fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-03 05:38:16 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

ac/nir: legacy vs/gs use nir_xfb_info to replace pipe_stream_output_info

Browse source 
pipe_stream_output_info is built from nir_xfb_info, why not just use
nir_xfb_info directly.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20015>
This commit is contained in:
Qiang Yu 2022-11-22 14:25:52 +08:00 • committed by Marge Bot
parent 59b29c5fc4
commit 3507cdc59c
2 changed files with 46 additions and 42 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

76
src/amd/common/ac_nir.c
View file

@ -23,6 +23,7 @@
#include "ac_nir.h" #include "ac_nir.h"
#include "nir_builder.h" #include "nir_builder.h"
#include "nir_xfb_info.h"
nir_ssa_def * nir_ssa_def *
ac_nir_load_arg(nir_builder *b, const struct ac_shader_args *ac_args, struct ac_arg arg) ac_nir_load_arg(nir_builder *b, const struct ac_shader_args *ac_args, struct ac_arg arg)
@ -110,7 +111,7 @@ ac_nir_lower_indirect_derefs(nir_shader *shader,
} }
static void static void
emit_streamout(nir_builder *b, const struct pipe_stream_output_info *info, unsigned stream, emit_streamout(nir_builder *b, unsigned stream, nir_xfb_info *info,
nir_ssa_def *const outputs[64][4]) nir_ssa_def *const outputs[64][4])
{ {
nir_ssa_def *so_vtx_count = nir_ubfe_imm(b, nir_load_streamout_config_amd(b), 16, 7); nir_ssa_def *so_vtx_count = nir_ubfe_imm(b, nir_load_streamout_config_amd(b), 16, 7);
@ -119,44 +120,43 @@ emit_streamout(nir_builder *b, const struct pipe_stream_output_info *info, unsig
nir_push_if(b, nir_ilt(b, tid, so_vtx_count)); nir_push_if(b, nir_ilt(b, tid, so_vtx_count));
nir_ssa_def *so_write_index = nir_load_streamout_write_index_amd(b); nir_ssa_def *so_write_index = nir_load_streamout_write_index_amd(b);
nir_ssa_def *so_buffers[PIPE_MAX_SO_BUFFERS]; nir_ssa_def *so_buffers[NIR_MAX_XFB_BUFFERS];
nir_ssa_def *so_write_offset[PIPE_MAX_SO_BUFFERS]; nir_ssa_def *so_write_offset[NIR_MAX_XFB_BUFFERS];
for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) { u_foreach_bit(i, info->buffers_written) {
uint16_t stride = info->stride[i];
if (!stride)
continue;
so_buffers[i] = nir_load_streamout_buffer_amd(b, i); so_buffers[i] = nir_load_streamout_buffer_amd(b, i);
unsigned stride = info->buffers[i].stride;
nir_ssa_def *offset = nir_load_streamout_offset_amd(b, i); nir_ssa_def *offset = nir_load_streamout_offset_amd(b, i);
offset = nir_iadd(b, nir_imul_imm(b, nir_iadd(b, so_write_index, tid), stride * 4), offset = nir_iadd(b, nir_imul_imm(b, nir_iadd(b, so_write_index, tid), stride),
nir_imul_imm(b, offset, 4)); nir_imul_imm(b, offset, 4));
so_write_offset[i] = offset; so_write_offset[i] = offset;
} }
nir_ssa_def *undef = nir_ssa_undef(b, 1, 32); nir_ssa_def *undef = nir_ssa_undef(b, 1, 32);
for (unsigned i = 0; i < info->num_outputs; i++) { for (unsigned i = 0; i < info->output_count; i++) {
const struct pipe_stream_output *output = &info->output[i]; const nir_xfb_output_info *output = info->outputs + i;
if (stream != output->stream) if (stream != info->buffer_to_stream[output->buffer])
continue; continue;
nir_ssa_def *vec[4] = {undef, undef, undef, undef}; nir_ssa_def *vec[4] = {undef, undef, undef, undef};
uint8_t mask = 0; uint8_t mask = 0;
for (unsigned j = 0; j < output->num_components; j++) { u_foreach_bit(j, output->component_mask) {
if (outputs[output->register_index][output->start_component + j]) { nir_ssa_def *src = outputs[output->location][j];
vec[j] = outputs[output->register_index][output->start_component + j]; if (src) {
mask |= 1 << j; unsigned comp = j - output->component_offset;
vec[comp] = src;
mask |= 1 << comp;
} }
} }
if (!mask) if (!mask)
continue; continue;
unsigned buffer = output->output_buffer; unsigned buffer = output->buffer;
nir_ssa_def *data = nir_vec(b, vec, output->num_components); nir_ssa_def *data = nir_vec(b, vec, util_last_bit(mask));
nir_ssa_def *zero = nir_imm_int(b, 0); nir_ssa_def *zero = nir_imm_int(b, 0);
nir_store_buffer_amd(b, data, so_buffers[buffer], so_write_offset[buffer], zero, zero, nir_store_buffer_amd(b, data, so_buffers[buffer], so_write_offset[buffer], zero, zero,
.base = output->dst_offset * 4, .slc_amd = true, .write_mask = mask, .base = output->offset, .slc_amd = true, .write_mask = mask,
.access = ACCESS_COHERENT); .access = ACCESS_COHERENT);
} }
@ -165,10 +165,11 @@ emit_streamout(nir_builder *b, const struct pipe_stream_output_info *info, unsig
nir_shader * nir_shader *
ac_nir_create_gs_copy_shader(const nir_shader *gs_nir, ac_nir_create_gs_copy_shader(const nir_shader *gs_nir,
const struct pipe_stream_output_info *so_info, size_t num_outputs, bool disable_streamout,
const uint8_t *output_usage_mask, const uint8_t *output_streams, size_t num_outputs,
const uint8_t *output_semantics, const uint8_t *output_usage_mask,
const uint8_t num_stream_output_components[4]) const uint8_t *output_streams,
const uint8_t *output_semantics)
{ {
assert(num_outputs <= 64); assert(num_outputs <= 64);
@ -180,15 +181,16 @@ ac_nir_create_gs_copy_shader(const nir_shader *gs_nir,
nir_ssa_def *gsvs_ring = nir_load_ring_gsvs_amd(&b); nir_ssa_def *gsvs_ring = nir_load_ring_gsvs_amd(&b);
nir_xfb_info *info = gs_nir->xfb_info;
nir_ssa_def *stream_id = NULL; nir_ssa_def *stream_id = NULL;
if (so_info->num_outputs) if (!disable_streamout && info)
stream_id = nir_ubfe_imm(&b, nir_load_streamout_config_amd(&b), 24, 2); stream_id = nir_ubfe_imm(&b, nir_load_streamout_config_amd(&b), 24, 2);
nir_ssa_def *vtx_offset = nir_imul_imm(&b, nir_load_vertex_id_zero_base(&b), 4); nir_ssa_def *vtx_offset = nir_imul_imm(&b, nir_load_vertex_id_zero_base(&b), 4);
nir_ssa_def *zero = nir_imm_zero(&b, 1, 32); nir_ssa_def *zero = nir_imm_zero(&b, 1, 32);
for (unsigned stream = 0; stream < 4; stream++) { for (unsigned stream = 0; stream < 4; stream++) {
if (stream > 0 && (!stream_id || !num_stream_output_components[stream])) if (stream > 0 && (!stream_id || !(info->streams_written & BITFIELD_BIT(stream))))
continue; continue;
if (stream_id) if (stream_id)
@ -202,13 +204,16 @@ ac_nir_create_gs_copy_shader(const nir_shader *gs_nir,
if (!mask) if (!mask)
continue; continue;
gl_varying_slot location = output_semantics ? output_semantics[i] : i;
u_foreach_bit (j, mask) { u_foreach_bit (j, mask) {
if (((output_streams[i] >> (j * 2)) & 0x3) != stream) if (((output_streams[i] >> (j * 2)) & 0x3) != stream)
continue; continue;
outputs[i][j] = nir_load_buffer_amd(&b, 1, 32, gsvs_ring, vtx_offset, zero, zero, outputs[location][j] =
.base = offset, .is_swizzled = false, nir_load_buffer_amd(&b, 1, 32, gsvs_ring, vtx_offset, zero, zero,
.slc_amd = true, .access = ACCESS_COHERENT); .base = offset, .is_swizzled = false,
.slc_amd = true, .access = ACCESS_COHERENT);
offset += gs_nir->info.gs.vertices_out * 16 * 4; offset += gs_nir->info.gs.vertices_out * 16 * 4;
} }
@ -217,15 +222,15 @@ ac_nir_create_gs_copy_shader(const nir_shader *gs_nir,
} }
if (stream_id) if (stream_id)
emit_streamout(&b, so_info, stream, outputs); emit_streamout(&b, stream, info, outputs);
if (stream == 0) { if (stream == 0) {
u_foreach_bit64 (i, output_mask) { u_foreach_bit64 (i, output_mask) {
gl_varying_slot location = output_semantics ? output_semantics[i] : i; gl_varying_slot location = output_semantics ? output_semantics[i] : i;
for (unsigned j = 0; j < 4; j++) { for (unsigned j = 0; j < 4; j++) {
if (outputs[i][j]) { if (outputs[location][j]) {
nir_store_output(&b, outputs[i][j], zero, nir_store_output(&b, outputs[location][j], zero,
.base = i, .base = i,
.component = j, .component = j,
.write_mask = 1, .write_mask = 1,
@ -267,7 +272,7 @@ gather_outputs(nir_builder *b, nir_function_impl *impl, nir_ssa_def *outputs[64]
assert(nir_src_is_const(intrin->src[1]) && !nir_src_as_uint(intrin->src[1])); assert(nir_src_is_const(intrin->src[1]) && !nir_src_as_uint(intrin->src[1]));
unsigned slot = nir_intrinsic_base(intrin); unsigned slot = nir_intrinsic_io_semantics(intrin).location;
u_foreach_bit (i, nir_intrinsic_write_mask(intrin)) { u_foreach_bit (i, nir_intrinsic_write_mask(intrin)) {
unsigned comp = nir_intrinsic_component(intrin) + i; unsigned comp = nir_intrinsic_component(intrin) + i;
outputs[slot][comp] = nir_channel(b, intrin->src[0].ssa, i); outputs[slot][comp] = nir_channel(b, intrin->src[0].ssa, i);
@ -277,8 +282,7 @@ gather_outputs(nir_builder *b, nir_function_impl *impl, nir_ssa_def *outputs[64]
} }
void void
ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location, ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location, bool disable_streamout)
const struct pipe_stream_output_info *so_info)
{ {
nir_function_impl *impl = nir_shader_get_entrypoint(nir); nir_function_impl *impl = nir_shader_get_entrypoint(nir);
nir_metadata preserved = nir_metadata_block_index | nir_metadata_dominance; nir_metadata preserved = nir_metadata_block_index | nir_metadata_dominance;
@ -306,7 +310,7 @@ ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location,
nir->info.outputs_written |= BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_ID); nir->info.outputs_written |= BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_ID);
} }
if (so_info && so_info->num_outputs) { if (!disable_streamout && nir->xfb_info) {
/* 26.1. Transform Feedback of Vulkan 1.3.229 spec: /* 26.1. Transform Feedback of Vulkan 1.3.229 spec:
* > The size of each component of an output variable must be at least 32-bits. * > The size of each component of an output variable must be at least 32-bits.
* We lower 64-bit outputs. * We lower 64-bit outputs.
@ -314,7 +318,7 @@ ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location,
nir_ssa_def *outputs[64][4] = {{0}}; nir_ssa_def *outputs[64][4] = {{0}};
gather_outputs(&b, impl, outputs); gather_outputs(&b, impl, outputs);
emit_streamout(&b, so_info, 0, outputs); emit_streamout(&b, 0, nir->xfb_info, outputs);
preserved = nir_metadata_none; preserved = nir_metadata_none;
} }

12
src/amd/common/ac_nir.h
View file

@ -185,14 +185,14 @@ bool ac_nir_lower_resinfo(nir_shader *nir, enum amd_gfx_level gfx_level);
nir_shader * nir_shader *
ac_nir_create_gs_copy_shader(const nir_shader *gs_nir, ac_nir_create_gs_copy_shader(const nir_shader *gs_nir,
const struct pipe_stream_output_info *so_info, size_t num_outputs, bool disable_streamout,
const uint8_t *output_usage_mask, const uint8_t *output_streams, size_t num_outputs,
const uint8_t *output_semantics, const uint8_t *output_usage_mask,
const uint8_t num_stream_output_components[4]); const uint8_t *output_streams,
const uint8_t *output_semantics);
void void
ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location, ac_nir_lower_legacy_vs(nir_shader *nir, int primitive_id_location, bool disable_streamout);
const struct pipe_stream_output_info *so_info);
#ifdef __cplusplus #ifdef __cplusplus
} }