drm/vc4: Store the validated shader state.

This massively reduces the overhead of CL submission in the kernel.

Signed-off-by: Eric Anholt <eric@anholt.net>

4 files changed, 12 insertions, 5 deletions

diff --git a/drivers/gpu/drm/vc4/vc4_bo.c b/drivers/gpu/drm/vc4/vc4_bo.c
index 8bfca3b0b3e1..93be736c87bf 100644
--- a/drivers/gpu/drm/vc4/vc4_bo.c
+++ b/drivers/gpu/drm/vc4/vc4_bo.c
@@ -174,6 +174,9 @@ vc4_free_object(struct drm_gem_object *gem_bo)
 		return;
 	}
 
+	kfree(bo->validated_shader);
+	bo->validated_shader = NULL;
+
 	bo->free_time = jiffies;
 	list_add(&bo->size_head, cache_list);
 	list_add(&bo->unref_head, &vc4->bo_cache.time_list);
diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h
index c7da68b854fe..0f71effb1cab 100644
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -99,6 +99,9 @@ struct vc4_bo {
 
 	/* List entry for the BO's position in vc4_dev->bo_cache.size_list */
 	struct list_head size_head;
+
+	/* Cached state from validation of the shader code. */
+	struct vc4_validated_shader_info *validated_shader;
 };
 
 static inline struct vc4_bo *
diff --git a/drivers/gpu/drm/vc4/vc4_validate.c b/drivers/gpu/drm/vc4/vc4_validate.c
index 0691a8d769ca..5ecd640f625a 100644
--- a/drivers/gpu/drm/vc4/vc4_validate.c
+++ b/drivers/gpu/drm/vc4/vc4_validate.c
@@ -1003,7 +1003,7 @@ validate_shader_rec(struct drm_device *dev,
 	struct drm_gem_cma_object *bo[ARRAY_SIZE(gl_relocs) + 8];
 	uint32_t nr_attributes = 0, nr_fixed_relocs, nr_relocs, packet_size;
 	int i;
-	struct vc4_validated_shader_info *validated_shader = NULL;
+	struct vc4_validated_shader_info *validated_shader;
 
 	if (state->packet == VC4_PACKET_NV_SHADER_STATE) {
 		relocs = nv_relocs;
@@ -1080,7 +1080,6 @@ validate_shader_rec(struct drm_device *dev,
 				goto fail;
 			}
 
-			kfree(validated_shader);
 			validated_shader = vc4_validate_shader(bo[i]);
 			if (!validated_shader)
 				goto fail;
@@ -1153,12 +1152,9 @@ validate_shader_rec(struct drm_device *dev,
 		*(uint32_t *)(pkt_v + o) = vbo->paddr + offset;
 	}
 
-	kfree(validated_shader);
-
 	return 0;
 
 fail:
-	kfree(validated_shader);
 	return -EINVAL;
 }
 
diff --git a/drivers/gpu/drm/vc4/vc4_validate_shaders.c b/drivers/gpu/drm/vc4/vc4_validate_shaders.c
index ad7ebb11f2d8..17c5bdcf5aa7 100644
--- a/drivers/gpu/drm/vc4/vc4_validate_shaders.c
+++ b/drivers/gpu/drm/vc4/vc4_validate_shaders.c
@@ -369,6 +369,7 @@ check_instruction_reads(uint64_t inst,
 struct vc4_validated_shader_info *
 vc4_validate_shader(struct drm_gem_cma_object *shader_obj)
 {
+	struct vc4_bo *shader_bo = to_vc4_bo(&shader_obj->base);
 	bool found_shader_end = false;
 	int shader_end_ip = 0;
 	uint32_t ip, max_ip;
@@ -377,6 +378,9 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj)
 	struct vc4_shader_validation_state validation_state;
 	int i;
 
+	if (shader_bo->validated_shader)
+		return shader_bo->validated_shader;
+
 	memset(&validation_state, 0, sizeof(validation_state));
 
 	for (i = 0; i < 8; i++)
@@ -456,6 +460,7 @@ vc4_validate_shader(struct drm_gem_cma_object *shader_obj)
 		(validated_shader->uniforms_size +
 		 4 * validated_shader->num_texture_samples);
 
+	shader_bo->validated_shader = validated_shader;
 	return validated_shader;
 
 fail: