#include "radv_cs.h"
#include "util/disk_cache.h"
#include "util/strtod.h"
-#include "util/vk_util.h"
+#include "vk_util.h"
#include <xf86drm.h>
#include <amdgpu.h>
#include <amdgpu_drm.h>
#include "ac_llvm_util.h"
#include "vk_format.h"
#include "sid.h"
+#include "gfx9d.h"
#include "util/debug.h"
static int
case CHIP_POLARIS11: return "AMD RADV POLARIS11";
case CHIP_POLARIS12: return "AMD RADV POLARIS12";
case CHIP_STONEY: return "AMD RADV STONEY";
+ case CHIP_VEGA10: return "AMD RADV VEGA";
+ case CHIP_RAVEN: return "AMD RADV RAVEN";
default: return "AMD RADV unknown";
}
}
radv_get_device_uuid(drm_device, device->device_uuid);
+ if (device->rad_info.family == CHIP_STONEY ||
+ device->rad_info.chip_class >= GFX9) {
+ device->has_rbplus = true;
+ device->rbplus_allowed = device->rad_info.family == CHIP_STONEY;
+ }
+
return VK_SUCCESS;
fail:
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures* pFeatures)
{
- // RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
-
+ RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ bool is_gfx9 = pdevice->rad_info.chip_class >= GFX9;
memset(pFeatures, 0, sizeof(*pFeatures));
*pFeatures = (VkPhysicalDeviceFeatures) {
.fullDrawIndexUint32 = true,
.imageCubeArray = true,
.independentBlend = true,
- .geometryShader = true,
- .tessellationShader = true,
+ .geometryShader = !is_gfx9,
+ .tessellationShader = !is_gfx9,
.sampleRateShading = false,
.dualSrcBlend = true,
.logicOp = true,
return radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
}
-static uint32_t radv_get_driver_version()
-{
- const char *minor_string = strchr(VERSION, '.');
- const char *patch_string = minor_string ? strchr(minor_string + 1, ','): NULL;
- int major = atoi(VERSION);
- int minor = minor_string ? atoi(minor_string + 1) : 0;
- int patch = patch_string ? atoi(patch_string + 1) : 0;
- if (strstr(VERSION, "devel")) {
- if (patch == 0) {
- patch = 99;
- if (minor == 0) {
- minor = 99;
- --major;
- } else
- --minor;
- } else
- --patch;
- }
- uint32_t version = VK_MAKE_VERSION(major, minor, patch);
- return version;
-}
-
void radv_GetPhysicalDeviceProperties(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceProperties* pProperties)
*pProperties = (VkPhysicalDeviceProperties) {
.apiVersion = VK_MAKE_VERSION(1, 0, 42),
- .driverVersion = radv_get_driver_version(),
+ .driverVersion = vk_get_driver_version(),
.vendorID = 0x1002,
.deviceID = pdevice->rad_info.pci_id,
- .deviceType = VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU,
+ .deviceType = pdevice->rad_info.has_dedicated_vram ? VK_PHYSICAL_DEVICE_TYPE_DISCRETE_GPU : VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
.limits = limits,
.sparseProperties = {0},
};
case CHIP_POLARIS10:
case CHIP_POLARIS11:
case CHIP_POLARIS12:
+ case CHIP_VEGA10:
+ case CHIP_RAVEN:
device->gs_table_depth = 32;
return;
default:
case RADV_QUEUE_COMPUTE:
si_cs_emit_cache_flush(device->flush_cs[family],
device->physical_device->rad_info.chip_class,
+ NULL, 0,
family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
RADV_CMD_FLAG_INV_ICACHE |
RADV_CMD_FLAG_INV_SMEM_L1 |
case RADV_QUEUE_COMPUTE:
si_cs_emit_cache_flush(device->flush_shader_cs[family],
device->physical_device->rad_info.chip_class,
+ NULL, 0,
family == RADV_QUEUE_COMPUTE && device->physical_device->rad_info.chip_class >= CIK,
family == RADV_QUEUE_COMPUTE ? RADV_CMD_FLAG_CS_PARTIAL_FLUSH : (RADV_CMD_FLAG_CS_PARTIAL_FLUSH | RADV_CMD_FLAG_PS_PARTIAL_FLUSH) |
RADV_CMD_FLAG_INV_ICACHE |
max_offchip_buffers = MIN2(max_offchip_buffers, 126);
break;
case CIK:
- max_offchip_buffers = MIN2(max_offchip_buffers, 508);
- break;
case VI:
+ case GFX9:
default:
- max_offchip_buffers = MIN2(max_offchip_buffers, 512);
+ max_offchip_buffers = MIN2(max_offchip_buffers, 508);
break;
}
S_030938_SIZE(tess_factor_ring_size / 4));
radeon_set_uconfig_reg(cs, R_030940_VGT_TF_MEMORY_BASE,
tf_va >> 8);
+ if (queue->device->physical_device->rad_info.chip_class >= GFX9) {
+ radeon_set_uconfig_reg(cs, R_030944_VGT_TF_MEMORY_BASE_HI,
+ tf_va >> 40);
+ }
radeon_set_uconfig_reg(cs, R_03093C_VGT_HS_OFFCHIP_PARAM, hs_offchip_param);
} else {
radeon_set_config_reg(cs, R_008988_VGT_TF_RING_SIZE,
if (!i) {
si_cs_emit_cache_flush(cs,
queue->device->physical_device->rad_info.chip_class,
+ NULL, 0,
queue->queue_family_index == RING_COMPUTE &&
queue->device->physical_device->rad_info.chip_class >= CIK,
RADV_CMD_FLAG_INV_ICACHE |
const struct vk_format_description *desc;
unsigned ntype, format, swap, endian;
unsigned blend_clamp = 0, blend_bypass = 0;
- unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
uint64_t va;
const struct radeon_surf *surf = &iview->image->surface;
- const struct legacy_surf_level *level_info = &surf->u.legacy.level[iview->base_mip];
desc = vk_format_description(iview->vk_format);
memset(cb, 0, sizeof(*cb));
+ /* Intensity is implemented as Red, so treat it that way. */
+ cb->cb_color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == VK_SWIZZLE_1);
+
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
- va += level_info->offset;
+
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ struct gfx9_surf_meta_flags meta;
+ if (iview->image->dcc_offset)
+ meta = iview->image->surface.u.gfx9.dcc;
+ else
+ meta = iview->image->surface.u.gfx9.cmask;
+
+ cb->cb_color_attrib |= S_028C74_COLOR_SW_MODE(iview->image->surface.u.gfx9.surf.swizzle_mode) |
+ S_028C74_FMASK_SW_MODE(iview->image->surface.u.gfx9.fmask.swizzle_mode) |
+ S_028C74_RB_ALIGNED(meta.rb_aligned) |
+ S_028C74_PIPE_ALIGNED(meta.pipe_aligned);
+
+ va += iview->image->surface.u.gfx9.surf_offset >> 8;
+ } else {
+ const struct legacy_surf_level *level_info = &surf->u.legacy.level[iview->base_mip];
+ unsigned pitch_tile_max, slice_tile_max, tile_mode_index;
+
+ va += level_info->offset;
+
+ pitch_tile_max = level_info->nblk_x / 8 - 1;
+ slice_tile_max = (level_info->nblk_x * level_info->nblk_y) / 64 - 1;
+ tile_mode_index = si_tile_mode_index(iview->image, iview->base_mip, false);
+
+ cb->cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
+ cb->cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
+ cb->cb_color_cmask_slice = iview->image->cmask.slice_tile_max;
+
+ cb->cb_color_attrib |= S_028C74_TILE_MODE_INDEX(tile_mode_index);
+ cb->micro_tile_mode = iview->image->surface.micro_tile_mode;
+
+ if (iview->image->fmask.size) {
+ if (device->physical_device->rad_info.chip_class >= CIK)
+ cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(iview->image->fmask.pitch_in_pixels / 8 - 1);
+ cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(iview->image->fmask.tile_mode_index);
+ cb->cb_color_fmask_slice = S_028C88_TILE_MAX(iview->image->fmask.slice_tile_max);
+ } else {
+ /* This must be set for fast clear to work without FMASK. */
+ if (device->physical_device->rad_info.chip_class >= CIK)
+ cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
+ cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
+ cb->cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
+ }
+ }
+
cb->cb_color_base = va >> 8;
/* CMASK variables */
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
va += iview->image->cmask.offset;
cb->cb_color_cmask = va >> 8;
- cb->cb_color_cmask_slice = iview->image->cmask.slice_tile_max;
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
va += iview->image->dcc_offset;
cb->cb_color_view = S_028C6C_SLICE_START(iview->base_layer) |
S_028C6C_SLICE_MAX(iview->base_layer + max_slice - 1);
- cb->micro_tile_mode = iview->image->surface.micro_tile_mode;
- pitch_tile_max = level_info->nblk_x / 8 - 1;
- slice_tile_max = (level_info->nblk_x * level_info->nblk_y) / 64 - 1;
- tile_mode_index = si_tile_mode_index(iview->image, iview->base_mip, false);
-
- cb->cb_color_pitch = S_028C64_TILE_MAX(pitch_tile_max);
- cb->cb_color_slice = S_028C68_TILE_MAX(slice_tile_max);
-
- /* Intensity is implemented as Red, so treat it that way. */
- cb->cb_color_attrib = S_028C74_FORCE_DST_ALPHA_1(desc->swizzle[3] == VK_SWIZZLE_1) |
- S_028C74_TILE_MODE_INDEX(tile_mode_index);
-
if (iview->image->info.samples > 1) {
unsigned log_samples = util_logbase2(iview->image->info.samples);
if (iview->image->fmask.size) {
va = device->ws->buffer_get_va(iview->bo) + iview->image->offset + iview->image->fmask.offset;
- if (device->physical_device->rad_info.chip_class >= CIK)
- cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(iview->image->fmask.pitch_in_pixels / 8 - 1);
- cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(iview->image->fmask.tile_mode_index);
cb->cb_color_fmask = va >> 8;
- cb->cb_color_fmask_slice = S_028C88_TILE_MAX(iview->image->fmask.slice_tile_max);
} else {
- /* This must be set for fast clear to work without FMASK. */
- if (device->physical_device->rad_info.chip_class >= CIK)
- cb->cb_color_pitch |= S_028C64_FMASK_TILE_MAX(pitch_tile_max);
- cb->cb_color_attrib |= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index);
cb->cb_color_fmask = cb->cb_color_base;
- cb->cb_color_fmask_slice = S_028C88_TILE_MAX(slice_tile_max);
}
ntype = radv_translate_color_numformat(iview->vk_format,
unsigned bankh = util_logbase2(iview->image->surface.u.legacy.bankh);
cb->cb_color_attrib |= S_028C74_FMASK_BANK_HEIGHT(bankh);
}
+
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ uint32_t max_slice = radv_surface_layer_count(iview);
+ unsigned mip0_depth = iview->base_layer + max_slice - 1;
+
+ cb->cb_color_view |= S_028C6C_MIP_LEVEL(iview->base_mip);
+ cb->cb_color_attrib |= S_028C74_MIP0_DEPTH(mip0_depth) |
+ S_028C74_RESOURCE_TYPE(iview->image->surface.u.gfx9.resource_type);
+ cb->cb_color_attrib2 = S_028C68_MIP0_WIDTH(iview->image->info.width - 1) |
+ S_028C68_MIP0_HEIGHT(iview->image->info.height - 1) |
+ S_028C68_MAX_MIP(iview->image->info.levels);
+
+ cb->gfx9_epitch = S_0287A0_EPITCH(iview->image->surface.u.gfx9.surf.epitch);
+
+ }
}
static void
struct radv_image_view *iview)
{
unsigned level = iview->base_mip;
- unsigned format;
+ unsigned format, stencil_format;
uint64_t va, s_offs, z_offs;
- const struct legacy_surf_level *level_info = &iview->image->surface.u.legacy.level[level];
bool stencil_only = false;
memset(ds, 0, sizeof(*ds));
switch (iview->vk_format) {
break;
case VK_FORMAT_S8_UINT:
stencil_only = true;
- level_info = &iview->image->surface.u.legacy.stencil_level[level];
break;
default:
break;
}
format = radv_translate_dbformat(iview->vk_format);
-
- va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
- s_offs = z_offs = va;
- z_offs += iview->image->surface.u.legacy.level[level].offset;
- s_offs += iview->image->surface.u.legacy.stencil_level[level].offset;
+ stencil_format = iview->image->surface.flags & RADEON_SURF_SBUFFER ?
+ V_028044_STENCIL_8 : V_028044_STENCIL_INVALID;
uint32_t max_slice = radv_surface_layer_count(iview);
ds->db_depth_view = S_028008_SLICE_START(iview->base_layer) |
S_028008_SLICE_MAX(iview->base_layer + max_slice - 1);
- ds->db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
- ds->db_z_info = S_028040_FORMAT(format) | S_028040_ZRANGE_PRECISION(1);
- if (iview->image->info.samples > 1)
- ds->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(iview->image->info.samples));
+ ds->db_htile_data_base = 0;
+ ds->db_htile_surface = 0;
- if (iview->image->surface.flags & RADEON_SURF_SBUFFER)
- ds->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_8);
- else
- ds->db_stencil_info = S_028044_FORMAT(V_028044_STENCIL_INVALID);
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset;
+ s_offs = z_offs = va;
- if (device->physical_device->rad_info.chip_class >= CIK) {
- struct radeon_info *info = &device->physical_device->rad_info;
- unsigned tiling_index = iview->image->surface.u.legacy.tiling_index[level];
- unsigned stencil_index = iview->image->surface.u.legacy.stencil_tiling_index[level];
- unsigned macro_index = iview->image->surface.u.legacy.macro_tile_index;
- unsigned tile_mode = info->si_tile_mode_array[tiling_index];
- unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
- unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
+ if (device->physical_device->rad_info.chip_class >= GFX9) {
+ assert(iview->image->surface.u.gfx9.surf_offset == 0);
+ s_offs += iview->image->surface.u.gfx9.stencil_offset;
+
+ ds->db_z_info = S_028038_FORMAT(format) |
+ S_028038_NUM_SAMPLES(util_logbase2(iview->image->info.samples)) |
+ S_028038_SW_MODE(iview->image->surface.u.gfx9.surf.swizzle_mode) |
+ S_028038_MAXMIP(iview->image->info.levels - 1);
+ ds->db_stencil_info = S_02803C_FORMAT(stencil_format) |
+ S_02803C_SW_MODE(iview->image->surface.u.gfx9.stencil.swizzle_mode);
+
+ ds->db_z_info2 = S_028068_EPITCH(iview->image->surface.u.gfx9.surf.epitch);
+ ds->db_stencil_info2 = S_02806C_EPITCH(iview->image->surface.u.gfx9.stencil.epitch);
+ ds->db_depth_view |= S_028008_MIPID(level);
+
+ ds->db_depth_size = S_02801C_X_MAX(iview->image->info.width - 1) |
+ S_02801C_Y_MAX(iview->image->info.height - 1);
+
+ /* Only use HTILE for the first level. */
+ if (iview->image->surface.htile_size && !level) {
+ ds->db_z_info |= S_028038_TILE_SURFACE_ENABLE(1);
+
+ if (!(iview->image->surface.flags & RADEON_SURF_SBUFFER))
+ /* Use all of the htile_buffer for depth if there's no stencil. */
+ ds->db_stencil_info |= S_02803C_TILE_STENCIL_DISABLE(1);
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
+ iview->image->htile_offset;
+ ds->db_htile_data_base = va >> 8;
+ ds->db_htile_surface = S_028ABC_FULL_CACHE(1) |
+ S_028ABC_PIPE_ALIGNED(iview->image->surface.u.gfx9.htile.pipe_aligned) |
+ S_028ABC_RB_ALIGNED(iview->image->surface.u.gfx9.htile.rb_aligned);
+ }
+ } else {
+ const struct legacy_surf_level *level_info = &iview->image->surface.u.legacy.level[level];
if (stencil_only)
- tile_mode = stencil_tile_mode;
-
- ds->db_depth_info |=
- S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
- S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
- S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
- S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
- S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
- S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
- ds->db_z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
- ds->db_stencil_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
- } else {
- unsigned tile_mode_index = si_tile_mode_index(iview->image, level, false);
- ds->db_z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
- tile_mode_index = si_tile_mode_index(iview->image, level, true);
- ds->db_stencil_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
- }
+ level_info = &iview->image->surface.u.legacy.stencil_level[level];
+
+ z_offs += iview->image->surface.u.legacy.level[level].offset;
+ s_offs += iview->image->surface.u.legacy.stencil_level[level].offset;
+
+ ds->db_depth_info = S_02803C_ADDR5_SWIZZLE_MASK(1);
+ ds->db_z_info = S_028040_FORMAT(format) | S_028040_ZRANGE_PRECISION(1);
+ ds->db_stencil_info = S_028044_FORMAT(stencil_format);
+
+ if (iview->image->info.samples > 1)
+ ds->db_z_info |= S_028040_NUM_SAMPLES(util_logbase2(iview->image->info.samples));
+
+ if (device->physical_device->rad_info.chip_class >= CIK) {
+ struct radeon_info *info = &device->physical_device->rad_info;
+ unsigned tiling_index = iview->image->surface.u.legacy.tiling_index[level];
+ unsigned stencil_index = iview->image->surface.u.legacy.stencil_tiling_index[level];
+ unsigned macro_index = iview->image->surface.u.legacy.macro_tile_index;
+ unsigned tile_mode = info->si_tile_mode_array[tiling_index];
+ unsigned stencil_tile_mode = info->si_tile_mode_array[stencil_index];
+ unsigned macro_mode = info->cik_macrotile_mode_array[macro_index];
+
+ if (stencil_only)
+ tile_mode = stencil_tile_mode;
+
+ ds->db_depth_info |=
+ S_02803C_ARRAY_MODE(G_009910_ARRAY_MODE(tile_mode)) |
+ S_02803C_PIPE_CONFIG(G_009910_PIPE_CONFIG(tile_mode)) |
+ S_02803C_BANK_WIDTH(G_009990_BANK_WIDTH(macro_mode)) |
+ S_02803C_BANK_HEIGHT(G_009990_BANK_HEIGHT(macro_mode)) |
+ S_02803C_MACRO_TILE_ASPECT(G_009990_MACRO_TILE_ASPECT(macro_mode)) |
+ S_02803C_NUM_BANKS(G_009990_NUM_BANKS(macro_mode));
+ ds->db_z_info |= S_028040_TILE_SPLIT(G_009910_TILE_SPLIT(tile_mode));
+ ds->db_stencil_info |= S_028044_TILE_SPLIT(G_009910_TILE_SPLIT(stencil_tile_mode));
+ } else {
+ unsigned tile_mode_index = si_tile_mode_index(iview->image, level, false);
+ ds->db_z_info |= S_028040_TILE_MODE_INDEX(tile_mode_index);
+ tile_mode_index = si_tile_mode_index(iview->image, level, true);
+ ds->db_stencil_info |= S_028044_TILE_MODE_INDEX(tile_mode_index);
+ }
- if (iview->image->surface.htile_size && !level) {
- ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1);
+ ds->db_depth_size = S_028058_PITCH_TILE_MAX((level_info->nblk_x / 8) - 1) |
+ S_028058_HEIGHT_TILE_MAX((level_info->nblk_y / 8) - 1);
+ ds->db_depth_slice = S_02805C_SLICE_TILE_MAX((level_info->nblk_x * level_info->nblk_y) / 64 - 1);
- if (!(iview->image->surface.flags & RADEON_SURF_SBUFFER))
- /* Use all of the htile_buffer for depth if there's no stencil. */
- ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
+ if (iview->image->surface.htile_size && !level) {
+ ds->db_z_info |= S_028040_TILE_SURFACE_ENABLE(1);
- va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
- iview->image->htile_offset;
- ds->db_htile_data_base = va >> 8;
- ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
- } else {
- ds->db_htile_data_base = 0;
- ds->db_htile_surface = 0;
+ if (!(iview->image->surface.flags & RADEON_SURF_SBUFFER))
+ /* Use all of the htile_buffer for depth if there's no stencil. */
+ ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
+
+ va = device->ws->buffer_get_va(iview->bo) + iview->image->offset +
+ iview->image->htile_offset;
+ ds->db_htile_data_base = va >> 8;
+ ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
+ }
}
ds->db_z_read_base = ds->db_z_write_base = z_offs >> 8;
ds->db_stencil_read_base = ds->db_stencil_write_base = s_offs >> 8;
-
- ds->db_depth_size = S_028058_PITCH_TILE_MAX((level_info->nblk_x / 8) - 1) |
- S_028058_HEIGHT_TILE_MAX((level_info->nblk_y / 8) - 1);
- ds->db_depth_slice = S_02805C_SLICE_TILE_MAX((level_info->nblk_x * level_info->nblk_y) / 64 - 1);
}
VkResult radv_CreateFramebuffer(
projects / mesa.git / blobdiff