6 files changed, 43 insertions, 41 deletions
diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c b/src/gallium/drivers/radeonsi/si_cp_dma.c
index 80673f3f5f2..59360c0d4aa 100644
--- a/src/gallium/drivers/radeonsi/si_cp_dma.c
+++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
@@ -581,3 +581,28 @@ void si_test_gds(struct si_context *sctx)
 	pipe_resource_reference(&dst, NULL);
 	exit(0);
 }
+
+void si_cp_write_data(struct si_context *sctx, struct r600_resource *buf,
+		      unsigned offset, unsigned size, unsigned dst_sel,
+		      unsigned engine, const void *data)
+{
+	struct radeon_cmdbuf *cs = sctx->gfx_cs;
+
+	assert(offset % 4 == 0);
+	assert(size % 4 == 0);
+
+	if (sctx->chip_class == SI && dst_sel == V_370_MEM)
+		dst_sel = V_370_MEM_GRBM;
+
+	radeon_add_to_buffer_list(sctx, cs, buf,
+				  RADEON_USAGE_WRITE, RADEON_PRIO_CP_DMA);
+	uint64_t va = buf->gpu_address + offset;
+
+	radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 2 + size/4, 0));
+	radeon_emit(cs, S_370_DST_SEL(dst_sel) |
+		    S_370_WR_CONFIRM(1) |
+		    S_370_ENGINE_SEL(engine));
+	radeon_emit(cs, va);
+	radeon_emit(cs, va >> 32);
+	radeon_emit_array(cs, (const uint32_t*)data, size/4);
+}
diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index 71ae00c53cb..ca62848296b 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -1821,7 +1821,6 @@ static void si_upload_bindless_descriptor(struct si_context *sctx,
 					  unsigned num_dwords)
 {
 	struct si_descriptors *desc = &sctx->bindless_descriptors;
-	struct radeon_cmdbuf *cs = sctx->gfx_cs;
 	unsigned desc_slot_offset = desc_slot * 16;
 	uint32_t *data;
 	uint64_t va;
@@ -1829,13 +1828,8 @@ static void si_upload_bindless_descriptor(struct si_context *sctx,
 	data = desc->list + desc_slot_offset;
 	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) |
-		    S_370_WR_CONFIRM(1) |
-		    S_370_ENGINE_SEL(V_370_ME));
-	radeon_emit(cs, va);
-	radeon_emit(cs, va >> 32);
-	radeon_emit_array(cs, data, num_dwords);
+	si_cp_write_data(sctx, desc->buffer, va - desc->buffer->gpu_address,
+			 num_dwords * 4, V_370_TC_L2, V_370_ME, data);
 }
 
 static void si_upload_bindless_descriptors(struct si_context *sctx)
diff --git a/src/gallium/drivers/radeonsi/si_fence.c b/src/gallium/drivers/radeonsi/si_fence.c
index 46d0289c90b..84bf4d10c20 100644
--- a/src/gallium/drivers/radeonsi/si_fence.c
+++ b/src/gallium/drivers/radeonsi/si_fence.c
@@ -259,21 +259,16 @@ static void si_fine_fence_set(struct si_context *ctx,
 
 	*fence_ptr = 0;
 
-	uint64_t fence_va = fine->buf->gpu_address + fine->offset;
-
-	radeon_add_to_buffer_list(ctx, ctx->gfx_cs, fine->buf,
-				  RADEON_USAGE_WRITE, RADEON_PRIO_QUERY);
 	if (flags & PIPE_FLUSH_TOP_OF_PIPE) {
-		struct radeon_cmdbuf *cs = ctx->gfx_cs;
-		radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
-		radeon_emit(cs, S_370_DST_SEL(ctx->chip_class >= CIK ? V_370_MEM
-								     : V_370_MEM_GRBM) |
-			S_370_WR_CONFIRM(1) |
-			S_370_ENGINE_SEL(V_370_PFP));
-		radeon_emit(cs, fence_va);
-		radeon_emit(cs, fence_va >> 32);
-		radeon_emit(cs, 0x80000000);
+		uint32_t value = 0x80000000;
+
+		si_cp_write_data(ctx, fine->buf, fine->offset, 4,
+				 V_370_MEM, V_370_PFP, &value);
 	} else if (flags & PIPE_FLUSH_BOTTOM_OF_PIPE) {
+		uint64_t fence_va = fine->buf->gpu_address + fine->offset;
+
+		radeon_add_to_buffer_list(ctx, ctx->gfx_cs, fine->buf,
+					  RADEON_USAGE_WRITE, RADEON_PRIO_QUERY);
 		si_cp_release_mem(ctx,
 				  V_028A90_BOTTOM_OF_PIPE_TS, 0,
 				  EOP_DST_SEL_MEM, EOP_INT_SEL_NONE,
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index b6953b8bd27..4edb25494ea 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -526,17 +526,8 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen,
 			goto fail;
 
 		/* Initialize the memory. */
-		struct radeon_cmdbuf *cs = sctx->gfx_cs;
-		radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
-		radeon_emit(cs, S_370_DST_SEL(sctx->chip_class >= CIK ? V_370_MEM
-								      : V_370_MEM_GRBM) |
-			    S_370_WR_CONFIRM(1) |
-			    S_370_ENGINE_SEL(V_370_ME));
-		radeon_emit(cs, sctx->wait_mem_scratch->gpu_address);
-		radeon_emit(cs, sctx->wait_mem_scratch->gpu_address >> 32);
-		radeon_emit(cs, sctx->wait_mem_number);
-		radeon_add_to_buffer_list(sctx, cs, sctx->wait_mem_scratch,
-					  RADEON_USAGE_WRITE, RADEON_PRIO_FENCE);
+		si_cp_write_data(sctx, sctx->wait_mem_scratch, 0, 4,
+				 V_370_MEM, V_370_ME, &sctx->wait_mem_number);
 	}
 
 	/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD doesn't skip loads
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index 957629e4633..89a93182ed3 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -1176,6 +1176,9 @@ void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf
 			      uint64_t offset, unsigned size);
 void cik_emit_prefetch_L2(struct si_context *sctx, bool vertex_stage_only);
 void si_test_gds(struct si_context *sctx);
+void si_cp_write_data(struct si_context *sctx, struct r600_resource *buf,
+		      unsigned offset, unsigned size, unsigned dst_sel,
+		      unsigned engine, const void *data);
 
 /* si_debug.c */
 void si_save_cs(struct radeon_winsys *ws, struct radeon_cmdbuf *cs,
diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index 9a80bd81327..1ff74e77433 100644
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -1592,17 +1592,11 @@ si_draw_rectangle(struct blitter_context *blitter,
 void si_trace_emit(struct si_context *sctx)
 {
 	struct radeon_cmdbuf *cs = sctx->gfx_cs;
-	uint64_t va = sctx->current_saved_cs->trace_buf->gpu_address;
 	uint32_t trace_id = ++sctx->current_saved_cs->trace_id;
 
-	radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
-	radeon_emit(cs, S_370_DST_SEL(sctx->chip_class >= CIK ? V_370_MEM
-							      : V_370_MEM_GRBM) |
-		    S_370_WR_CONFIRM(1) |
-		    S_370_ENGINE_SEL(V_370_ME));
-	radeon_emit(cs, va);
-	radeon_emit(cs, va >> 32);
-	radeon_emit(cs, trace_id);
+	si_cp_write_data(sctx, sctx->current_saved_cs->trace_buf,
+			 0, 4, V_370_MEM, V_370_ME, &trace_id);
+
 	radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
 	radeon_emit(cs, AC_ENCODE_TRACE_POINT(trace_id));