[Mesa-dev] [PATCH 4/5] radeonsi/compute: Pass kernel arguments in a buffer

[Tom Stellard]

From: Tom Stellard <thomas.stellard at amd.com>

---
 src/gallium/drivers/radeonsi/r600_buffer.c      | 31 +++++++++++++++++++++++++
 src/gallium/drivers/radeonsi/radeonsi_compute.c | 26 ++++++++++-----------
 src/gallium/drivers/radeonsi/si_state.c         | 29 +++++++----------------
 3 files changed, 51 insertions(+), 35 deletions(-)

diff --git a/src/gallium/drivers/radeonsi/r600_buffer.c b/src/gallium/drivers/radeonsi/r600_buffer.c
index cdf9988..87763c3 100644
--- a/src/gallium/drivers/radeonsi/r600_buffer.c
+++ b/src/gallium/drivers/radeonsi/r600_buffer.c
@@ -25,6 +25,8 @@
  *      Corbin Simpson <MostAwesomeDude at gmail.com>
  */
 
+#include <byteswap.h>
+
 #include "pipe/p_screen.h"
 #include "util/u_format.h"
 #include "util/u_math.h"
@@ -168,3 +170,32 @@ void r600_upload_index_buffer(struct r600_context *rctx,
 	u_upload_data(rctx->uploader, 0, count * ib->index_size,
 		      ib->user_buffer, &ib->offset, &ib->buffer);
 }
+
+void r600_upload_const_buffer(struct r600_context *rctx, struct si_resource **rbuffer,
+			const uint8_t *ptr, unsigned size,
+			uint32_t *const_offset)
+{
+	*rbuffer = NULL;
+
+	if (R600_BIG_ENDIAN) {
+		uint32_t *tmpPtr;
+		unsigned i;
+
+		if (!(tmpPtr = malloc(size))) {
+			R600_ERR("Failed to allocate BE swap buffer.\n");
+			return;
+		}
+
+		for (i = 0; i < size / 4; ++i) {
+			tmpPtr[i] = bswap_32(((uint32_t *)ptr)[i]);
+		}
+
+		u_upload_data(rctx->uploader, 0, size, tmpPtr, const_offset,
+				(struct pipe_resource**)rbuffer);
+
+		free(tmpPtr);
+	} else {
+		u_upload_data(rctx->uploader, 0, size, ptr, const_offset,
+					(struct pipe_resource**)rbuffer);
+	}
+}
diff --git a/src/gallium/drivers/radeonsi/radeonsi_compute.c b/src/gallium/drivers/radeonsi/radeonsi_compute.c
index 3fb6eb1..035076d 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_compute.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_compute.c
@@ -91,8 +91,11 @@ static void radeonsi_launch_grid(
 	struct r600_context *rctx = (struct r600_context*)ctx;
 	struct si_pipe_compute *program = rctx->cs_shader_state.program;
 	struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
+	struct si_resource *input_buffer;
+	uint32_t input_offset = 0;
+	uint64_t input_va;
 	uint64_t shader_va;
-	unsigned arg_user_sgpr_count;
+	unsigned arg_user_sgpr_count = 2;
 	unsigned i;
 	struct si_pipe_shader *shader = &program->kernels[pc];
 
@@ -109,21 +112,16 @@ static void radeonsi_launch_grid(
 	si_pm4_inval_shader_cache(pm4);
 	si_cmd_surface_sync(pm4, pm4->cp_coher_cntl);
 
-	arg_user_sgpr_count = program->input_size / 4;
-	if (program->input_size % 4 != 0) {
-		arg_user_sgpr_count++;
-	}
+	/* Upload the input data */
+	r600_upload_const_buffer(rctx, &input_buffer, input,
+					program->input_size, &input_offset);
+	input_va = r600_resource_va(ctx->screen, (struct pipe_resource*)input_buffer);
+	input_va += input_offset;
 
-	/* XXX: We should store arguments in memory if we run out of user sgprs.
-	 */
-	assert(arg_user_sgpr_count < 16);
+	si_pm4_add_bo(pm4, input_buffer, RADEON_USAGE_READ);
 
-	for (i = 0; i < arg_user_sgpr_count; i++) {
-		uint32_t *args = (uint32_t*)input;
-		si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0 +
-					(i * 4),
-					args[i]);
-	}
+	si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0, input_va);
+	si_pm4_set_reg(pm4, R_00B900_COMPUTE_USER_DATA_0 + 4, S_008F04_BASE_ADDRESS_HI (input_va >> 32) | S_008F04_STRIDE(0));
 
 	si_pm4_set_reg(pm4, R_00B810_COMPUTE_START_X, 0);
 	si_pm4_set_reg(pm4, R_00B814_COMPUTE_START_Y, 0);
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index dec535c..1e94f7e 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -24,8 +24,6 @@
  *      Christian König <christian.koenig at amd.com>
  */
 
-#include <byteswap.h>
-
 #include "util/u_memory.h"
 #include "util/u_framebuffer.h"
 #include "util/u_blitter.h"
@@ -2526,25 +2524,14 @@ static void si_set_constant_buffer(struct pipe_context *ctx, uint shader, uint i
 	ptr = input->user_buffer;
 
 	if (ptr) {
-		/* Upload the user buffer. */
-		if (R600_BIG_ENDIAN) {
-			uint32_t *tmpPtr;
-			unsigned i, size = input->buffer_size;
-
-			if (!(tmpPtr = malloc(size))) {
-				R600_ERR("Failed to allocate BE swap buffer.\n");
-				return;
-			}
-
-			for (i = 0; i < size / 4; ++i) {
-				tmpPtr[i] = bswap_32(((uint32_t *)ptr)[i]);
-			}
-
-			u_upload_data(rctx->uploader, 0, size, tmpPtr, &cb->buffer_offset, &cb->buffer);
-			free(tmpPtr);
-		} else {
-			u_upload_data(rctx->uploader, 0, input->buffer_size, ptr, &cb->buffer_offset, &cb->buffer);
-		}
+		/* XXX:
+		 * Using this function here causes a memory leak in X and makes
+		 * it impossible to do a full piglit run.  I'm not sure why this
+		 * is happening since we were using this function prior to
+		 * eb19163a4dd3d7bfeed63229820c926f99ed00d9
+		 */
+		r600_upload_const_buffer(rctx, &cb->buffer, ptr, cb->buffer_size,
+							&cb->buffer_offset);
 	} else {
 		/* Setup the hw buffer. */
 		cb->buffer_offset = input->buffer_offset;
-- 
1.8.1.5