[Mesa-dev] [PATCH 21/24] radv: use ac_gpu_info
Nicolai Hähnle
nhaehnle at gmail.com
Sun May 14 21:47:07 UTC 2017
From: Nicolai Hähnle <nicolai.haehnle at amd.com>
---
src/amd/vulkan/radv_device.c | 10 +-
src/amd/vulkan/radv_private.h | 1 +
src/amd/vulkan/radv_radeon_winsys.h | 55 +------
src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c | 175 +---------------------
src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h | 1 +
5 files changed, 13 insertions(+), 229 deletions(-)
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index b4f2331..102b9fd 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -731,21 +731,21 @@ void radv_GetPhysicalDeviceProperties2KHR(
}
}
static void radv_get_physical_device_queue_family_properties(
struct radv_physical_device* pdevice,
uint32_t* pCount,
VkQueueFamilyProperties** pQueueFamilyProperties)
{
int num_queue_families = 1;
int idx;
- if (pdevice->rad_info.compute_rings > 0 &&
+ if (pdevice->rad_info.num_compute_rings > 0 &&
pdevice->rad_info.chip_class >= CIK &&
!(pdevice->instance->debug_flags & RADV_DEBUG_NO_COMPUTE_QUEUE))
num_queue_families++;
if (pQueueFamilyProperties == NULL) {
*pCount = num_queue_families;
return;
}
if (!*pCount)
@@ -758,29 +758,29 @@ static void radv_get_physical_device_queue_family_properties(
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT |
VK_QUEUE_SPARSE_BINDING_BIT,
.queueCount = 1,
.timestampValidBits = 64,
.minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
};
idx++;
}
- if (pdevice->rad_info.compute_rings > 0 &&
+ if (pdevice->rad_info.num_compute_rings > 0 &&
pdevice->rad_info.chip_class >= CIK &&
!(pdevice->instance->debug_flags & RADV_DEBUG_NO_COMPUTE_QUEUE)) {
if (*pCount > idx) {
*pQueueFamilyProperties[idx] = (VkQueueFamilyProperties) {
.queueFlags = VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT |
VK_QUEUE_SPARSE_BINDING_BIT,
- .queueCount = pdevice->rad_info.compute_rings,
+ .queueCount = pdevice->rad_info.num_compute_rings,
.timestampValidBits = 64,
.minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
};
idx++;
}
}
*pCount = idx;
}
void radv_GetPhysicalDeviceQueueFamilyProperties(
@@ -850,25 +850,25 @@ void radv_GetPhysicalDeviceMemoryProperties(
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
.heapIndex = RADV_MEM_HEAP_GTT,
};
STATIC_ASSERT(RADV_MEM_HEAP_COUNT <= VK_MAX_MEMORY_HEAPS);
pMemoryProperties->memoryHeapCount = RADV_MEM_HEAP_COUNT;
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM] = (VkMemoryHeap) {
.size = physical_device->rad_info.vram_size -
- physical_device->rad_info.visible_vram_size,
+ physical_device->rad_info.vram_vis_size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
};
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_VRAM_CPU_ACCESS] = (VkMemoryHeap) {
- .size = physical_device->rad_info.visible_vram_size,
+ .size = physical_device->rad_info.vram_vis_size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
};
pMemoryProperties->memoryHeaps[RADV_MEM_HEAP_GTT] = (VkMemoryHeap) {
.size = physical_device->rad_info.gart_size,
.flags = 0,
};
}
void radv_GetPhysicalDeviceMemoryProperties2KHR(
VkPhysicalDevice physicalDevice,
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 518ed6c..302fd47 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -46,20 +46,21 @@
#include <amdgpu.h>
#include "compiler/shader_enums.h"
#include "util/macros.h"
#include "util/list.h"
#include "util/vk_alloc.h"
#include "main/macros.h"
#include "radv_radeon_winsys.h"
#include "ac_binary.h"
#include "ac_nir_to_llvm.h"
+#include "ac_gpu_info.h"
#include "ac_surface.h"
#include "radv_debug.h"
#include "radv_descriptor_set.h"
#include <llvm-c/TargetMachine.h>
/* Pre-declarations needed for WSI entrypoints */
struct wl_surface;
struct wl_display;
typedef struct xcb_connection_t xcb_connection_t;
diff --git a/src/amd/vulkan/radv_radeon_winsys.h b/src/amd/vulkan/radv_radeon_winsys.h
index 855a2d7..397f28e 100644
--- a/src/amd/vulkan/radv_radeon_winsys.h
+++ b/src/amd/vulkan/radv_radeon_winsys.h
@@ -28,20 +28,21 @@
#ifndef RADV_RADEON_WINSYS_H
#define RADV_RADEON_WINSYS_H
#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
#include "main/macros.h"
#include "amd_family.h"
+struct radeon_info;
struct ac_surf_info;
struct radeon_surf;
#define FREE(x) free(x)
enum radeon_bo_domain { /* bitfield */
RADEON_DOMAIN_GTT = 2,
RADEON_DOMAIN_VRAM = 4,
RADEON_DOMAIN_VRAM_GTT = RADEON_DOMAIN_VRAM | RADEON_DOMAIN_GTT
};
@@ -67,74 +68,20 @@ enum ring_type {
RING_VCE,
RING_LAST,
};
struct radeon_winsys_cs {
unsigned cdw; /* Number of used dwords. */
unsigned max_dw; /* Maximum number of dwords. */
uint32_t *buf; /* The base pointer of the chunk. */
};
-struct radeon_info {
- /* PCI info: domain:bus:dev:func */
- uint32_t pci_domain;
- uint32_t pci_bus;
- uint32_t pci_dev;
- uint32_t pci_func;
-
- /* Device info. */
- uint32_t pci_id;
- enum radeon_family family;
- enum chip_class chip_class;
- uint32_t gart_page_size;
- uint64_t gart_size;
- uint64_t vram_size;
- uint64_t visible_vram_size;
- bool has_dedicated_vram;
- bool has_virtual_memory;
- bool gfx_ib_pad_with_type2;
- bool has_uvd;
- uint32_t sdma_rings;
- uint32_t compute_rings;
- uint32_t vce_fw_version;
- uint32_t vce_harvest_config;
- uint32_t clock_crystal_freq; /* in kHz */
-
- /* Kernel info. */
- uint32_t drm_major; /* version */
- uint32_t drm_minor;
- uint32_t drm_patchlevel;
- bool has_userptr;
-
- /* Shader cores. */
- uint32_t r600_max_quad_pipes; /* wave size / 16 */
- uint32_t max_shader_clock;
- uint32_t num_good_compute_units;
- uint32_t max_se; /* shader engines */
- uint32_t max_sh_per_se; /* shader arrays per shader engine */
-
- /* Render backends (color + depth blocks). */
- uint32_t r300_num_gb_pipes;
- uint32_t r300_num_z_pipes;
- uint32_t r600_gb_backend_map; /* R600 harvest config */
- bool r600_gb_backend_map_valid;
- uint32_t r600_num_banks;
- uint32_t num_render_backends;
- uint32_t num_tile_pipes; /* pipe count from PIPE_CONFIG */
- uint32_t pipe_interleave_bytes;
- uint32_t enabled_rb_mask; /* GCN harvest config */
-
- /* Tile modes. */
- uint32_t si_tile_mode_array[32];
- uint32_t cik_macrotile_mode_array[16];
-};
-
#define RADEON_SURF_TYPE_MASK 0xFF
#define RADEON_SURF_TYPE_SHIFT 0
#define RADEON_SURF_TYPE_1D 0
#define RADEON_SURF_TYPE_2D 1
#define RADEON_SURF_TYPE_3D 2
#define RADEON_SURF_TYPE_CUBEMAP 3
#define RADEON_SURF_TYPE_1D_ARRAY 4
#define RADEON_SURF_TYPE_2D_ARRAY 5
#define RADEON_SURF_MODE_MASK 0xFF
#define RADEON_SURF_MODE_SHIFT 8
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
index d144d03..2ebb856 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.c
@@ -32,153 +32,28 @@
#include "xf86drm.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <amdgpu_drm.h>
#include <assert.h>
#include "radv_amdgpu_cs.h"
#include "radv_amdgpu_bo.h"
#include "radv_amdgpu_surface.h"
-#define CIK_TILE_MODE_COLOR_2D 14
-
-#define CIK__GB_TILE_MODE__PIPE_CONFIG(x) (((x) >> 6) & 0x1f)
-#define CIK__PIPE_CONFIG__ADDR_SURF_P2 0
-#define CIK__PIPE_CONFIG__ADDR_SURF_P4_8x16 4
-#define CIK__PIPE_CONFIG__ADDR_SURF_P4_16x16 5
-#define CIK__PIPE_CONFIG__ADDR_SURF_P4_16x32 6
-#define CIK__PIPE_CONFIG__ADDR_SURF_P4_32x32 7
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_16x16_8x16 8
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_16x32_8x16 9
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_8x16 10
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_16x32_16x16 11
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_16x16 12
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_16x32 13
-#define CIK__PIPE_CONFIG__ADDR_SURF_P8_32x64_32x32 14
-#define CIK__PIPE_CONFIG__ADDR_SURF_P16_32X32_8X16 16
-#define CIK__PIPE_CONFIG__ADDR_SURF_P16_32X32_16X16 17
-
-static unsigned radv_cik_get_num_tile_pipes(struct amdgpu_gpu_info *info)
-{
- unsigned mode2d = info->gb_tile_mode[CIK_TILE_MODE_COLOR_2D];
-
- switch (CIK__GB_TILE_MODE__PIPE_CONFIG(mode2d)) {
- case CIK__PIPE_CONFIG__ADDR_SURF_P2:
- return 2;
- case CIK__PIPE_CONFIG__ADDR_SURF_P4_8x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P4_16x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P4_16x32:
- case CIK__PIPE_CONFIG__ADDR_SURF_P4_32x32:
- return 4;
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_16x16_8x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_16x32_8x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_8x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_16x32_16x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_16x16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_32x32_16x32:
- case CIK__PIPE_CONFIG__ADDR_SURF_P8_32x64_32x32:
- return 8;
- case CIK__PIPE_CONFIG__ADDR_SURF_P16_32X32_8X16:
- case CIK__PIPE_CONFIG__ADDR_SURF_P16_32X32_16X16:
- return 16;
- default:
- fprintf(stderr, "Invalid CIK pipe configuration, assuming P2\n");
- assert(!"this should never occur");
- return 2;
- }
-}
-
static bool
do_winsys_init(struct radv_amdgpu_winsys *ws, int fd)
{
- struct amdgpu_buffer_size_alignments alignment_info = {};
- struct amdgpu_heap_info vram, visible_vram, gtt;
- struct drm_amdgpu_info_hw_ip dma = {};
- struct drm_amdgpu_info_hw_ip compute = {};
- drmDevicePtr devinfo;
- int r;
- int i, j;
- /* Get PCI info. */
- r = drmGetDevice2(fd, 0, &devinfo);
- if (r) {
- fprintf(stderr, "amdgpu: drmGetDevice2 failed.\n");
- goto fail;
- }
- ws->info.pci_domain = devinfo->businfo.pci->domain;
- ws->info.pci_bus = devinfo->businfo.pci->bus;
- ws->info.pci_dev = devinfo->businfo.pci->dev;
- ws->info.pci_func = devinfo->businfo.pci->func;
- drmFreeDevice(&devinfo);
-
- /* Query hardware and driver information. */
- r = amdgpu_query_gpu_info(ws->dev, &ws->amdinfo);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_gpu_info failed.\n");
+ if (!ac_query_gpu_info(fd, ws->dev, &ws->info, &ws->amdinfo))
goto fail;
- }
-
- r = amdgpu_query_buffer_size_alignment(ws->dev, &alignment_info);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_buffer_size_alignment failed.\n");
- goto fail;
- }
-
- r = amdgpu_query_heap_info(ws->dev, AMDGPU_GEM_DOMAIN_VRAM, 0, &vram);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_heap_info(vram) failed.\n");
- goto fail;
- }
-
- r = amdgpu_query_heap_info(ws->dev, AMDGPU_GEM_DOMAIN_VRAM,
- AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED, &visible_vram);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_heap_info(visible_vram) failed.\n");
- goto fail;
- }
- r = amdgpu_query_heap_info(ws->dev, AMDGPU_GEM_DOMAIN_GTT, 0, >t);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_heap_info(gtt) failed.\n");
- goto fail;
- }
-
- r = amdgpu_query_hw_ip_info(ws->dev, AMDGPU_HW_IP_DMA, 0, &dma);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_hw_ip_info(dma) failed.\n");
- goto fail;
- }
-
- r = amdgpu_query_hw_ip_info(ws->dev, AMDGPU_HW_IP_COMPUTE, 0, &compute);
- if (r) {
- fprintf(stderr, "amdgpu: amdgpu_query_hw_ip_info(compute) failed.\n");
- goto fail;
- }
- ws->info.pci_id = ws->amdinfo.asic_id; /* TODO: is this correct? */
- ws->info.vce_harvest_config = ws->amdinfo.vce_harvest_config;
-
- switch (ws->info.pci_id) {
-#define CHIPSET(pci_id, name, cfamily) case pci_id: ws->info.family = CHIP_##cfamily; break;
-#include "pci_ids/radeonsi_pci_ids.h"
-#undef CHIPSET
- default:
- fprintf(stderr, "amdgpu: Invalid PCI ID.\n");
- goto fail;
- }
-
- if (ws->info.family >= CHIP_TONGA)
- ws->info.chip_class = VI;
- else if (ws->info.family >= CHIP_BONAIRE)
- ws->info.chip_class = CIK;
- else if (ws->info.family >= CHIP_TAHITI)
- ws->info.chip_class = SI;
- else {
- fprintf(stderr, "amdgpu: Unknown family.\n");
+ if (ws->info.chip_class >= GFX9) {
+ fprintf(stderr, "radv: GFX9 is not supported.\n");
goto fail;
}
/* family and rev_id are for addrlib */
switch (ws->info.family) {
case CHIP_TAHITI:
ws->family = FAMILY_SI;
ws->rev_id = SI_TAHITI_P_A0;
break;
case CHIP_PITCAIRN:
@@ -253,62 +128,22 @@ do_winsys_init(struct radv_amdgpu_winsys *ws, int fd)
fprintf(stderr, "amdgpu: Unknown family.\n");
goto fail;
}
ws->addrlib = radv_amdgpu_addr_create(&ws->amdinfo, ws->family, ws->rev_id, ws->info.chip_class);
if (!ws->addrlib) {
fprintf(stderr, "amdgpu: Cannot create addrlib.\n");
goto fail;
}
- assert(util_is_power_of_two(dma.available_rings + 1));
- assert(util_is_power_of_two(compute.available_rings + 1));
-
- /* Set hardware information. */
- ws->info.gart_size = gtt.heap_size;
- ws->info.vram_size = vram.heap_size;
- ws->info.visible_vram_size = visible_vram.heap_size;
- /* convert the shader clock from KHz to MHz */
- ws->info.max_shader_clock = ws->amdinfo.max_engine_clk / 1000;
- ws->info.max_se = ws->amdinfo.num_shader_engines;
- ws->info.max_sh_per_se = ws->amdinfo.num_shader_arrays_per_engine;
- ws->info.has_uvd = 0;
- ws->info.vce_fw_version = 0;
- ws->info.has_userptr = TRUE;
- ws->info.num_render_backends = ws->amdinfo.rb_pipes;
- ws->info.clock_crystal_freq = ws->amdinfo.gpu_counter_freq;
- ws->info.num_tile_pipes = radv_cik_get_num_tile_pipes(&ws->amdinfo);
- ws->info.pipe_interleave_bytes = 256 << ((ws->amdinfo.gb_addr_cfg >> 4) & 0x7);
- ws->info.has_virtual_memory = TRUE;
- ws->info.sdma_rings = MIN2(util_bitcount(dma.available_rings),
- MAX_RINGS_PER_TYPE);
- ws->info.compute_rings = MIN2(util_bitcount(compute.available_rings),
- MAX_RINGS_PER_TYPE);
-
- /* Get the number of good compute units. */
- ws->info.num_good_compute_units = 0;
- for (i = 0; i < ws->info.max_se; i++)
- for (j = 0; j < ws->info.max_sh_per_se; j++)
- ws->info.num_good_compute_units +=
- util_bitcount(ws->amdinfo.cu_bitmap[i][j]);
-
- memcpy(ws->info.si_tile_mode_array, ws->amdinfo.gb_tile_mode,
- sizeof(ws->amdinfo.gb_tile_mode));
- ws->info.enabled_rb_mask = ws->amdinfo.enabled_rb_pipes_mask;
-
- memcpy(ws->info.cik_macrotile_mode_array, ws->amdinfo.gb_macro_tile_mode,
- sizeof(ws->amdinfo.gb_macro_tile_mode));
-
- ws->info.gart_page_size = alignment_info.size_remote;
-
- if (ws->info.chip_class == SI)
- ws->info.gfx_ib_pad_with_type2 = TRUE;
+ ws->info.num_sdma_rings = MIN2(ws->info.num_sdma_rings, MAX_RINGS_PER_TYPE);
+ ws->info.num_compute_rings = MIN2(ws->info.num_compute_rings, MAX_RINGS_PER_TYPE);
ws->use_ib_bos = ws->family >= FAMILY_CI;
return true;
fail:
return false;
}
static void radv_amdgpu_winsys_query_info(struct radeon_winsys *rws,
struct radeon_info *info)
{
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
index abb238b..93f4c2c 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_winsys.h
@@ -22,20 +22,21 @@
* THE AUTHORS OR COPYRIGHT HOLDERS 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.
*/
#ifndef RADV_AMDGPU_WINSYS_H
#define RADV_AMDGPU_WINSYS_H
#include "radv_radeon_winsys.h"
+#include "ac_gpu_info.h"
#include "addrlib/addrinterface.h"
#include <amdgpu.h>
#include "util/list.h"
struct radv_amdgpu_winsys {
struct radeon_winsys base;
amdgpu_device_handle dev;
struct radeon_info info;
struct amdgpu_gpu_info amdinfo;
--
2.9.3
More information about the mesa-dev
mailing list