset -= first_slot_offset;
+	buffer_offset -= first_slot_offset;
+	desc->gpu_address = desc->buffer->gpu_address + buffer_offset;
 
 	si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom);
 	return true;
@@ -997,14 +1001,18 @@ bool si_upload_vertex_buffer_descriptors(struct si_context *sctx)
 	 * directly through a staging buffer and don't go through
 	 * the fine-grained upload path.
 	 */
+	unsigned buffer_offset = 0;
 	u_upload_alloc(sctx->b.b.const_uploader, 0,
 		       desc_list_byte_size,
 		       si_optimal_tcc_alignment(sctx, desc_list_byte_size),
-		       (unsigned*)&desc->buffer_offset,
+		       &buffer_offset,
 		       (struct pipe_resource**)&desc->buffer, (void**)&ptr);
-	if (!desc->buffer)
+	if (!desc->buffer) {
+		desc->gpu_address = 0;
 		return false;
+	}
 
+	desc->gpu_address = desc->buffer->gpu_address + buffer_offset;
 	desc->list = ptr;
 	radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
 			      desc->buffer, RADEON_USAGE_READ,
@@ -1714,9 +1722,7 @@ static void si_upload_bindless_descriptor(struct si_context *sctx,
 	uint64_t va;
 
 	data = desc->list + desc_slot_offset;
-
-	va = desc->buffer->gpu_address + desc->buffer_offset +
-	     desc_slot_offset * 4;
+	va = desc->gpu_address + desc_slot_offset * 4;
 
 	radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 2 + num_dwords, 0));
 	radeon_emit(cs, S_370_DST_SEL(V_370_TC_L2) |
@@ -1967,10 +1973,7 @@ static void si_emit_shader_pointer_head(struct radeon_winsys_cs *cs,
 static void si_emit_shader_pointer_body(struct radeon_winsys_cs *cs,
 					struct si_descriptors *desc)
 {
-	uint64_t va = 0;
-
-	if (desc->buffer)
-		va = desc->buffer->gpu_address + desc->buffer_offset;
+	uint64_t va = desc->gpu_address;
 
 	radeon_emit(cs, va);
 	radeon_emit(cs, va >> 32);
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index 8ce3cdb66e..eb1901b5a9 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -263,7 +263,7 @@ struct si_descriptors {
 
 	/* The buffer where the descriptors have been uploaded. */
 	struct r600_resource *buffer;
-	int buffer_offset; /* can be negative if not using lower slots */
+	uint64_t gpu_address;
 
 	/* The maximum number of descriptors. */
 	uint32_t num_elements;




More information about the mesa-commit mailing list