[Mesa-dev] [PATCH 11/23] radeonsi: move all get functions to si_get.c; disk_cache_create to si_pipe.c

Marek Olšák maraeo at gmail.com
Tue Nov 28 21:38:39 UTC 2017


From: Marek Olšák <marek.olsak at amd.com>

---
 src/gallium/drivers/radeon/r600_pipe_common.c | 411 +------------
 src/gallium/drivers/radeonsi/Makefile.sources |   1 +
 src/gallium/drivers/radeonsi/meson.build      |   1 +
 src/gallium/drivers/radeonsi/si_get.c         | 852 ++++++++++++++++++++++++++
 src/gallium/drivers/radeonsi/si_pipe.c        | 486 ++-------------
 src/gallium/drivers/radeonsi/si_pipe.h        |   4 +
 6 files changed, 903 insertions(+), 852 deletions(-)
 create mode 100644 src/gallium/drivers/radeonsi/si_get.c

diff --git a/src/gallium/drivers/radeon/r600_pipe_common.c b/src/gallium/drivers/radeon/r600_pipe_common.c
index f410b27..b4e76a3 100644
--- a/src/gallium/drivers/radeon/r600_pipe_common.c
+++ b/src/gallium/drivers/radeon/r600_pipe_common.c
@@ -29,21 +29,20 @@
 #include "util/u_memory.h"
 #include "util/u_format_s3tc.h"
 #include "util/u_upload_mgr.h"
 #include "util/os_time.h"
 #include "vl/vl_decoder.h"
 #include "vl/vl_video_buffer.h"
 #include "radeon/radeon_video.h"
 #include "amd/common/ac_llvm_util.h"
 #include "amd/common/sid.h"
 #include <inttypes.h>
-#include <sys/utsname.h>
 
 #include <llvm-c/TargetMachine.h>
 
 
 /*
  * shader binary helpers.
  */
 void si_radeon_shader_binary_init(struct ac_shader_binary *b)
 {
 	memset(b, 0, sizeof(*b));
@@ -623,475 +622,67 @@ static const struct debug_named_value common_debug_options[] = {
 	{ "nodpbb", DBG(NO_DPBB), "Disable DPBB." },
 	{ "nodfsm", DBG(NO_DFSM), "Disable DFSM." },
 	{ "dpbb", DBG(DPBB), "Enable DPBB." },
 	{ "dfsm", DBG(DFSM), "Enable DFSM." },
 	{ "nooutoforder", DBG(NO_OUT_OF_ORDER), "Disable out-of-order rasterization" },
 	{ "reserve_vmid", DBG(RESERVE_VMID), "Force VMID reservation per context." },
 
 	DEBUG_NAMED_VALUE_END /* must be last */
 };
 
-static const char* r600_get_vendor(struct pipe_screen* pscreen)
-{
-	return "X.Org";
-}
-
-static const char* r600_get_device_vendor(struct pipe_screen* pscreen)
-{
-	return "AMD";
-}
-
-static const char *r600_get_marketing_name(struct radeon_winsys *ws)
-{
-	if (!ws->get_chip_name)
-		return NULL;
-	return ws->get_chip_name(ws);
-}
-
-static const char *r600_get_family_name(const struct r600_common_screen *rscreen)
-{
-	switch (rscreen->info.family) {
-	case CHIP_TAHITI: return "AMD TAHITI";
-	case CHIP_PITCAIRN: return "AMD PITCAIRN";
-	case CHIP_VERDE: return "AMD CAPE VERDE";
-	case CHIP_OLAND: return "AMD OLAND";
-	case CHIP_HAINAN: return "AMD HAINAN";
-	case CHIP_BONAIRE: return "AMD BONAIRE";
-	case CHIP_KAVERI: return "AMD KAVERI";
-	case CHIP_KABINI: return "AMD KABINI";
-	case CHIP_HAWAII: return "AMD HAWAII";
-	case CHIP_MULLINS: return "AMD MULLINS";
-	case CHIP_TONGA: return "AMD TONGA";
-	case CHIP_ICELAND: return "AMD ICELAND";
-	case CHIP_CARRIZO: return "AMD CARRIZO";
-	case CHIP_FIJI: return "AMD FIJI";
-	case CHIP_POLARIS10: return "AMD POLARIS10";
-	case CHIP_POLARIS11: return "AMD POLARIS11";
-	case CHIP_POLARIS12: return "AMD POLARIS12";
-	case CHIP_STONEY: return "AMD STONEY";
-	case CHIP_VEGA10: return "AMD VEGA10";
-	case CHIP_RAVEN: return "AMD RAVEN";
-	default: return "AMD unknown";
-	}
-}
-
-static void r600_disk_cache_create(struct r600_common_screen *rscreen)
-{
-	/* Don't use the cache if shader dumping is enabled. */
-	if (rscreen->debug_flags & DBG_ALL_SHADERS)
-		return;
-
-	/* TODO: remove this once gallium supports a nir cache */
-	if (rscreen->debug_flags & DBG(NIR))
-		return;
-
-	uint32_t mesa_timestamp;
-	if (disk_cache_get_function_timestamp(r600_disk_cache_create,
-					      &mesa_timestamp)) {
-		char *timestamp_str;
-		int res = -1;
-		uint32_t llvm_timestamp;
-
-		if (disk_cache_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo,
-						      &llvm_timestamp)) {
-			res = asprintf(&timestamp_str, "%u_%u",
-				       mesa_timestamp, llvm_timestamp);
-		}
-
-		if (res != -1) {
-			/* These flags affect shader compilation. */
-			uint64_t shader_debug_flags =
-				rscreen->debug_flags &
-				(DBG(FS_CORRECT_DERIVS_AFTER_KILL) |
-				 DBG(SI_SCHED) |
-				 DBG(UNSAFE_MATH));
-
-			rscreen->disk_shader_cache =
-				disk_cache_create(r600_get_family_name(rscreen),
-						  timestamp_str,
-						  shader_debug_flags);
-			free(timestamp_str);
-		}
-	}
-}
-
-static struct disk_cache *r600_get_disk_shader_cache(struct pipe_screen *pscreen)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen*)pscreen;
-	return rscreen->disk_shader_cache;
-}
-
-static const char* r600_get_name(struct pipe_screen* pscreen)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen*)pscreen;
-
-	return rscreen->renderer_string;
-}
-
-static float r600_get_paramf(struct pipe_screen* pscreen,
-			     enum pipe_capf param)
-{
-	switch (param) {
-	case PIPE_CAPF_MAX_LINE_WIDTH:
-	case PIPE_CAPF_MAX_LINE_WIDTH_AA:
-	case PIPE_CAPF_MAX_POINT_WIDTH:
-	case PIPE_CAPF_MAX_POINT_WIDTH_AA:
-		return 8192.0f;
-	case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
-		return 16.0f;
-	case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
-		return 16.0f;
-	case PIPE_CAPF_GUARD_BAND_LEFT:
-	case PIPE_CAPF_GUARD_BAND_TOP:
-	case PIPE_CAPF_GUARD_BAND_RIGHT:
-	case PIPE_CAPF_GUARD_BAND_BOTTOM:
-		return 0.0f;
-	}
-	return 0.0f;
-}
-
-static int r600_get_video_param(struct pipe_screen *screen,
-				enum pipe_video_profile profile,
-				enum pipe_video_entrypoint entrypoint,
-				enum pipe_video_cap param)
-{
-	switch (param) {
-	case PIPE_VIDEO_CAP_SUPPORTED:
-		return vl_profile_supported(screen, profile, entrypoint);
-	case PIPE_VIDEO_CAP_NPOT_TEXTURES:
-		return 1;
-	case PIPE_VIDEO_CAP_MAX_WIDTH:
-	case PIPE_VIDEO_CAP_MAX_HEIGHT:
-		return vl_video_buffer_max_size(screen);
-	case PIPE_VIDEO_CAP_PREFERED_FORMAT:
-		return PIPE_FORMAT_NV12;
-	case PIPE_VIDEO_CAP_PREFERS_INTERLACED:
-		return false;
-	case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED:
-		return false;
-	case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE:
-		return true;
-	case PIPE_VIDEO_CAP_MAX_LEVEL:
-		return vl_level_supported(screen, profile);
-	default:
-		return 0;
-	}
-}
-
-static unsigned get_max_threads_per_block(struct r600_common_screen *screen,
-					  enum pipe_shader_ir ir_type)
-{
-	if (ir_type != PIPE_SHADER_IR_TGSI)
-		return 256;
-
-	/* Only 16 waves per thread-group on gfx9. */
-	if (screen->chip_class >= GFX9)
-		return 1024;
-
-	/* Up to 40 waves per thread-group on GCN < gfx9. Expose a nice
-	 * round number.
-	 */
-	return 2048;
-}
-
-static int r600_get_compute_param(struct pipe_screen *screen,
-        enum pipe_shader_ir ir_type,
-        enum pipe_compute_cap param,
-        void *ret)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen *)screen;
-
-	//TODO: select these params by asic
-	switch (param) {
-	case PIPE_COMPUTE_CAP_IR_TARGET: {
-		const char *gpu;
-		const char *triple;
-
-		if (HAVE_LLVM < 0x0400)
-			triple = "amdgcn--";
-		else
-			triple = "amdgcn-mesa-mesa3d";
-
-		gpu = ac_get_llvm_processor_name(rscreen->family);
-		if (ret) {
-			sprintf(ret, "%s-%s", gpu, triple);
-		}
-		/* +2 for dash and terminating NIL byte */
-		return (strlen(triple) + strlen(gpu) + 2) * sizeof(char);
-	}
-	case PIPE_COMPUTE_CAP_GRID_DIMENSION:
-		if (ret) {
-			uint64_t *grid_dimension = ret;
-			grid_dimension[0] = 3;
-		}
-		return 1 * sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
-		if (ret) {
-			uint64_t *grid_size = ret;
-			grid_size[0] = 65535;
-			grid_size[1] = 65535;
-			grid_size[2] = 65535;
-		}
-		return 3 * sizeof(uint64_t) ;
-
-	case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
-		if (ret) {
-			uint64_t *block_size = ret;
-			unsigned threads_per_block = get_max_threads_per_block(rscreen, ir_type);
-			block_size[0] = threads_per_block;
-			block_size[1] = threads_per_block;
-			block_size[2] = threads_per_block;
-		}
-		return 3 * sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
-		if (ret) {
-			uint64_t *max_threads_per_block = ret;
-			*max_threads_per_block = get_max_threads_per_block(rscreen, ir_type);
-		}
-		return sizeof(uint64_t);
-	case PIPE_COMPUTE_CAP_ADDRESS_BITS:
-		if (ret) {
-			uint32_t *address_bits = ret;
-			address_bits[0] = 64;
-		}
-		return 1 * sizeof(uint32_t);
-
-	case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
-		if (ret) {
-			uint64_t *max_global_size = ret;
-			uint64_t max_mem_alloc_size;
-
-			r600_get_compute_param(screen, ir_type,
-				PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
-				&max_mem_alloc_size);
-
-			/* In OpenCL, the MAX_MEM_ALLOC_SIZE must be at least
-			 * 1/4 of the MAX_GLOBAL_SIZE.  Since the
-			 * MAX_MEM_ALLOC_SIZE is fixed for older kernels,
-			 * make sure we never report more than
-			 * 4 * MAX_MEM_ALLOC_SIZE.
-			 */
-			*max_global_size = MIN2(4 * max_mem_alloc_size,
-						MAX2(rscreen->info.gart_size,
-						     rscreen->info.vram_size));
-		}
-		return sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
-		if (ret) {
-			uint64_t *max_local_size = ret;
-			/* Value reported by the closed source driver. */
-			*max_local_size = 32768;
-		}
-		return sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
-		if (ret) {
-			uint64_t *max_input_size = ret;
-			/* Value reported by the closed source driver. */
-			*max_input_size = 1024;
-		}
-		return sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
-		if (ret) {
-			uint64_t *max_mem_alloc_size = ret;
-
-			*max_mem_alloc_size = rscreen->info.max_alloc_size;
-		}
-		return sizeof(uint64_t);
-
-	case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
-		if (ret) {
-			uint32_t *max_clock_frequency = ret;
-			*max_clock_frequency = rscreen->info.max_shader_clock;
-		}
-		return sizeof(uint32_t);
-
-	case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
-		if (ret) {
-			uint32_t *max_compute_units = ret;
-			*max_compute_units = rscreen->info.num_good_compute_units;
-		}
-		return sizeof(uint32_t);
-
-	case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
-		if (ret) {
-			uint32_t *images_supported = ret;
-			*images_supported = 0;
-		}
-		return sizeof(uint32_t);
-	case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
-		break; /* unused */
-	case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
-		if (ret) {
-			uint32_t *subgroup_size = ret;
-			*subgroup_size = 64;
-		}
-		return sizeof(uint32_t);
-	case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
-		if (ret) {
-			uint64_t *max_variable_threads_per_block = ret;
-			if (ir_type == PIPE_SHADER_IR_TGSI)
-				*max_variable_threads_per_block = SI_MAX_VARIABLE_THREADS_PER_BLOCK;
-			else
-				*max_variable_threads_per_block = 0;
-		}
-		return sizeof(uint64_t);
-	}
-
-        fprintf(stderr, "unknown PIPE_COMPUTE_CAP %d\n", param);
-        return 0;
-}
-
-static uint64_t r600_get_timestamp(struct pipe_screen *screen)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
-
-	return 1000000 * rscreen->ws->query_value(rscreen->ws, RADEON_TIMESTAMP) /
-			rscreen->info.clock_crystal_freq;
-}
-
-static void r600_query_memory_info(struct pipe_screen *screen,
-				   struct pipe_memory_info *info)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen*)screen;
-	struct radeon_winsys *ws = rscreen->ws;
-	unsigned vram_usage, gtt_usage;
-
-	info->total_device_memory = rscreen->info.vram_size / 1024;
-	info->total_staging_memory = rscreen->info.gart_size / 1024;
-
-	/* The real TTM memory usage is somewhat random, because:
-	 *
-	 * 1) TTM delays freeing memory, because it can only free it after
-	 *    fences expire.
-	 *
-	 * 2) The memory usage can be really low if big VRAM evictions are
-	 *    taking place, but the real usage is well above the size of VRAM.
-	 *
-	 * Instead, return statistics of this process.
-	 */
-	vram_usage = ws->query_value(ws, RADEON_REQUESTED_VRAM_MEMORY) / 1024;
-	gtt_usage =  ws->query_value(ws, RADEON_REQUESTED_GTT_MEMORY) / 1024;
-
-	info->avail_device_memory =
-		vram_usage <= info->total_device_memory ?
-				info->total_device_memory - vram_usage : 0;
-	info->avail_staging_memory =
-		gtt_usage <= info->total_staging_memory ?
-				info->total_staging_memory - gtt_usage : 0;
-
-	info->device_memory_evicted =
-		ws->query_value(ws, RADEON_NUM_BYTES_MOVED) / 1024;
-
-	if (rscreen->info.drm_major == 3 && rscreen->info.drm_minor >= 4)
-		info->nr_device_memory_evictions =
-			ws->query_value(ws, RADEON_NUM_EVICTIONS);
-	else
-		/* Just return the number of evicted 64KB pages. */
-		info->nr_device_memory_evictions = info->device_memory_evicted / 64;
-}
-
 struct pipe_resource *si_resource_create_common(struct pipe_screen *screen,
 						const struct pipe_resource *templ)
 {
 	if (templ->target == PIPE_BUFFER) {
 		return si_buffer_create(screen, templ, 256);
 	} else {
 		return si_texture_create(screen, templ);
 	}
 }
 
 bool si_common_screen_init(struct r600_common_screen *rscreen,
 			   struct radeon_winsys *ws)
 {
-	char family_name[32] = {}, llvm_string[32] = {}, kernel_version[128] = {};
-	struct utsname uname_data;
-	const char *chip_name;
-
-	ws->query_info(ws, &rscreen->info);
-	rscreen->ws = ws;
-
-	if ((chip_name = r600_get_marketing_name(ws)))
-		snprintf(family_name, sizeof(family_name), "%s / ",
-			 r600_get_family_name(rscreen) + 4);
-	else
-		chip_name = r600_get_family_name(rscreen);
-
-	if (uname(&uname_data) == 0)
-		snprintf(kernel_version, sizeof(kernel_version),
-			 " / %s", uname_data.release);
-
-	if (HAVE_LLVM > 0) {
-		snprintf(llvm_string, sizeof(llvm_string),
-			 ", LLVM %i.%i.%i", (HAVE_LLVM >> 8) & 0xff,
-			 HAVE_LLVM & 0xff, MESA_LLVM_VERSION_PATCH);
-	}
-
-	snprintf(rscreen->renderer_string, sizeof(rscreen->renderer_string),
-		 "%s (%sDRM %i.%i.%i%s%s)",
-		 chip_name, family_name, rscreen->info.drm_major,
-		 rscreen->info.drm_minor, rscreen->info.drm_patchlevel,
-		 kernel_version, llvm_string);
-
-	rscreen->b.get_name = r600_get_name;
-	rscreen->b.get_vendor = r600_get_vendor;
-	rscreen->b.get_device_vendor = r600_get_device_vendor;
-	rscreen->b.get_disk_shader_cache = r600_get_disk_shader_cache;
-	rscreen->b.get_compute_param = r600_get_compute_param;
-	rscreen->b.get_paramf = r600_get_paramf;
-	rscreen->b.get_timestamp = r600_get_timestamp;
 	rscreen->b.resource_destroy = u_resource_destroy_vtbl;
 	rscreen->b.resource_from_user_memory = si_buffer_from_user_memory;
-	rscreen->b.query_memory_info = r600_query_memory_info;
 
 	if (rscreen->info.has_hw_decode) {
-		rscreen->b.get_video_param = si_vid_get_video_param;
 		rscreen->b.is_video_format_supported = si_vid_is_format_supported;
 	} else {
-		rscreen->b.get_video_param = r600_get_video_param;
 		rscreen->b.is_video_format_supported = vl_video_buffer_is_format_supported;
 	}
 
 	si_init_screen_texture_functions(rscreen);
 	si_init_screen_query_functions(rscreen);
 
-	rscreen->family = rscreen->info.family;
-	rscreen->chip_class = rscreen->info.chip_class;
 	rscreen->debug_flags |= debug_get_flags_option("R600_DEBUG", common_debug_options, 0);
 	rscreen->has_rbplus = false;
 	rscreen->rbplus_allowed = false;
 
-	r600_disk_cache_create(rscreen);
-
 	slab_create_parent(&rscreen->pool_transfers, sizeof(struct r600_transfer), 64);
 
 	rscreen->force_aniso = MIN2(16, debug_get_num_option("R600_TEX_ANISO", -1));
 	if (rscreen->force_aniso >= 0) {
 		printf("radeon: Forcing anisotropy filter to %ix\n",
 		       /* round down to a power of two */
 		       1 << util_logbase2(rscreen->force_aniso));
 	}
 
 	(void) mtx_init(&rscreen->aux_context_lock, mtx_plain);
 	(void) mtx_init(&rscreen->gpu_load_mutex, mtx_plain);
 
 	if (rscreen->debug_flags & DBG(INFO)) {
 		printf("pci (domain:bus:dev.func): %04x:%02x:%02x.%x\n",
 		       rscreen->info.pci_domain, rscreen->info.pci_bus,
 		       rscreen->info.pci_dev, rscreen->info.pci_func);
 		printf("pci_id = 0x%x\n", rscreen->info.pci_id);
-		printf("family = %i (%s)\n", rscreen->info.family,
-		       r600_get_family_name(rscreen));
+		printf("family = %i\n", rscreen->info.family);
 		printf("chip_class = %i\n", rscreen->info.chip_class);
 		printf("pte_fragment_size = %u\n", rscreen->info.pte_fragment_size);
 		printf("gart_page_size = %u\n", rscreen->info.gart_page_size);
 		printf("gart_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.gart_size, 1024*1024));
 		printf("vram_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_size, 1024*1024));
 		printf("vram_vis_size = %i MB\n", (int)DIV_ROUND_UP(rscreen->info.vram_vis_size, 1024*1024));
 		printf("max_alloc_size = %i MB\n",
 		       (int)DIV_ROUND_UP(rscreen->info.max_alloc_size, 1024*1024));
 		printf("min_alloc_size = %u\n", rscreen->info.min_alloc_size);
 		printf("has_dedicated_vram = %u\n", rscreen->info.has_dedicated_vram);
diff --git a/src/gallium/drivers/radeonsi/Makefile.sources b/src/gallium/drivers/radeonsi/Makefile.sources
index 10c99f6..b193a60 100644
--- a/src/gallium/drivers/radeonsi/Makefile.sources
+++ b/src/gallium/drivers/radeonsi/Makefile.sources
@@ -7,20 +7,21 @@ C_SOURCES := \
 	driinfo_radeonsi.h \
 	si_blit.c \
 	si_clear.c \
 	si_compute.c \
 	si_compute.h \
 	si_cp_dma.c \
 	si_debug.c \
 	si_descriptors.c \
 	si_dma.c \
 	si_fence.c \
+	si_get.c \
 	si_hw_context.c \
 	si_pipe.c \
 	si_pipe.h \
 	si_pm4.c \
 	si_pm4.h \
 	si_perfcounter.c \
 	si_public.h \
 	si_shader.c \
 	si_shader.h \
 	si_shader_internal.h \
diff --git a/src/gallium/drivers/radeonsi/meson.build b/src/gallium/drivers/radeonsi/meson.build
index 7eae9d6..c4915d0 100644
--- a/src/gallium/drivers/radeonsi/meson.build
+++ b/src/gallium/drivers/radeonsi/meson.build
@@ -23,20 +23,21 @@ files_libradeonsi = files(
   'driinfo_radeonsi.h',
   'si_blit.c',
   'si_clear.c',
   'si_compute.c',
   'si_compute.h',
   'si_cp_dma.c',
   'si_debug.c',
   'si_descriptors.c',
   'si_dma.c',
   'si_fence.c',
+  'si_get.c',
   'si_hw_context.c',
   'si_pipe.c',
   'si_pipe.h',
   'si_pm4.c',
   'si_pm4.h',
   'si_perfcounter.c',
   'si_public.h',
   'si_shader.c',
   'si_shader.h',
   'si_shader_internal.h',
diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
new file mode 100644
index 0000000..349f7dd
--- /dev/null
+++ b/src/gallium/drivers/radeonsi/si_get.c
@@ -0,0 +1,852 @@
+/*
+ * Copyright 2017 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "si_pipe.h"
+#include "radeon/radeon_video.h"
+#include "ac_llvm_util.h"
+#include "vl/vl_decoder.h"
+#include "vl/vl_video_buffer.h"
+#include "compiler/nir/nir.h"
+
+#include <sys/utsname.h>
+
+static const char *si_get_vendor(struct pipe_screen *pscreen)
+{
+	return "X.Org";
+}
+
+static const char *si_get_device_vendor(struct pipe_screen *pscreen)
+{
+	return "AMD";
+}
+
+static const char *si_get_marketing_name(struct radeon_winsys *ws)
+{
+	if (!ws->get_chip_name)
+		return NULL;
+	return ws->get_chip_name(ws);
+}
+
+const char *si_get_family_name(const struct si_screen *sscreen)
+{
+	switch (sscreen->b.info.family) {
+	case CHIP_TAHITI: return "AMD TAHITI";
+	case CHIP_PITCAIRN: return "AMD PITCAIRN";
+	case CHIP_VERDE: return "AMD CAPE VERDE";
+	case CHIP_OLAND: return "AMD OLAND";
+	case CHIP_HAINAN: return "AMD HAINAN";
+	case CHIP_BONAIRE: return "AMD BONAIRE";
+	case CHIP_KAVERI: return "AMD KAVERI";
+	case CHIP_KABINI: return "AMD KABINI";
+	case CHIP_HAWAII: return "AMD HAWAII";
+	case CHIP_MULLINS: return "AMD MULLINS";
+	case CHIP_TONGA: return "AMD TONGA";
+	case CHIP_ICELAND: return "AMD ICELAND";
+	case CHIP_CARRIZO: return "AMD CARRIZO";
+	case CHIP_FIJI: return "AMD FIJI";
+	case CHIP_POLARIS10: return "AMD POLARIS10";
+	case CHIP_POLARIS11: return "AMD POLARIS11";
+	case CHIP_POLARIS12: return "AMD POLARIS12";
+	case CHIP_STONEY: return "AMD STONEY";
+	case CHIP_VEGA10: return "AMD VEGA10";
+	case CHIP_RAVEN: return "AMD RAVEN";
+	default: return "AMD unknown";
+	}
+}
+
+static bool si_have_tgsi_compute(struct si_screen *sscreen)
+{
+	/* Old kernels disallowed some register writes for SI
+	 * that are used for indirect dispatches. */
+	return (sscreen->b.chip_class >= CIK ||
+		sscreen->b.info.drm_major == 3 ||
+		(sscreen->b.info.drm_major == 2 &&
+		 sscreen->b.info.drm_minor >= 45));
+}
+
+static int si_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
+{
+	struct si_screen *sscreen = (struct si_screen *)pscreen;
+
+	switch (param) {
+	/* Supported features (boolean caps). */
+	case PIPE_CAP_ACCELERATED:
+	case PIPE_CAP_TWO_SIDED_STENCIL:
+	case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
+	case PIPE_CAP_ANISOTROPIC_FILTER:
+	case PIPE_CAP_POINT_SPRITE:
+	case PIPE_CAP_OCCLUSION_QUERY:
+	case PIPE_CAP_TEXTURE_SHADOW_MAP:
+	case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
+	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
+	case PIPE_CAP_TEXTURE_SWIZZLE:
+	case PIPE_CAP_DEPTH_CLIP_DISABLE:
+	case PIPE_CAP_SHADER_STENCIL_EXPORT:
+	case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
+	case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
+	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
+	case PIPE_CAP_SM3:
+	case PIPE_CAP_SEAMLESS_CUBE_MAP:
+	case PIPE_CAP_PRIMITIVE_RESTART:
+	case PIPE_CAP_CONDITIONAL_RENDER:
+	case PIPE_CAP_TEXTURE_BARRIER:
+	case PIPE_CAP_INDEP_BLEND_ENABLE:
+	case PIPE_CAP_INDEP_BLEND_FUNC:
+	case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
+	case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
+	case PIPE_CAP_USER_CONSTANT_BUFFERS:
+	case PIPE_CAP_START_INSTANCE:
+	case PIPE_CAP_NPOT_TEXTURES:
+	case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
+	case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
+	case PIPE_CAP_VERTEX_COLOR_CLAMPED:
+	case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
+	case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
+	case PIPE_CAP_TGSI_INSTANCEID:
+	case PIPE_CAP_COMPUTE:
+	case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
+	case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
+	case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
+	case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
+	case PIPE_CAP_CUBE_MAP_ARRAY:
+	case PIPE_CAP_SAMPLE_SHADING:
+	case PIPE_CAP_DRAW_INDIRECT:
+	case PIPE_CAP_CLIP_HALFZ:
+	case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
+	case PIPE_CAP_POLYGON_OFFSET_CLAMP:
+	case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
+	case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
+	case PIPE_CAP_TGSI_TEXCOORD:
+	case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
+	case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
+	case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
+	case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
+	case PIPE_CAP_SHAREABLE_SHADERS:
+	case PIPE_CAP_DEPTH_BOUNDS_TEST:
+	case PIPE_CAP_SAMPLER_VIEW_TARGET:
+	case PIPE_CAP_TEXTURE_QUERY_LOD:
+	case PIPE_CAP_TEXTURE_GATHER_SM5:
+	case PIPE_CAP_TGSI_TXQS:
+	case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
+	case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
+	case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
+	case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
+	case PIPE_CAP_INVALIDATE_BUFFER:
+	case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
+	case PIPE_CAP_QUERY_MEMORY_INFO:
+	case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
+	case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
+	case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
+	case PIPE_CAP_GENERATE_MIPMAP:
+	case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
+	case PIPE_CAP_STRING_MARKER:
+	case PIPE_CAP_CLEAR_TEXTURE:
+	case PIPE_CAP_CULL_DISTANCE:
+	case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
+	case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
+	case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
+	case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
+	case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
+	case PIPE_CAP_DOUBLES:
+	case PIPE_CAP_TGSI_TEX_TXF_LZ:
+	case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
+	case PIPE_CAP_BINDLESS_TEXTURE:
+	case PIPE_CAP_QUERY_TIMESTAMP:
+	case PIPE_CAP_QUERY_TIME_ELAPSED:
+	case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
+	case PIPE_CAP_QUERY_SO_OVERFLOW:
+	case PIPE_CAP_MEMOBJ:
+	case PIPE_CAP_LOAD_CONSTBUF:
+	case PIPE_CAP_INT64:
+	case PIPE_CAP_INT64_DIVMOD:
+	case PIPE_CAP_TGSI_CLOCK:
+	case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
+	case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
+	case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
+	case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
+		return 1;
+
+	case PIPE_CAP_TGSI_VOTE:
+		return HAVE_LLVM >= 0x0400;
+
+	case PIPE_CAP_TGSI_BALLOT:
+		return HAVE_LLVM >= 0x0500;
+
+	case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
+		return !SI_BIG_ENDIAN && sscreen->b.info.has_userptr;
+
+	case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
+		return (sscreen->b.info.drm_major == 2 &&
+			sscreen->b.info.drm_minor >= 43) ||
+		       sscreen->b.info.drm_major == 3;
+
+	case PIPE_CAP_TEXTURE_MULTISAMPLE:
+		/* 2D tiling on CIK is supported since DRM 2.35.0 */
+		return sscreen->b.chip_class < CIK ||
+		       (sscreen->b.info.drm_major == 2 &&
+			sscreen->b.info.drm_minor >= 35) ||
+		       sscreen->b.info.drm_major == 3;
+
+        case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
+                return R600_MAP_BUFFER_ALIGNMENT;
+
+	case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
+	case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
+	case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
+	case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
+	case PIPE_CAP_MAX_VERTEX_STREAMS:
+	case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
+		return 4;
+
+	case PIPE_CAP_GLSL_FEATURE_LEVEL:
+		if (sscreen->b.debug_flags & DBG(NIR))
+			return 140; /* no geometry and tessellation shaders yet */
+		if (si_have_tgsi_compute(sscreen))
+			return 450;
+		return 420;
+
+	case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
+		return MIN2(sscreen->b.info.max_alloc_size, INT_MAX);
+
+	case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
+	case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
+	case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
+		/* SI doesn't support unaligned loads.
+		 * CIK needs DRM 2.50.0 on radeon. */
+		return sscreen->b.chip_class == SI ||
+		       (sscreen->b.info.drm_major == 2 &&
+			sscreen->b.info.drm_minor < 50);
+
+	case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
+		/* TODO: GFX9 hangs. */
+		if (sscreen->b.chip_class >= GFX9)
+			return 0;
+		/* Disable on SI due to VM faults in CP DMA. Enable once these
+		 * faults are mitigated in software.
+		 */
+		if (sscreen->b.chip_class >= CIK &&
+		    sscreen->b.info.drm_major == 3 &&
+		    sscreen->b.info.drm_minor >= 13)
+			return RADEON_SPARSE_PAGE_SIZE;
+		return 0;
+
+	/* Unsupported features. */
+	case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
+	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
+	case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
+	case PIPE_CAP_USER_VERTEX_BUFFERS:
+	case PIPE_CAP_FAKE_SW_MSAA:
+	case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
+	case PIPE_CAP_VERTEXID_NOBASE:
+	case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
+	case PIPE_CAP_MAX_WINDOW_RECTANGLES:
+	case PIPE_CAP_TGSI_FS_FBFETCH:
+	case PIPE_CAP_TGSI_MUL_ZERO_WINS:
+	case PIPE_CAP_UMA:
+	case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
+	case PIPE_CAP_POST_DEPTH_COVERAGE:
+	case PIPE_CAP_TILE_RASTER_ORDER:
+	case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
+		return 0;
+
+	case PIPE_CAP_NATIVE_FENCE_FD:
+		return sscreen->b.info.has_sync_file;
+
+	case PIPE_CAP_QUERY_BUFFER_OBJECT:
+		return si_have_tgsi_compute(sscreen);
+
+	case PIPE_CAP_DRAW_PARAMETERS:
+	case PIPE_CAP_MULTI_DRAW_INDIRECT:
+	case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
+		return sscreen->has_draw_indirect_multi;
+
+	case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
+		return 30;
+
+	case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
+		return sscreen->b.chip_class <= VI ?
+			PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600 : 0;
+
+	/* Stream output. */
+	case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
+	case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
+		return 32*4;
+
+	/* Geometry shader output. */
+	case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
+		return 1024;
+	case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
+		return 4095;
+
+	case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
+		return 2048;
+
+	/* Texturing. */
+	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
+	case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
+		return 15; /* 16384 */
+	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
+		/* textures support 8192, but layered rendering supports 2048 */
+		return 12;
+	case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
+		/* textures support 8192, but layered rendering supports 2048 */
+		return 2048;
+
+	/* Viewports and render targets. */
+	case PIPE_CAP_MAX_VIEWPORTS:
+		return SI_MAX_VIEWPORTS;
+	case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
+	case PIPE_CAP_MAX_RENDER_TARGETS:
+		return 8;
+
+	case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
+	case PIPE_CAP_MIN_TEXEL_OFFSET:
+		return -32;
+
+	case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
+	case PIPE_CAP_MAX_TEXEL_OFFSET:
+		return 31;
+
+	case PIPE_CAP_ENDIANNESS:
+		return PIPE_ENDIAN_LITTLE;
+
+	case PIPE_CAP_VENDOR_ID:
+		return ATI_VENDOR_ID;
+	case PIPE_CAP_DEVICE_ID:
+		return sscreen->b.info.pci_id;
+	case PIPE_CAP_VIDEO_MEMORY:
+		return sscreen->b.info.vram_size >> 20;
+	case PIPE_CAP_PCI_GROUP:
+		return sscreen->b.info.pci_domain;
+	case PIPE_CAP_PCI_BUS:
+		return sscreen->b.info.pci_bus;
+	case PIPE_CAP_PCI_DEVICE:
+		return sscreen->b.info.pci_dev;
+	case PIPE_CAP_PCI_FUNCTION:
+		return sscreen->b.info.pci_func;
+	}
+	return 0;
+}
+
+static float si_get_paramf(struct pipe_screen* pscreen, enum pipe_capf param)
+{
+	switch (param) {
+	case PIPE_CAPF_MAX_LINE_WIDTH:
+	case PIPE_CAPF_MAX_LINE_WIDTH_AA:
+	case PIPE_CAPF_MAX_POINT_WIDTH:
+	case PIPE_CAPF_MAX_POINT_WIDTH_AA:
+		return 8192.0f;
+	case PIPE_CAPF_MAX_TEXTURE_ANISOTROPY:
+		return 16.0f;
+	case PIPE_CAPF_MAX_TEXTURE_LOD_BIAS:
+		return 16.0f;
+	case PIPE_CAPF_GUARD_BAND_LEFT:
+	case PIPE_CAPF_GUARD_BAND_TOP:
+	case PIPE_CAPF_GUARD_BAND_RIGHT:
+	case PIPE_CAPF_GUARD_BAND_BOTTOM:
+		return 0.0f;
+	}
+	return 0.0f;
+}
+
+static int si_get_shader_param(struct pipe_screen* pscreen,
+			       enum pipe_shader_type shader,
+			       enum pipe_shader_cap param)
+{
+	struct si_screen *sscreen = (struct si_screen *)pscreen;
+
+	switch(shader)
+	{
+	case PIPE_SHADER_FRAGMENT:
+	case PIPE_SHADER_VERTEX:
+	case PIPE_SHADER_GEOMETRY:
+	case PIPE_SHADER_TESS_CTRL:
+	case PIPE_SHADER_TESS_EVAL:
+		break;
+	case PIPE_SHADER_COMPUTE:
+		switch (param) {
+		case PIPE_SHADER_CAP_PREFERRED_IR:
+			return PIPE_SHADER_IR_NATIVE;
+
+		case PIPE_SHADER_CAP_SUPPORTED_IRS: {
+			int ir = 1 << PIPE_SHADER_IR_NATIVE;
+
+			if (si_have_tgsi_compute(sscreen))
+				ir |= 1 << PIPE_SHADER_IR_TGSI;
+
+			return ir;
+		}
+
+		case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE: {
+			uint64_t max_const_buffer_size;
+			pscreen->get_compute_param(pscreen, PIPE_SHADER_IR_TGSI,
+				PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
+				&max_const_buffer_size);
+			return MIN2(max_const_buffer_size, INT_MAX);
+		}
+		default:
+			/* If compute shaders don't require a special value
+			 * for this cap, we can return the same value we
+			 * do for other shader types. */
+			break;
+		}
+		break;
+	default:
+		return 0;
+	}
+
+	switch (param) {
+	/* Shader limits. */
+	case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
+	case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
+	case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
+		return 16384;
+	case PIPE_SHADER_CAP_MAX_INPUTS:
+		return shader == PIPE_SHADER_VERTEX ? SI_MAX_ATTRIBS : 32;
+	case PIPE_SHADER_CAP_MAX_OUTPUTS:
+		return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
+	case PIPE_SHADER_CAP_MAX_TEMPS:
+		return 256; /* Max native temporaries. */
+	case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
+		return 4096 * sizeof(float[4]); /* actually only memory limits this */
+	case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
+		return SI_NUM_CONST_BUFFERS;
+	case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
+	case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
+		return SI_NUM_SAMPLERS;
+	case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
+		return SI_NUM_SHADER_BUFFERS;
+	case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
+		return SI_NUM_IMAGES;
+	case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
+		return 32;
+	case PIPE_SHADER_CAP_PREFERRED_IR:
+		if (sscreen->b.debug_flags & DBG(NIR) &&
+		    (shader == PIPE_SHADER_VERTEX ||
+		     shader == PIPE_SHADER_FRAGMENT))
+			return PIPE_SHADER_IR_NIR;
+		return PIPE_SHADER_IR_TGSI;
+	case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
+		return 4;
+
+	/* Supported boolean features. */
+	case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
+	case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
+	case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
+	case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
+	case PIPE_SHADER_CAP_INTEGERS:
+	case PIPE_SHADER_CAP_INT64_ATOMICS:
+	case PIPE_SHADER_CAP_FP16:
+	case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
+	case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
+	case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
+	case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
+	case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
+	case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
+		return 1;
+
+	case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
+		/* TODO: Indirect indexing of GS inputs is unimplemented. */
+		return shader != PIPE_SHADER_GEOMETRY &&
+		       (sscreen->llvm_has_working_vgpr_indexing ||
+			/* TCS and TES load inputs directly from LDS or
+			 * offchip memory, so indirect indexing is trivial. */
+			shader == PIPE_SHADER_TESS_CTRL ||
+			shader == PIPE_SHADER_TESS_EVAL);
+
+	case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
+		return sscreen->llvm_has_working_vgpr_indexing ||
+		       /* TCS stores outputs directly to memory. */
+		       shader == PIPE_SHADER_TESS_CTRL;
+
+	/* Unsupported boolean features. */
+	case PIPE_SHADER_CAP_SUBROUTINES:
+	case PIPE_SHADER_CAP_SUPPORTED_IRS:
+	case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
+	case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
+		return 0;
+	}
+	return 0;
+}
+
+static const struct nir_shader_compiler_options nir_options = {
+	.vertex_id_zero_based = true,
+	.lower_scmp = true,
+	.lower_flrp32 = true,
+	.lower_fsat = true,
+	.lower_fdiv = true,
+	.lower_sub = true,
+	.lower_ffma = true,
+	.lower_pack_snorm_2x16 = true,
+	.lower_pack_snorm_4x8 = true,
+	.lower_pack_unorm_2x16 = true,
+	.lower_pack_unorm_4x8 = true,
+	.lower_unpack_snorm_2x16 = true,
+	.lower_unpack_snorm_4x8 = true,
+	.lower_unpack_unorm_2x16 = true,
+	.lower_unpack_unorm_4x8 = true,
+	.lower_extract_byte = true,
+	.lower_extract_word = true,
+	.max_unroll_iterations = 32,
+	.native_integers = true,
+};
+
+static const void *
+si_get_compiler_options(struct pipe_screen *screen,
+			enum pipe_shader_ir ir,
+			enum pipe_shader_type shader)
+{
+	assert(ir == PIPE_SHADER_IR_NIR);
+	return &nir_options;
+}
+
+static void si_get_driver_uuid(struct pipe_screen *pscreen, char *uuid)
+{
+	ac_compute_driver_uuid(uuid, PIPE_UUID_SIZE);
+}
+
+static void si_get_device_uuid(struct pipe_screen *pscreen, char *uuid)
+{
+	struct si_screen *sscreen = (struct si_screen *)pscreen;
+
+	ac_compute_device_uuid(&sscreen->b.info, uuid, PIPE_UUID_SIZE);
+}
+
+static const char* si_get_name(struct pipe_screen *pscreen)
+{
+	struct si_screen *sscreen = (struct si_screen*)pscreen;
+
+	return sscreen->b.renderer_string;
+}
+
+static int si_get_video_param_no_decode(struct pipe_screen *screen,
+					enum pipe_video_profile profile,
+					enum pipe_video_entrypoint entrypoint,
+					enum pipe_video_cap param)
+{
+	switch (param) {
+	case PIPE_VIDEO_CAP_SUPPORTED:
+		return vl_profile_supported(screen, profile, entrypoint);
+	case PIPE_VIDEO_CAP_NPOT_TEXTURES:
+		return 1;
+	case PIPE_VIDEO_CAP_MAX_WIDTH:
+	case PIPE_VIDEO_CAP_MAX_HEIGHT:
+		return vl_video_buffer_max_size(screen);
+	case PIPE_VIDEO_CAP_PREFERED_FORMAT:
+		return PIPE_FORMAT_NV12;
+	case PIPE_VIDEO_CAP_PREFERS_INTERLACED:
+		return false;
+	case PIPE_VIDEO_CAP_SUPPORTS_INTERLACED:
+		return false;
+	case PIPE_VIDEO_CAP_SUPPORTS_PROGRESSIVE:
+		return true;
+	case PIPE_VIDEO_CAP_MAX_LEVEL:
+		return vl_level_supported(screen, profile);
+	default:
+		return 0;
+	}
+}
+
+static unsigned get_max_threads_per_block(struct si_screen *screen,
+					  enum pipe_shader_ir ir_type)
+{
+	if (ir_type != PIPE_SHADER_IR_TGSI)
+		return 256;
+
+	/* Only 16 waves per thread-group on gfx9. */
+	if (screen->b.chip_class >= GFX9)
+		return 1024;
+
+	/* Up to 40 waves per thread-group on GCN < gfx9. Expose a nice
+	 * round number.
+	 */
+	return 2048;
+}
+
+static int si_get_compute_param(struct pipe_screen *screen,
+				enum pipe_shader_ir ir_type,
+				enum pipe_compute_cap param,
+				void *ret)
+{
+	struct si_screen *sscreen = (struct si_screen *)screen;
+
+	//TODO: select these params by asic
+	switch (param) {
+	case PIPE_COMPUTE_CAP_IR_TARGET: {
+		const char *gpu;
+		const char *triple;
+
+		if (HAVE_LLVM < 0x0400)
+			triple = "amdgcn--";
+		else
+			triple = "amdgcn-mesa-mesa3d";
+
+		gpu = ac_get_llvm_processor_name(sscreen->b.family);
+		if (ret) {
+			sprintf(ret, "%s-%s", gpu, triple);
+		}
+		/* +2 for dash and terminating NIL byte */
+		return (strlen(triple) + strlen(gpu) + 2) * sizeof(char);
+	}
+	case PIPE_COMPUTE_CAP_GRID_DIMENSION:
+		if (ret) {
+			uint64_t *grid_dimension = ret;
+			grid_dimension[0] = 3;
+		}
+		return 1 * sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_GRID_SIZE:
+		if (ret) {
+			uint64_t *grid_size = ret;
+			grid_size[0] = 65535;
+			grid_size[1] = 65535;
+			grid_size[2] = 65535;
+		}
+		return 3 * sizeof(uint64_t) ;
+
+	case PIPE_COMPUTE_CAP_MAX_BLOCK_SIZE:
+		if (ret) {
+			uint64_t *block_size = ret;
+			unsigned threads_per_block = get_max_threads_per_block(sscreen, ir_type);
+			block_size[0] = threads_per_block;
+			block_size[1] = threads_per_block;
+			block_size[2] = threads_per_block;
+		}
+		return 3 * sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK:
+		if (ret) {
+			uint64_t *max_threads_per_block = ret;
+			*max_threads_per_block = get_max_threads_per_block(sscreen, ir_type);
+		}
+		return sizeof(uint64_t);
+	case PIPE_COMPUTE_CAP_ADDRESS_BITS:
+		if (ret) {
+			uint32_t *address_bits = ret;
+			address_bits[0] = 64;
+		}
+		return 1 * sizeof(uint32_t);
+
+	case PIPE_COMPUTE_CAP_MAX_GLOBAL_SIZE:
+		if (ret) {
+			uint64_t *max_global_size = ret;
+			uint64_t max_mem_alloc_size;
+
+			si_get_compute_param(screen, ir_type,
+				PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
+				&max_mem_alloc_size);
+
+			/* In OpenCL, the MAX_MEM_ALLOC_SIZE must be at least
+			 * 1/4 of the MAX_GLOBAL_SIZE.  Since the
+			 * MAX_MEM_ALLOC_SIZE is fixed for older kernels,
+			 * make sure we never report more than
+			 * 4 * MAX_MEM_ALLOC_SIZE.
+			 */
+			*max_global_size = MIN2(4 * max_mem_alloc_size,
+						MAX2(sscreen->b.info.gart_size,
+						     sscreen->b.info.vram_size));
+		}
+		return sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_LOCAL_SIZE:
+		if (ret) {
+			uint64_t *max_local_size = ret;
+			/* Value reported by the closed source driver. */
+			*max_local_size = 32768;
+		}
+		return sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_INPUT_SIZE:
+		if (ret) {
+			uint64_t *max_input_size = ret;
+			/* Value reported by the closed source driver. */
+			*max_input_size = 1024;
+		}
+		return sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE:
+		if (ret) {
+			uint64_t *max_mem_alloc_size = ret;
+
+			*max_mem_alloc_size = sscreen->b.info.max_alloc_size;
+		}
+		return sizeof(uint64_t);
+
+	case PIPE_COMPUTE_CAP_MAX_CLOCK_FREQUENCY:
+		if (ret) {
+			uint32_t *max_clock_frequency = ret;
+			*max_clock_frequency = sscreen->b.info.max_shader_clock;
+		}
+		return sizeof(uint32_t);
+
+	case PIPE_COMPUTE_CAP_MAX_COMPUTE_UNITS:
+		if (ret) {
+			uint32_t *max_compute_units = ret;
+			*max_compute_units = sscreen->b.info.num_good_compute_units;
+		}
+		return sizeof(uint32_t);
+
+	case PIPE_COMPUTE_CAP_IMAGES_SUPPORTED:
+		if (ret) {
+			uint32_t *images_supported = ret;
+			*images_supported = 0;
+		}
+		return sizeof(uint32_t);
+	case PIPE_COMPUTE_CAP_MAX_PRIVATE_SIZE:
+		break; /* unused */
+	case PIPE_COMPUTE_CAP_SUBGROUP_SIZE:
+		if (ret) {
+			uint32_t *subgroup_size = ret;
+			*subgroup_size = 64;
+		}
+		return sizeof(uint32_t);
+	case PIPE_COMPUTE_CAP_MAX_VARIABLE_THREADS_PER_BLOCK:
+		if (ret) {
+			uint64_t *max_variable_threads_per_block = ret;
+			if (ir_type == PIPE_SHADER_IR_TGSI)
+				*max_variable_threads_per_block = SI_MAX_VARIABLE_THREADS_PER_BLOCK;
+			else
+				*max_variable_threads_per_block = 0;
+		}
+		return sizeof(uint64_t);
+	}
+
+        fprintf(stderr, "unknown PIPE_COMPUTE_CAP %d\n", param);
+        return 0;
+}
+
+static uint64_t si_get_timestamp(struct pipe_screen *screen)
+{
+	struct si_screen *sscreen = (struct si_screen*)screen;
+
+	return 1000000 * sscreen->b.ws->query_value(sscreen->b.ws, RADEON_TIMESTAMP) /
+			sscreen->b.info.clock_crystal_freq;
+}
+
+static void si_query_memory_info(struct pipe_screen *screen,
+				 struct pipe_memory_info *info)
+{
+	struct si_screen *sscreen = (struct si_screen*)screen;
+	struct radeon_winsys *ws = sscreen->b.ws;
+	unsigned vram_usage, gtt_usage;
+
+	info->total_device_memory = sscreen->b.info.vram_size / 1024;
+	info->total_staging_memory = sscreen->b.info.gart_size / 1024;
+
+	/* The real TTM memory usage is somewhat random, because:
+	 *
+	 * 1) TTM delays freeing memory, because it can only free it after
+	 *    fences expire.
+	 *
+	 * 2) The memory usage can be really low if big VRAM evictions are
+	 *    taking place, but the real usage is well above the size of VRAM.
+	 *
+	 * Instead, return statistics of this process.
+	 */
+	vram_usage = ws->query_value(ws, RADEON_REQUESTED_VRAM_MEMORY) / 1024;
+	gtt_usage =  ws->query_value(ws, RADEON_REQUESTED_GTT_MEMORY) / 1024;
+
+	info->avail_device_memory =
+		vram_usage <= info->total_device_memory ?
+				info->total_device_memory - vram_usage : 0;
+	info->avail_staging_memory =
+		gtt_usage <= info->total_staging_memory ?
+				info->total_staging_memory - gtt_usage : 0;
+
+	info->device_memory_evicted =
+		ws->query_value(ws, RADEON_NUM_BYTES_MOVED) / 1024;
+
+	if (sscreen->b.info.drm_major == 3 && sscreen->b.info.drm_minor >= 4)
+		info->nr_device_memory_evictions =
+			ws->query_value(ws, RADEON_NUM_EVICTIONS);
+	else
+		/* Just return the number of evicted 64KB pages. */
+		info->nr_device_memory_evictions = info->device_memory_evicted / 64;
+}
+
+static struct disk_cache *si_get_disk_shader_cache(struct pipe_screen *pscreen)
+{
+	struct si_screen *sscreen = (struct si_screen*)pscreen;
+
+	return sscreen->b.disk_shader_cache;
+}
+
+static void si_init_renderer_string(struct si_screen *sscreen)
+{
+	struct radeon_winsys *ws = sscreen->b.ws;
+	char family_name[32] = {}, llvm_string[32] = {}, kernel_version[128] = {};
+	struct utsname uname_data;
+
+	const char *chip_name = si_get_marketing_name(ws);
+
+	if (chip_name)
+		snprintf(family_name, sizeof(family_name), "%s / ",
+			 si_get_family_name(sscreen) + 4);
+	else
+		chip_name = si_get_family_name(sscreen);
+
+	if (uname(&uname_data) == 0)
+		snprintf(kernel_version, sizeof(kernel_version),
+			 " / %s", uname_data.release);
+
+	if (HAVE_LLVM > 0) {
+		snprintf(llvm_string, sizeof(llvm_string),
+			 ", LLVM %i.%i.%i", (HAVE_LLVM >> 8) & 0xff,
+			 HAVE_LLVM & 0xff, MESA_LLVM_VERSION_PATCH);
+	}
+
+	snprintf(sscreen->b.renderer_string, sizeof(sscreen->b.renderer_string),
+		 "%s (%sDRM %i.%i.%i%s%s)",
+		 chip_name, family_name, sscreen->b.info.drm_major,
+		 sscreen->b.info.drm_minor, sscreen->b.info.drm_patchlevel,
+		 kernel_version, llvm_string);
+}
+
+void si_init_screen_get_functions(struct si_screen *sscreen)
+{
+	sscreen->b.b.get_name = si_get_name;
+	sscreen->b.b.get_vendor = si_get_vendor;
+	sscreen->b.b.get_device_vendor = si_get_device_vendor;
+	sscreen->b.b.get_param = si_get_param;
+	sscreen->b.b.get_paramf = si_get_paramf;
+	sscreen->b.b.get_compute_param = si_get_compute_param;
+	sscreen->b.b.get_timestamp = si_get_timestamp;
+	sscreen->b.b.get_shader_param = si_get_shader_param;
+	sscreen->b.b.get_compiler_options = si_get_compiler_options;
+	sscreen->b.b.get_device_uuid = si_get_device_uuid;
+	sscreen->b.b.get_driver_uuid = si_get_driver_uuid;
+	sscreen->b.b.query_memory_info = si_query_memory_info;
+	sscreen->b.b.get_disk_shader_cache = si_get_disk_shader_cache;
+
+	if (sscreen->b.info.has_hw_decode) {
+		sscreen->b.b.get_video_param = si_vid_get_video_param;
+	} else {
+		sscreen->b.b.get_video_param = si_get_video_param_no_decode;
+	}
+
+	si_init_renderer_string(sscreen);
+}
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index eb0283a..dfb35ed 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -29,22 +29,20 @@
 #include "radeon/radeon_uvd.h"
 #include "util/hash_table.h"
 #include "util/u_log.h"
 #include "util/u_memory.h"
 #include "util/u_suballoc.h"
 #include "util/u_tests.h"
 #include "util/xmlconfig.h"
 #include "vl/vl_decoder.h"
 #include "../ddebug/dd_util.h"
 
-#include "compiler/nir/nir.h"
-
 /*
  * pipe_context
  */
 static void si_destroy_context(struct pipe_context *context)
 {
 	struct si_context *sctx = (struct si_context *)context;
 	int i;
 
 	/* Unreference the framebuffer normally to disable related logic
 	 * properly.
@@ -395,448 +393,20 @@ static struct pipe_context *si_pipe_create_context(struct pipe_screen *screen,
 	 * implementation for fence_server_sync is incomplete. */
 	return threaded_context_create(ctx, &sscreen->b.pool_transfers,
 				       si_replace_buffer_storage,
 				       sscreen->b.info.drm_major >= 3 ? si_create_fence : NULL,
 				       &((struct si_context*)ctx)->b.tc);
 }
 
 /*
  * pipe_screen
  */
-static bool si_have_tgsi_compute(struct si_screen *sscreen)
-{
-	/* Old kernels disallowed some register writes for SI
-	 * that are used for indirect dispatches. */
-	return (sscreen->b.chip_class >= CIK ||
-		sscreen->b.info.drm_major == 3 ||
-		(sscreen->b.info.drm_major == 2 &&
-		 sscreen->b.info.drm_minor >= 45));
-}
-
-static int si_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
-{
-	struct si_screen *sscreen = (struct si_screen *)pscreen;
-
-	switch (param) {
-	/* Supported features (boolean caps). */
-	case PIPE_CAP_ACCELERATED:
-	case PIPE_CAP_TWO_SIDED_STENCIL:
-	case PIPE_CAP_MAX_DUAL_SOURCE_RENDER_TARGETS:
-	case PIPE_CAP_ANISOTROPIC_FILTER:
-	case PIPE_CAP_POINT_SPRITE:
-	case PIPE_CAP_OCCLUSION_QUERY:
-	case PIPE_CAP_TEXTURE_SHADOW_MAP:
-	case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
-	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
-	case PIPE_CAP_TEXTURE_SWIZZLE:
-	case PIPE_CAP_DEPTH_CLIP_DISABLE:
-	case PIPE_CAP_SHADER_STENCIL_EXPORT:
-	case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
-	case PIPE_CAP_MIXED_COLORBUFFER_FORMATS:
-	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
-	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
-	case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER:
-	case PIPE_CAP_SM3:
-	case PIPE_CAP_SEAMLESS_CUBE_MAP:
-	case PIPE_CAP_PRIMITIVE_RESTART:
-	case PIPE_CAP_CONDITIONAL_RENDER:
-	case PIPE_CAP_TEXTURE_BARRIER:
-	case PIPE_CAP_INDEP_BLEND_ENABLE:
-	case PIPE_CAP_INDEP_BLEND_FUNC:
-	case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
-	case PIPE_CAP_VERTEX_COLOR_UNCLAMPED:
-	case PIPE_CAP_USER_CONSTANT_BUFFERS:
-	case PIPE_CAP_START_INSTANCE:
-	case PIPE_CAP_NPOT_TEXTURES:
-	case PIPE_CAP_MIXED_FRAMEBUFFER_SIZES:
-	case PIPE_CAP_MIXED_COLOR_DEPTH_BITS:
-	case PIPE_CAP_VERTEX_COLOR_CLAMPED:
-	case PIPE_CAP_FRAGMENT_COLOR_CLAMPED:
-        case PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER:
-	case PIPE_CAP_TGSI_INSTANCEID:
-	case PIPE_CAP_COMPUTE:
-	case PIPE_CAP_TEXTURE_BUFFER_OBJECTS:
-        case PIPE_CAP_TGSI_VS_LAYER_VIEWPORT:
-	case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
-	case PIPE_CAP_BUFFER_MAP_PERSISTENT_COHERENT:
-	case PIPE_CAP_CUBE_MAP_ARRAY:
-	case PIPE_CAP_SAMPLE_SHADING:
-	case PIPE_CAP_DRAW_INDIRECT:
-	case PIPE_CAP_CLIP_HALFZ:
-	case PIPE_CAP_TGSI_VS_WINDOW_SPACE_POSITION:
-	case PIPE_CAP_POLYGON_OFFSET_CLAMP:
-	case PIPE_CAP_MULTISAMPLE_Z_RESOLVE:
-	case PIPE_CAP_QUADS_FOLLOW_PROVOKING_VERTEX_CONVENTION:
-	case PIPE_CAP_TGSI_TEXCOORD:
-	case PIPE_CAP_TGSI_FS_FINE_DERIVATIVE:
-	case PIPE_CAP_CONDITIONAL_RENDER_INVERTED:
-	case PIPE_CAP_TEXTURE_FLOAT_LINEAR:
-	case PIPE_CAP_TEXTURE_HALF_FLOAT_LINEAR:
-	case PIPE_CAP_SHAREABLE_SHADERS:
-	case PIPE_CAP_DEPTH_BOUNDS_TEST:
-	case PIPE_CAP_SAMPLER_VIEW_TARGET:
-	case PIPE_CAP_TEXTURE_QUERY_LOD:
-	case PIPE_CAP_TEXTURE_GATHER_SM5:
-	case PIPE_CAP_TGSI_TXQS:
-	case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
-	case PIPE_CAP_COPY_BETWEEN_COMPRESSED_AND_PLAIN_FORMATS:
-	case PIPE_CAP_TGSI_FS_POSITION_IS_SYSVAL:
-	case PIPE_CAP_TGSI_FS_FACE_IS_INTEGER_SYSVAL:
-	case PIPE_CAP_INVALIDATE_BUFFER:
-	case PIPE_CAP_SURFACE_REINTERPRET_BLOCKS:
-	case PIPE_CAP_QUERY_MEMORY_INFO:
-	case PIPE_CAP_TGSI_PACK_HALF_FLOAT:
-	case PIPE_CAP_FRAMEBUFFER_NO_ATTACHMENT:
-	case PIPE_CAP_ROBUST_BUFFER_ACCESS_BEHAVIOR:
-	case PIPE_CAP_GENERATE_MIPMAP:
-	case PIPE_CAP_POLYGON_OFFSET_UNITS_UNSCALED:
-	case PIPE_CAP_STRING_MARKER:
-	case PIPE_CAP_CLEAR_TEXTURE:
-	case PIPE_CAP_CULL_DISTANCE:
-	case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
-	case PIPE_CAP_TGSI_CAN_READ_OUTPUTS:
-	case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
-	case PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME:
-	case PIPE_CAP_STREAM_OUTPUT_INTERLEAVE_BUFFERS:
-	case PIPE_CAP_DOUBLES:
-	case PIPE_CAP_TGSI_TEX_TXF_LZ:
-	case PIPE_CAP_TGSI_TES_LAYER_VIEWPORT:
-	case PIPE_CAP_BINDLESS_TEXTURE:
-	case PIPE_CAP_QUERY_TIMESTAMP:
-	case PIPE_CAP_QUERY_TIME_ELAPSED:
-	case PIPE_CAP_NIR_SAMPLERS_AS_DEREF:
-	case PIPE_CAP_QUERY_SO_OVERFLOW:
-	case PIPE_CAP_MEMOBJ:
-	case PIPE_CAP_LOAD_CONSTBUF:
-	case PIPE_CAP_INT64:
-	case PIPE_CAP_INT64_DIVMOD:
-	case PIPE_CAP_TGSI_CLOCK:
-	case PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX:
-	case PIPE_CAP_ALLOW_MAPPED_BUFFERS_DURING_EXECUTION:
-	case PIPE_CAP_TGSI_ANY_REG_AS_ADDRESS:
-	case PIPE_CAP_SIGNED_VERTEX_BUFFER_OFFSET:
-		return 1;
-
-	case PIPE_CAP_TGSI_VOTE:
-		return HAVE_LLVM >= 0x0400;
-
-	case PIPE_CAP_TGSI_BALLOT:
-		return HAVE_LLVM >= 0x0500;
-
-	case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
-		return !SI_BIG_ENDIAN && sscreen->b.info.has_userptr;
-
-	case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
-		return (sscreen->b.info.drm_major == 2 &&
-			sscreen->b.info.drm_minor >= 43) ||
-		       sscreen->b.info.drm_major == 3;
-
-	case PIPE_CAP_TEXTURE_MULTISAMPLE:
-		/* 2D tiling on CIK is supported since DRM 2.35.0 */
-		return sscreen->b.chip_class < CIK ||
-		       (sscreen->b.info.drm_major == 2 &&
-			sscreen->b.info.drm_minor >= 35) ||
-		       sscreen->b.info.drm_major == 3;
-
-        case PIPE_CAP_MIN_MAP_BUFFER_ALIGNMENT:
-                return R600_MAP_BUFFER_ALIGNMENT;
-
-	case PIPE_CAP_CONSTANT_BUFFER_OFFSET_ALIGNMENT:
-	case PIPE_CAP_TEXTURE_BUFFER_OFFSET_ALIGNMENT:
-	case PIPE_CAP_MAX_TEXTURE_GATHER_COMPONENTS:
-	case PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS:
-	case PIPE_CAP_MAX_VERTEX_STREAMS:
-	case PIPE_CAP_SHADER_BUFFER_OFFSET_ALIGNMENT:
-		return 4;
-
-	case PIPE_CAP_GLSL_FEATURE_LEVEL:
-		if (sscreen->b.debug_flags & DBG(NIR))
-			return 140; /* no geometry and tessellation shaders yet */
-		if (si_have_tgsi_compute(sscreen))
-			return 450;
-		return 420;
-
-	case PIPE_CAP_MAX_TEXTURE_BUFFER_SIZE:
-		return MIN2(sscreen->b.info.max_alloc_size, INT_MAX);
-
-	case PIPE_CAP_VERTEX_BUFFER_OFFSET_4BYTE_ALIGNED_ONLY:
-	case PIPE_CAP_VERTEX_BUFFER_STRIDE_4BYTE_ALIGNED_ONLY:
-	case PIPE_CAP_VERTEX_ELEMENT_SRC_OFFSET_4BYTE_ALIGNED_ONLY:
-		/* SI doesn't support unaligned loads.
-		 * CIK needs DRM 2.50.0 on radeon. */
-		return sscreen->b.chip_class == SI ||
-		       (sscreen->b.info.drm_major == 2 &&
-			sscreen->b.info.drm_minor < 50);
-
-	case PIPE_CAP_SPARSE_BUFFER_PAGE_SIZE:
-		/* TODO: GFX9 hangs. */
-		if (sscreen->b.chip_class >= GFX9)
-			return 0;
-		/* Disable on SI due to VM faults in CP DMA. Enable once these
-		 * faults are mitigated in software.
-		 */
-		if (sscreen->b.chip_class >= CIK &&
-		    sscreen->b.info.drm_major == 3 &&
-		    sscreen->b.info.drm_minor >= 13)
-			return RADEON_SPARSE_PAGE_SIZE;
-		return 0;
-
-	/* Unsupported features. */
-	case PIPE_CAP_BUFFER_SAMPLER_VIEW_RGBA_ONLY:
-	case PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT:
-	case PIPE_CAP_TGSI_CAN_COMPACT_CONSTANTS:
-	case PIPE_CAP_USER_VERTEX_BUFFERS:
-	case PIPE_CAP_FAKE_SW_MSAA:
-	case PIPE_CAP_TEXTURE_GATHER_OFFSETS:
-	case PIPE_CAP_VERTEXID_NOBASE:
-	case PIPE_CAP_PRIMITIVE_RESTART_FOR_PATCHES:
-	case PIPE_CAP_MAX_WINDOW_RECTANGLES:
-	case PIPE_CAP_TGSI_FS_FBFETCH:
-	case PIPE_CAP_TGSI_MUL_ZERO_WINS:
-	case PIPE_CAP_UMA:
-	case PIPE_CAP_POLYGON_MODE_FILL_RECTANGLE:
-	case PIPE_CAP_POST_DEPTH_COVERAGE:
-	case PIPE_CAP_TILE_RASTER_ORDER:
-	case PIPE_CAP_MAX_COMBINED_SHADER_OUTPUT_RESOURCES:
-		return 0;
-
-	case PIPE_CAP_NATIVE_FENCE_FD:
-		return sscreen->b.info.has_sync_file;
-
-	case PIPE_CAP_QUERY_BUFFER_OBJECT:
-		return si_have_tgsi_compute(sscreen);
-
-	case PIPE_CAP_DRAW_PARAMETERS:
-	case PIPE_CAP_MULTI_DRAW_INDIRECT:
-	case PIPE_CAP_MULTI_DRAW_INDIRECT_PARAMS:
-		return sscreen->has_draw_indirect_multi;
-
-	case PIPE_CAP_MAX_SHADER_PATCH_VARYINGS:
-		return 30;
-
-	case PIPE_CAP_TEXTURE_BORDER_COLOR_QUIRK:
-		return sscreen->b.chip_class <= VI ?
-			PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_R600 : 0;
-
-	/* Stream output. */
-	case PIPE_CAP_MAX_STREAM_OUTPUT_SEPARATE_COMPONENTS:
-	case PIPE_CAP_MAX_STREAM_OUTPUT_INTERLEAVED_COMPONENTS:
-		return 32*4;
-
-	/* Geometry shader output. */
-	case PIPE_CAP_MAX_GEOMETRY_OUTPUT_VERTICES:
-		return 1024;
-	case PIPE_CAP_MAX_GEOMETRY_TOTAL_OUTPUT_COMPONENTS:
-		return 4095;
-
-	case PIPE_CAP_MAX_VERTEX_ATTRIB_STRIDE:
-		return 2048;
-
-	/* Texturing. */
-	case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-	case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-		return 15; /* 16384 */
-	case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-		/* textures support 8192, but layered rendering supports 2048 */
-		return 12;
-	case PIPE_CAP_MAX_TEXTURE_ARRAY_LAYERS:
-		/* textures support 8192, but layered rendering supports 2048 */
-		return 2048;
-
-	/* Viewports and render targets. */
-	case PIPE_CAP_MAX_VIEWPORTS:
-		return SI_MAX_VIEWPORTS;
-	case PIPE_CAP_VIEWPORT_SUBPIXEL_BITS:
-	case PIPE_CAP_MAX_RENDER_TARGETS:
-		return 8;
-
- 	case PIPE_CAP_MIN_TEXTURE_GATHER_OFFSET:
-	case PIPE_CAP_MIN_TEXEL_OFFSET:
-		return -32;
-
- 	case PIPE_CAP_MAX_TEXTURE_GATHER_OFFSET:
-	case PIPE_CAP_MAX_TEXEL_OFFSET:
-		return 31;
-
-	case PIPE_CAP_ENDIANNESS:
-		return PIPE_ENDIAN_LITTLE;
-
-	case PIPE_CAP_VENDOR_ID:
-		return ATI_VENDOR_ID;
-	case PIPE_CAP_DEVICE_ID:
-		return sscreen->b.info.pci_id;
-	case PIPE_CAP_VIDEO_MEMORY:
-		return sscreen->b.info.vram_size >> 20;
-	case PIPE_CAP_PCI_GROUP:
-		return sscreen->b.info.pci_domain;
-	case PIPE_CAP_PCI_BUS:
-		return sscreen->b.info.pci_bus;
-	case PIPE_CAP_PCI_DEVICE:
-		return sscreen->b.info.pci_dev;
-	case PIPE_CAP_PCI_FUNCTION:
-		return sscreen->b.info.pci_func;
-	}
-	return 0;
-}
-
-static int si_get_shader_param(struct pipe_screen* pscreen,
-			       enum pipe_shader_type shader,
-			       enum pipe_shader_cap param)
-{
-	struct si_screen *sscreen = (struct si_screen *)pscreen;
-
-	switch(shader)
-	{
-	case PIPE_SHADER_FRAGMENT:
-	case PIPE_SHADER_VERTEX:
-	case PIPE_SHADER_GEOMETRY:
-	case PIPE_SHADER_TESS_CTRL:
-	case PIPE_SHADER_TESS_EVAL:
-		break;
-	case PIPE_SHADER_COMPUTE:
-		switch (param) {
-		case PIPE_SHADER_CAP_PREFERRED_IR:
-			return PIPE_SHADER_IR_NATIVE;
-
-		case PIPE_SHADER_CAP_SUPPORTED_IRS: {
-			int ir = 1 << PIPE_SHADER_IR_NATIVE;
-
-			if (si_have_tgsi_compute(sscreen))
-				ir |= 1 << PIPE_SHADER_IR_TGSI;
-
-			return ir;
-		}
-
-		case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE: {
-			uint64_t max_const_buffer_size;
-			pscreen->get_compute_param(pscreen, PIPE_SHADER_IR_TGSI,
-				PIPE_COMPUTE_CAP_MAX_MEM_ALLOC_SIZE,
-				&max_const_buffer_size);
-			return MIN2(max_const_buffer_size, INT_MAX);
-		}
-		default:
-			/* If compute shaders don't require a special value
-			 * for this cap, we can return the same value we
-			 * do for other shader types. */
-			break;
-		}
-		break;
-	default:
-		return 0;
-	}
-
-	switch (param) {
-	/* Shader limits. */
-	case PIPE_SHADER_CAP_MAX_INSTRUCTIONS:
-	case PIPE_SHADER_CAP_MAX_ALU_INSTRUCTIONS:
-	case PIPE_SHADER_CAP_MAX_TEX_INSTRUCTIONS:
-	case PIPE_SHADER_CAP_MAX_TEX_INDIRECTIONS:
-	case PIPE_SHADER_CAP_MAX_CONTROL_FLOW_DEPTH:
-		return 16384;
-	case PIPE_SHADER_CAP_MAX_INPUTS:
-		return shader == PIPE_SHADER_VERTEX ? SI_MAX_ATTRIBS : 32;
-	case PIPE_SHADER_CAP_MAX_OUTPUTS:
-		return shader == PIPE_SHADER_FRAGMENT ? 8 : 32;
-	case PIPE_SHADER_CAP_MAX_TEMPS:
-		return 256; /* Max native temporaries. */
-	case PIPE_SHADER_CAP_MAX_CONST_BUFFER_SIZE:
-		return 4096 * sizeof(float[4]); /* actually only memory limits this */
-	case PIPE_SHADER_CAP_MAX_CONST_BUFFERS:
-		return SI_NUM_CONST_BUFFERS;
-	case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
-	case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
-		return SI_NUM_SAMPLERS;
-	case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
-		return SI_NUM_SHADER_BUFFERS;
-	case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
-		return SI_NUM_IMAGES;
-	case PIPE_SHADER_CAP_MAX_UNROLL_ITERATIONS_HINT:
-		return 32;
-	case PIPE_SHADER_CAP_PREFERRED_IR:
-		if (sscreen->b.debug_flags & DBG(NIR) &&
-		    (shader == PIPE_SHADER_VERTEX ||
-		     shader == PIPE_SHADER_FRAGMENT))
-			return PIPE_SHADER_IR_NIR;
-		return PIPE_SHADER_IR_TGSI;
-	case PIPE_SHADER_CAP_LOWER_IF_THRESHOLD:
-		return 4;
-
-	/* Supported boolean features. */
-	case PIPE_SHADER_CAP_TGSI_CONT_SUPPORTED:
-	case PIPE_SHADER_CAP_TGSI_SQRT_SUPPORTED:
-	case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
-	case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
-	case PIPE_SHADER_CAP_INTEGERS:
-	case PIPE_SHADER_CAP_INT64_ATOMICS:
-	case PIPE_SHADER_CAP_FP16:
-	case PIPE_SHADER_CAP_TGSI_FMA_SUPPORTED:
-	case PIPE_SHADER_CAP_TGSI_ANY_INOUT_DECL_RANGE:
-	case PIPE_SHADER_CAP_TGSI_SKIP_MERGE_REGISTERS:
-	case PIPE_SHADER_CAP_TGSI_DROUND_SUPPORTED:
-	case PIPE_SHADER_CAP_TGSI_LDEXP_SUPPORTED:
-	case PIPE_SHADER_CAP_TGSI_DFRACEXP_DLDEXP_SUPPORTED:
-		return 1;
-
-	case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
-		/* TODO: Indirect indexing of GS inputs is unimplemented. */
-		return shader != PIPE_SHADER_GEOMETRY &&
-		       (sscreen->llvm_has_working_vgpr_indexing ||
-			/* TCS and TES load inputs directly from LDS or
-			 * offchip memory, so indirect indexing is trivial. */
-			shader == PIPE_SHADER_TESS_CTRL ||
-			shader == PIPE_SHADER_TESS_EVAL);
-
-	case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
-		return sscreen->llvm_has_working_vgpr_indexing ||
-		       /* TCS stores outputs directly to memory. */
-		       shader == PIPE_SHADER_TESS_CTRL;
-
-	/* Unsupported boolean features. */
-	case PIPE_SHADER_CAP_SUBROUTINES:
-	case PIPE_SHADER_CAP_SUPPORTED_IRS:
-	case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTERS:
-	case PIPE_SHADER_CAP_MAX_HW_ATOMIC_COUNTER_BUFFERS:
-		return 0;
-	}
-	return 0;
-}
-
-static const struct nir_shader_compiler_options nir_options = {
-	.vertex_id_zero_based = true,
-	.lower_scmp = true,
-	.lower_flrp32 = true,
-	.lower_fsat = true,
-	.lower_fdiv = true,
-	.lower_sub = true,
-	.lower_ffma = true,
-	.lower_pack_snorm_2x16 = true,
-	.lower_pack_snorm_4x8 = true,
-	.lower_pack_unorm_2x16 = true,
-	.lower_pack_unorm_4x8 = true,
-	.lower_unpack_snorm_2x16 = true,
-	.lower_unpack_snorm_4x8 = true,
-	.lower_unpack_unorm_2x16 = true,
-	.lower_unpack_unorm_4x8 = true,
-	.lower_extract_byte = true,
-	.lower_extract_word = true,
-	.max_unroll_iterations = 32,
-	.native_integers = true,
-};
-
-static const void *
-si_get_compiler_options(struct pipe_screen *screen,
-			enum pipe_shader_ir ir,
-			enum pipe_shader_type shader)
-{
-	assert(ir == PIPE_SHADER_IR_NIR);
-	return &nir_options;
-}
 
 static void si_destroy_screen(struct pipe_screen* pscreen)
 {
 	struct si_screen *sscreen = (struct si_screen *)pscreen;
 	struct si_shader_part *parts[] = {
 		sscreen->vs_prologs,
 		sscreen->tcs_epilogs,
 		sscreen->gs_prologs,
 		sscreen->ps_prologs,
 		sscreen->ps_epilogs
@@ -962,71 +532,103 @@ static void si_test_vmfault(struct si_screen *sscreen)
 		ctx->flush(ctx, NULL, 0);
 		puts("VM fault test: SDMA - done.");
 	}
 	if (sscreen->b.debug_flags & DBG(TEST_VMFAULT_SHADER)) {
 		util_test_constant_buffer(ctx, buf);
 		puts("VM fault test: Shader - done.");
 	}
 	exit(0);
 }
 
-static void radeonsi_get_driver_uuid(struct pipe_screen *pscreen, char *uuid)
+static void si_disk_cache_create(struct si_screen *sscreen)
 {
-	ac_compute_driver_uuid(uuid, PIPE_UUID_SIZE);
-}
+	/* Don't use the cache if shader dumping is enabled. */
+	if (sscreen->b.debug_flags & DBG_ALL_SHADERS)
+		return;
 
-static void radeonsi_get_device_uuid(struct pipe_screen *pscreen, char *uuid)
-{
-	struct r600_common_screen *rscreen = (struct r600_common_screen *)pscreen;
+	/* TODO: remove this once gallium supports a nir cache */
+	if (sscreen->b.debug_flags & DBG(NIR))
+		return;
 
-	ac_compute_device_uuid(&rscreen->info, uuid, PIPE_UUID_SIZE);
+	uint32_t mesa_timestamp;
+	if (disk_cache_get_function_timestamp(si_disk_cache_create,
+					      &mesa_timestamp)) {
+		char *timestamp_str;
+		int res = -1;
+		uint32_t llvm_timestamp;
+
+		if (disk_cache_get_function_timestamp(LLVMInitializeAMDGPUTargetInfo,
+						      &llvm_timestamp)) {
+			res = asprintf(&timestamp_str, "%u_%u",
+				       mesa_timestamp, llvm_timestamp);
+		}
+
+		if (res != -1) {
+			/* These flags affect shader compilation. */
+			uint64_t shader_debug_flags =
+				sscreen->b.debug_flags &
+				(DBG(FS_CORRECT_DERIVS_AFTER_KILL) |
+				 DBG(SI_SCHED) |
+				 DBG(UNSAFE_MATH));
+
+			sscreen->b.disk_shader_cache =
+				disk_cache_create(si_get_family_name(sscreen),
+						  timestamp_str,
+						  shader_debug_flags);
+			free(timestamp_str);
+		}
+	}
 }
 
 struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws,
 					   const struct pipe_screen_config *config)
 {
 	struct si_screen *sscreen = CALLOC_STRUCT(si_screen);
 	unsigned num_threads, num_compiler_threads, num_compiler_threads_lowprio, i;
 
 	if (!sscreen) {
 		return NULL;
 	}
 
+	sscreen->b.ws = ws;
+	ws->query_info(ws, &sscreen->b.info);
+
+	sscreen->b.family = sscreen->b.info.family;
+	sscreen->b.chip_class = sscreen->b.info.chip_class;
+
 	/* Set functions first. */
 	sscreen->b.b.context_create = si_pipe_create_context;
 	sscreen->b.b.destroy = si_destroy_screen;
-	sscreen->b.b.get_param = si_get_param;
-	sscreen->b.b.get_shader_param = si_get_shader_param;
-	sscreen->b.b.get_compiler_options = si_get_compiler_options;
-	sscreen->b.b.get_device_uuid = radeonsi_get_device_uuid;
-	sscreen->b.b.get_driver_uuid = radeonsi_get_driver_uuid;
 	sscreen->b.b.resource_create = si_resource_create_common;
 
+	si_init_screen_get_functions(sscreen);
 	si_init_screen_fence_functions(sscreen);
 	si_init_screen_state_functions(sscreen);
 
 	/* Set these flags in debug_flags early, so that the shader cache takes
 	 * them into account.
 	 */
 	if (driQueryOptionb(config->options,
 			    "glsl_correct_derivatives_after_discard"))
 		sscreen->b.debug_flags |= DBG(FS_CORRECT_DERIVS_AFTER_KILL);
 	if (driQueryOptionb(config->options, "radeonsi_enable_sisched"))
 		sscreen->b.debug_flags |= DBG(SI_SCHED);
 
 	if (!si_common_screen_init(&sscreen->b, ws) ||
 	    !si_init_gs_info(sscreen) ||
 	    !si_init_shader_cache(sscreen)) {
 		FREE(sscreen);
 		return NULL;
 	}
 
+	si_disk_cache_create(sscreen);
+
 	/* Only enable as many threads as we have target machines, but at most
 	 * the number of CPUs - 1 if there is more than one.
 	 */
 	num_threads = sysconf(_SC_NPROCESSORS_ONLN);
 	num_threads = MAX2(1, num_threads - 1);
 	num_compiler_threads = MIN2(num_threads, ARRAY_SIZE(sscreen->tm));
 	num_compiler_threads_lowprio =
 		MIN2(num_threads, ARRAY_SIZE(sscreen->tm_low_priority));
 
 	if (!util_queue_init(&sscreen->shader_compiler_queue, "si_shader",
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index bdf146f..3131a8b 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -634,20 +634,24 @@ bool si_replace_shader(unsigned num, struct ac_shader_binary *binary);
 
 /* si_dma.c */
 void si_init_dma_functions(struct si_context *sctx);
 
 /* si_fence.c */
 void si_init_fence_functions(struct si_context *ctx);
 void si_init_screen_fence_functions(struct si_screen *screen);
 struct pipe_fence_handle *si_create_fence(struct pipe_context *ctx,
 					  struct tc_unflushed_batch_token *tc_token);
 
+/* si_get.c */
+const char *si_get_family_name(const struct si_screen *sscreen);
+void si_init_screen_get_functions(struct si_screen *sscreen);
+
 /* si_hw_context.c */
 void si_destroy_saved_cs(struct si_saved_cs *scs);
 void si_context_gfx_flush(void *context, unsigned flags,
 			  struct pipe_fence_handle **fence);
 void si_begin_new_cs(struct si_context *ctx);
 void si_need_cs_space(struct si_context *ctx);
 
 /* si_compute.c */
 void si_init_compute_functions(struct si_context *sctx);
 
-- 
2.7.4



More information about the mesa-dev mailing list