radv: introduce perf test env var and allow to enable chaining

We have some features that seem to slow things down or cause other
possible undesireable side effects, but it would be nice to test
games etc with them easily.

I forsee multisample DCC and maybe some shader opt changes using this.

For now use it for batch chaining.

Reviewed-by: Bas Nieuwenhuizen <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>

7 files changed, 20 insertions, 4 deletions

diff --git a/src/amd/vulkan/radv_debug.h b/src/amd/vulkan/radv_debug.h
index 4d1398e191d..c986c7bc9d7 100644
--- a/src/amd/vulkan/radv_debug.h
+++ b/src/amd/vulkan/radv_debug.h
@@ -37,4 +37,7 @@ enum {
 	RADV_DEBUG_NO_IBS            = 0x200,
 };
 
+enum {
+	RADV_PERFTEST_BATCHCHAIN     =   0x1,
+};
 #endif
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 5fdb894146c..1ea69608a14 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -270,7 +270,8 @@ radv_physical_device_init(struct radv_physical_device *device,
 	assert(strlen(path) < ARRAY_SIZE(device->path));
 	strncpy(device->path, path, ARRAY_SIZE(device->path));
 
-	device->ws = radv_amdgpu_winsys_create(fd, instance->debug_flags);
+	device->ws = radv_amdgpu_winsys_create(fd, instance->debug_flags,
+					       instance->perftest_flags);
 	if (!device->ws) {
 		result = VK_ERROR_INCOMPATIBLE_DRIVER;
 		goto fail;
@@ -367,6 +368,11 @@ static const struct debug_control radv_debug_options[] = {
 	{NULL, 0}
 };
 
+static const struct debug_control radv_perftest_options[] = {
+	{"batchchain", RADV_PERFTEST_BATCHCHAIN},
+	{NULL, 0}
+};
+
 VkResult radv_CreateInstance(
 	const VkInstanceCreateInfo*                 pCreateInfo,
 	const VkAllocationCallbacks*                pAllocator,
@@ -424,6 +430,9 @@ VkResult radv_CreateInstance(
 	instance->debug_flags = parse_debug_string(getenv("RADV_DEBUG"),
 						   radv_debug_options);
 
+	instance->perftest_flags = parse_debug_string(getenv("RADV_PERFTEST"),
+						   radv_perftest_options);
+
 	*pInstance = radv_instance_to_handle(instance);
 
 	return VK_SUCCESS;
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index c21b17e8be4..87cb0a67fe7 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -288,6 +288,7 @@ struct radv_instance {
 	struct radv_physical_device                 physicalDevices[RADV_MAX_DRM_DEVICES];
 
 	uint64_t debug_flags;
+	uint64_t perftest_flags;
 };
 
 VkResult radv_init_wsi(struct radv_physical_device *physical_device);
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index 5ea12188e87..7b749700d1c 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -931,7 +931,7 @@ static int radv_amdgpu_winsys_cs_submit(struct radeon_winsys_ctx *_ctx,
 	if (!cs->ws->use_ib_bos) {
 		ret = radv_amdgpu_winsys_cs_submit_sysmem(_ctx, queue_idx, cs_array,
 							   cs_count, initial_preamble_cs, continue_preamble_cs, _fence);
-	} else if (can_patch && cs_count > AMDGPU_CS_MAX_IBS_PER_SUBMIT && false) {
+	} else if (can_patch && cs_count > AMDGPU_CS_MAX_IBS_PER_SUBMIT && cs->ws->batchchain) {
 		ret = radv_amdgpu_winsys_cs_submit_chained(_ctx, queue_idx, cs_array,
 							    cs_count, initial_preamble_cs, continue_preamble_cs, _fence);
 	} else {
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
index bce8dd2cceb..c7688cf4c9b 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
@@ -82,7 +82,7 @@ static void radv_amdgpu_winsys_destroy(struct radeon_winsys *rws)
 }
 
 struct radeon_winsys *
-radv_amdgpu_winsys_create(int fd, uint32_t debug_flags)
+radv_amdgpu_winsys_create(int fd, uint64_t debug_flags, uint64_t perftest_flags)
 {
 	uint32_t drm_major, drm_minor, r;
 	amdgpu_device_handle dev;
@@ -106,6 +106,7 @@ radv_amdgpu_winsys_create(int fd, uint32_t debug_flags)
 	if (debug_flags & RADV_DEBUG_NO_IBS)
 		ws->use_ib_bos = false;
 
+	ws->batchchain = !!(perftest_flags & RADV_PERFTEST_BATCHCHAIN);
 	LIST_INITHEAD(&ws->global_bo_list);
 	pthread_mutex_init(&ws->global_bo_list_lock, NULL);
 	ws->base.query_info = radv_amdgpu_winsys_query_info;
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
index 59e2730b124..426cf692ec0 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
@@ -43,6 +43,7 @@ struct radv_amdgpu_winsys {
 	ADDR_HANDLE addrlib;
 
 	bool debug_all_bos;
+	bool batchchain;
 	pthread_mutex_t global_bo_list_lock;
 	struct list_head global_bo_list;
 	unsigned num_buffers;
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys_public.h b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys_public.h
index d5d0ff52c21..854e216551f 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys_public.h
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys_public.h
@@ -29,6 +29,7 @@
 #ifndef RADV_AMDGPU_WINSYS_PUBLIC_H
 #define RADV_AMDGPU_WINSYS_PUBLIC_H
 
-struct radeon_winsys *radv_amdgpu_winsys_create(int fd, uint32_t debug_flags);
+struct radeon_winsys *radv_amdgpu_winsys_create(int fd, uint64_t debug_flags,
+						uint64_t perftest_flags);
 
 #endif /* RADV_AMDGPU_WINSYS_PUBLIC_H */