OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

tu: Support VK_EXT_fragment_density_map on a750

Browse Source 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29938>
This commit is contained in:
Connor Abbott
2024-06-27 06:49:03 -04:00
committed by Marge Bot
parent b0599a7fe2
commit 2988f43420
3 changed files with 109 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

130
src/freedreno/vulkan/tu_cmd_buffer.cc
View File

@@ -1090,9 +1090,14 @@ tu6_emit_tile_select(struct tu_cmd_buffer *cmd,
}
/* Make the CP wait until the CP_MEM_WRITE's to the command buffers
* land.
* land. When loading FS params via UBOs, we also need to invalidate
* UCHE because the FS param patchpoint is read through UCHE.
*/
tu_cs_emit_pkt7(cs, CP_WAIT_MEM_WRITES, 0);
if (cmd->device->compiler->load_shader_consts_via_preamble) {
tu_emit_event_write<CHIP>(cmd, cs, FD_CACHE_INVALIDATE);
tu_cs_emit_wfi(cs);
}
tu_cs_emit_pkt7(cs, CP_WAIT_FOR_ME, 0);
}
}
@@ -5117,6 +5122,38 @@ fdm_apply_fs_params(struct tu_cmd_buffer *cmd,
}
}
static void
tu_emit_fdm_params(struct tu_cmd_buffer *cmd,
struct tu_cs *cs, struct tu_shader *fs,
unsigned num_units)
{
STATIC_ASSERT(IR3_DP_FS_FRAG_INVOCATION_COUNT == IR3_DP_FS_DYNAMIC);
tu_cs_emit(cs, fs->fs.per_samp ?
cmd->vk.dynamic_graphics_state.ms.rasterization_samples : 1);
tu_cs_emit(cs, 0);
tu_cs_emit(cs, 0);
tu_cs_emit(cs, 0);
STATIC_ASSERT(IR3_DP_FS_FRAG_SIZE == IR3_DP_FS_DYNAMIC + 4);
STATIC_ASSERT(IR3_DP_FS_FRAG_OFFSET == IR3_DP_FS_DYNAMIC + 6);
if (num_units > 1) {
if (fs->fs.has_fdm) {
struct apply_fs_params_state state = {
.num_consts = num_units - 1,
};
tu_create_fdm_bin_patchpoint(cmd, cs, 4 * (num_units - 1),
fdm_apply_fs_params, state);
} else {
for (unsigned i = 1; i < num_units; i++) {
tu_cs_emit(cs, 1);
tu_cs_emit(cs, 1);
tu_cs_emit(cs, fui(0.0f));
tu_cs_emit(cs, fui(0.0f));
}
}
}
}
static void
tu6_emit_fs_params(struct tu_cmd_buffer *cmd)
{
@@ -5151,31 +5188,7 @@ tu6_emit_fs_params(struct tu_cmd_buffer *cmd)
tu_cs_emit(&cs, 0);
tu_cs_emit(&cs, 0);
STATIC_ASSERT(IR3_DP_FS_FRAG_INVOCATION_COUNT == IR3_DP_FS_DYNAMIC);
tu_cs_emit(&cs, fs->fs.per_samp ?
cmd->vk.dynamic_graphics_state.ms.rasterization_samples : 1);
tu_cs_emit(&cs, 0);
tu_cs_emit(&cs, 0);
tu_cs_emit(&cs, 0);
STATIC_ASSERT(IR3_DP_FS_FRAG_SIZE == IR3_DP_FS_DYNAMIC + 4);
STATIC_ASSERT(IR3_DP_FS_FRAG_OFFSET == IR3_DP_FS_DYNAMIC + 6);
if (num_units > 1) {
if (fs->fs.has_fdm) {
struct apply_fs_params_state state = {
.num_consts = num_units - 1,
};
tu_create_fdm_bin_patchpoint(cmd, &cs, 4 * (num_units - 1),
fdm_apply_fs_params, state);
} else {
for (unsigned i = 1; i < num_units; i++) {
tu_cs_emit(&cs, 1);
tu_cs_emit(&cs, 1);
tu_cs_emit(&cs, fui(0.0f));
tu_cs_emit(&cs, fui(0.0f));
}
}
}
tu_emit_fdm_params(cmd, &cs, fs, num_units);
cmd->state.fs_params = tu_cs_end_draw_state(&cmd->sub_cs, &cs);
@@ -5183,6 +5196,69 @@ tu6_emit_fs_params(struct tu_cmd_buffer *cmd)
tu_cs_set_writeable(&cmd->sub_cs, false);
}
static void
tu7_emit_fs_params(struct tu_cmd_buffer *cmd)
{
struct tu_shader *fs = cmd->state.shaders[MESA_SHADER_FRAGMENT];
int ubo_offset = fs->const_state.fdm_ubo.idx;
if (ubo_offset < 0) {
cmd->state.fs_params = (struct tu_draw_state) {};
return;
}
unsigned num_units = DIV_ROUND_UP(fs->const_state.fdm_ubo.size, 4);
if (fs->fs.has_fdm)
tu_cs_set_writeable(&cmd->sub_cs, true);
struct tu_cs cs;
VkResult result =
tu_cs_begin_sub_stream_aligned(&cmd->sub_cs, num_units, 4, &cs);
if (result != VK_SUCCESS) {
tu_cs_set_writeable(&cmd->sub_cs, false);
vk_command_buffer_set_error(&cmd->vk, result);
return;
}
tu_emit_fdm_params(cmd, &cs, fs, num_units);
struct tu_draw_state fdm_ubo = tu_cs_end_draw_state(&cmd->sub_cs, &cs);
if (fs->fs.has_fdm)
tu_cs_set_writeable(&cmd->sub_cs, false);
result = tu_cs_begin_sub_stream(&cmd->sub_cs, 6, &cs);
if (result != VK_SUCCESS) {
vk_command_buffer_set_error(&cmd->vk, result);
return;
}
tu_cs_emit_pkt7(&cs, CP_LOAD_STATE6_FRAG, 5);
tu_cs_emit(&cs,
CP_LOAD_STATE6_0_DST_OFF(ubo_offset) |
CP_LOAD_STATE6_0_STATE_TYPE(ST6_UBO)|
CP_LOAD_STATE6_0_STATE_SRC(SS6_DIRECT) |
CP_LOAD_STATE6_0_STATE_BLOCK(SB6_FS_SHADER) |
CP_LOAD_STATE6_0_NUM_UNIT(1));
tu_cs_emit(&cs, CP_LOAD_STATE6_1_EXT_SRC_ADDR(0));
tu_cs_emit(&cs, CP_LOAD_STATE6_2_EXT_SRC_ADDR_HI(0));
tu_cs_emit_qw(&cs,
fdm_ubo.iova |
(uint64_t)A6XX_UBO_1_SIZE(num_units) << 32);
cmd->state.fs_params = tu_cs_end_draw_state(&cmd->sub_cs, &cs);
}
static void
tu_emit_fs_params(struct tu_cmd_buffer *cmd)
{
if (cmd->device->compiler->load_shader_consts_via_preamble)
tu7_emit_fs_params(cmd);
else
tu6_emit_fs_params(cmd);
}
template <chip CHIP>
static VkResult
tu6_draw_common(struct tu_cmd_buffer *cmd,
@@ -5352,7 +5428,7 @@ tu6_draw_common(struct tu_cmd_buffer *cmd,
if (BITSET_TEST(cmd->vk.dynamic_graphics_state.dirty,
MESA_VK_DYNAMIC_MS_RASTERIZATION_SAMPLES) ||
(cmd->state.dirty & (TU_CMD_DIRTY_PROGRAM | TU_CMD_DIRTY_FDM))) {
tu6_emit_fs_params(cmd);
tu_emit_fs_params(cmd);
dirty_fs_params = true;
}

2
src/freedreno/vulkan/tu_device.cc
View File

@@ -242,7 +242,7 @@ get_device_extensions(const struct tu_physical_device *device,
.EXT_extended_dynamic_state3 = true,
.EXT_external_memory_dma_buf = true,
.EXT_filter_cubic = device->info->a6xx.has_tex_filter_cubic,
.EXT_fragment_density_map = !device->info->a7xx.load_shader_consts_via_preamble,
.EXT_fragment_density_map = true,
.EXT_global_priority = true,
.EXT_global_priority_query = true,
.EXT_graphics_pipeline_library = true,

7
src/freedreno/vulkan/tu_shader.cc
View File

@@ -477,10 +477,12 @@ lower_intrinsic(nir_builder *b, nir_intrinsic_instr *instr,
instr->intrinsic == nir_intrinsic_load_frag_size_ir3 ?
IR3_DP_FS_FRAG_SIZE : IR3_DP_FS_FRAG_OFFSET;
unsigned offset = param - IR3_DP_FS_DYNAMIC;
nir_def *view = instr->src[0].ssa;
nir_def *result =
ir3_load_driver_ubo_indirect(b, 2, &shader->const_state.fdm_ubo,
param, view, nir_intrinsic_range(instr));
offset, view, nir_intrinsic_range(instr));
nir_def_replace(&instr->def, result);
return true;
@@ -491,7 +493,8 @@ lower_intrinsic(nir_builder *b, nir_intrinsic_instr *instr,
nir_def *result =
ir3_load_driver_ubo(b, 1, &shader->const_state.fdm_ubo,
IR3_DP_FS_FRAG_INVOCATION_COUNT);
IR3_DP_FS_FRAG_INVOCATION_COUNT -
IR3_DP_FS_DYNAMIC);
nir_def_replace(&instr->def, result);
return true;