r600g/compute: Use a constant buffer to store kernel parameters v2

v2:
  - Fix usage of set_constant_buffer()
  - Fix typo in comment

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <maraeo@gmail.com>

diff --git a/src/gallium/drivers/r600/evergreen_compute.c b/src/gallium/drivers/r600/evergreen_compute.c
index 05e5e3d1ef2d6bd24a948784671718fca59ad3ea..b0151a86f995746d260199c91b3ce602066e86ac 100644 (file)
--- a/src/gallium/drivers/r600/evergreen_compute.c
+++ b/src/gallium/drivers/r600/evergreen_compute.c
@@ -102,6 +102,22 @@ static void evergreen_cs_set_vertex_buffer(
        state->atom.dirty = true;
 }
 
+static void evergreen_cs_set_constant_buffer(
+       struct r600_context * rctx,
+       unsigned cb_index,
+       unsigned offset,
+       unsigned size,
+       struct pipe_resource * buffer)
+{
+       struct pipe_constant_buffer cb;
+       cb.buffer_size = size;
+       cb.buffer_offset = offset;
+       cb.buffer = buffer;
+       cb.user_buffer = NULL;
+
+       rctx->context.set_constant_buffer(&rctx->context, PIPE_SHADER_COMPUTE, cb_index, &cb);
+}
+
 static const struct u_resource_vtbl r600_global_buffer_vtbl =
 {
        u_default_resource_get_handle, /* get_handle */
@@ -189,7 +205,10 @@ void evergreen_compute_upload_input(
        struct r600_context *ctx = (struct r600_context *)ctx_;
        struct r600_pipe_compute *shader = ctx->cs_shader_state.shader;
        int i;
-       unsigned kernel_parameters_offset_bytes = 36;
+       /* We need to reserve 9 dwords (36 bytes) for implicit kernel
+        * parameters.
+        */
+       unsigned input_size = shader->input_size + 36;
        uint32_t * num_work_groups_start;
        uint32_t * global_size_start;
        uint32_t * local_size_start;
@@ -200,12 +219,9 @@ void evergreen_compute_upload_input(
        }
 
        if (!shader->kernel_param) {
-               unsigned buffer_size = shader->input_size;
-
                /* Add space for the grid dimensions */
-               buffer_size += kernel_parameters_offset_bytes;
                shader->kernel_param = r600_compute_buffer_alloc_vram(
-                                               ctx->screen, buffer_size);
+                                               ctx->screen, input_size);
        }
 
        num_work_groups_start = r600_buffer_mmap_sync_with_rings(ctx, shader->kernel_param, PIPE_TRANSFER_WRITE);
@@ -227,20 +243,16 @@ void evergreen_compute_upload_input(
        /* Copy the kernel inputs */
        memcpy(kernel_parameters_start, input, shader->input_size);
 
-       for (i = 0; i < (kernel_parameters_offset_bytes / 4) +
-                                       (shader->input_size / 4); i++) {
+       for (i = 0; i < (input_size / 4); i++) {
                COMPUTE_DBG(ctx->screen, "input %i : %i\n", i,
                        ((unsigned*)num_work_groups_start)[i]);
        }
 
        ctx->ws->buffer_unmap(shader->kernel_param->cs_buf);
 
-       ///ID=0 is reserved for the parameters
-       evergreen_cs_set_vertex_buffer(ctx, 0, 0,
+       /* ID=0 is reserved for the parameters */
+       evergreen_cs_set_constant_buffer(ctx, 0, 0, input_size,
                        (struct pipe_resource*)shader->kernel_param);
-       ///ID=0 is reserved for parameters
-       evergreen_set_const_cache(shader, 0, shader->kernel_param,
-                                               shader->input_size, 0);
 }
 
 static void evergreen_emit_direct_dispatch(
@@ -369,6 +381,9 @@ static void compute_emit_cs(struct r600_context *ctx, const uint *block_layout,
        ctx->cs_vertex_buffer_state.atom.num_dw = 12 * util_bitcount(ctx->cs_vertex_buffer_state.dirty_mask);
        r600_emit_atom(ctx, &ctx->cs_vertex_buffer_state.atom);
 
+       /* Emit constant buffer state */
+       r600_emit_atom(ctx, &ctx->constbuf_state[PIPE_SHADER_COMPUTE].atom);
+
        /* Emit compute shader state */
        r600_emit_atom(ctx, &ctx->cs_shader_state.atom);
 
@@ -783,10 +798,9 @@ void evergreen_init_compute_state_functions(struct r600_context *ctx)
        ctx->context.set_global_binding = evergreen_set_global_binding;
        ctx->context.launch_grid = evergreen_launch_grid;
 
-       /* We always use at least two vertex buffers for compute, one for
-         * parameters and one for global memory */
+       /* We always use at least one vertex buffer for parameters (id = 1)*/
        ctx->cs_vertex_buffer_state.enabled_mask =
-       ctx->cs_vertex_buffer_state.dirty_mask = 1 | 2;
+       ctx->cs_vertex_buffer_state.dirty_mask = 0x2;
 }
 
 

diff --git a/src/gallium/drivers/r600/evergreen_compute_internal.h b/src/gallium/drivers/r600/evergreen_compute_internal.h
index b1a180f609af71e7003d113d499bba51f7b36ace..463cc7d89e1c1cce4701ab377cddbdaeeebd6272 100644 (file)
--- a/src/gallium/drivers/r600/evergreen_compute_internal.h
+++ b/src/gallium/drivers/r600/evergreen_compute_internal.h
@@ -113,7 +113,7 @@ void evergreen_set_loop_const(struct r600_pipe_compute *pipe, int id, int count,
 void evergreen_set_tmp_ring(struct r600_pipe_compute *pipe, struct r600_resource* bo, int offset, int size, int se);
 void evergreen_set_tex_resource(struct r600_pipe_compute *pipe, struct r600_pipe_sampler_view* view, int id);
 void evergreen_set_sampler_resource(struct r600_pipe_compute *pipe, struct compute_sampler_state *sampler, int id);
-void evergreen_set_const_cache(struct r600_pipe_compute *pipe, int cache_id, struct r600_resource* cbo, int size, int offset);
+void evergreen_set_const_cache(struct r600_pipe_compute *pipe, int cache_id, struct r600_resource* cbo, int size);
 
 struct r600_resource* r600_compute_buffer_alloc_vram(struct r600_screen *screen, unsigned size);