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

nir: get ffma support from NIR options for nir_lower_flrp

Browse Source 
This also fixes the inverted last parameter of nir_lower_flrp in most drivers.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6599>
This commit is contained in:
Marek Olšák
2020-07-22 22:13:16 -04:00
committed by Marge Bot
parent efd439e0d6
commit ac55b1a9a6
14 changed files with 20 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.gitlab-ci/traces-radeonsi.yml
View File

@@ -173,7 +173,7 @@ traces:
- path: godot/Material Testers.x86_64_2020.04.08_13.38_frame799.rdc
expectations:
- device: gl-radeonsi-stoney
checksum: 2ddcda6b1c136ce1448714a3ff8432eb
checksum: 5164e238381e7d77a64e3de771cc005f
- path: gputest/gimark.trace
expectations:
- device: gl-radeonsi-stoney

3
src/amd/vulkan/radv_shader.c
View File

@@ -261,8 +261,7 @@ radv_optimize_nir(struct nir_shader *shader, bool optimize_conservatively,
shader,
nir_lower_flrp,
lower_flrp,
false /* always_precise */,
shader->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, shader,
nir_opt_constant_folding);

3
src/broadcom/compiler/nir_to_vir.c
View File

@@ -1451,8 +1451,7 @@ v3d_optimize_nir(struct nir_shader *s)
NIR_PASS(lower_flrp_progress, s, nir_lower_flrp,
lower_flrp,
false /* always_precise */,
s->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, s, nir_opt_constant_folding);
progress = true;

2
src/compiler/nir/nir.h
View File

@@ -4314,7 +4314,7 @@ void nir_lower_alpha_test(nir_shader *shader, enum compare_func func,
bool nir_lower_alu(nir_shader *shader);
bool nir_lower_flrp(nir_shader *shader, unsigned lowering_mask,
bool always_precise, bool have_ffma);
bool always_precise);
bool nir_lower_alu_to_scalar(nir_shader *shader, nir_instr_filter_cb cb, const void *data);
bool nir_lower_bool_to_bitsize(nir_shader *shader);

16
src/compiler/nir/nir_lower_flrp.c
View File

@@ -364,9 +364,10 @@ static void
convert_flrp_instruction(nir_builder *bld,
struct u_vector *dead_flrp,
nir_alu_instr *alu,
bool always_precise,
bool have_ffma)
bool always_precise)
{
bool have_ffma = !bld->shader->options->lower_ffma;
bld->cursor = nir_before_instr(&alu->instr);
/* There are two methods to implement flrp(x, y, t). The strictly correct
@@ -586,8 +587,7 @@ static void
lower_flrp_impl(nir_function_impl *impl,
struct u_vector *dead_flrp,
unsigned lowering_mask,
bool always_precise,
bool have_ffma)
bool always_precise)
{
nir_builder b;
nir_builder_init(&b, impl);
@@ -599,8 +599,7 @@ lower_flrp_impl(nir_function_impl *impl,
if (alu->op == nir_op_flrp &&
(alu->dest.dest.ssa.bit_size & lowering_mask)) {
convert_flrp_instruction(&b, dead_flrp, alu, always_precise,
have_ffma);
convert_flrp_instruction(&b, dead_flrp, alu, always_precise);
}
}
}
@@ -622,8 +621,7 @@ lower_flrp_impl(nir_function_impl *impl,
bool
nir_lower_flrp(nir_shader *shader,
unsigned lowering_mask,
bool always_precise,
bool have_ffma)
bool always_precise)
{
struct u_vector dead_flrp;
@@ -633,7 +631,7 @@ nir_lower_flrp(nir_shader *shader,
nir_foreach_function(function, shader) {
if (function->impl) {
lower_flrp_impl(function->impl, &dead_flrp, lowering_mask,
always_precise, have_ffma);
always_precise);
}
}

3
src/freedreno/ir3/ir3_nir.c
View File

@@ -185,8 +185,7 @@ ir3_optimize_loop(nir_shader *s)
if (lower_flrp != 0) {
if (OPT(s, nir_lower_flrp,
lower_flrp,
false /* always_precise */,
s->options->lower_ffma)) {
false /* always_precise */)) {
OPT(s, nir_opt_constant_folding);
progress = true;
}

2
src/gallium/drivers/r600/r600_shader.c
View File

@@ -203,7 +203,7 @@ int r600_pipe_shader_create(struct pipe_context *ctx,
NIR_PASS_V(sel->nir, nir_lower_int64);
NIR_PASS_V(sel->nir, nir_opt_vectorize, NULL, NULL);
}
NIR_PASS_V(sel->nir, nir_lower_flrp, ~0, false, false);
NIR_PASS_V(sel->nir, nir_lower_flrp, ~0, false);
}
nir_tgsi_scan_shader(sel->nir, &sel->info, true);

3
src/gallium/drivers/radeonsi/si_shader_nir.c
View File

@@ -644,8 +644,7 @@ static void si_nir_opts(struct nir_shader *nir, bool first)
assert(lower_flrp);
bool lower_flrp_progress = false;
NIR_PASS(lower_flrp_progress, nir, nir_lower_flrp, lower_flrp, false /* always_precise */,
nir->options->lower_ffma);
NIR_PASS(lower_flrp_progress, nir, nir_lower_flrp, lower_flrp, false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, nir, nir_opt_constant_folding);
progress = true;

3
src/gallium/drivers/vc4/vc4_program.c
View File

@@ -1546,8 +1546,7 @@ vc4_optimize_nir(struct nir_shader *s)
NIR_PASS(lower_flrp_progress, s, nir_lower_flrp,
lower_flrp,
false /* always_precise */,
s->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, s, nir_opt_constant_folding);
progress = true;

2
src/gallium/frontends/vallium/val_pipeline.c
View File

@@ -590,7 +590,7 @@ val_shader_compile_to_ir(struct val_pipeline *pipeline,
do {
progress = false;
progress |= OPT(nir_lower_flrp, 32|64, true, false);
progress |= OPT(nir_lower_flrp, 32|64, true);
progress |= OPT(nir_split_array_vars, nir_var_function_temp);
progress |= OPT(nir_shrink_vec_array_vars, nir_var_function_temp);
progress |= OPT(nir_opt_deref);

3
src/intel/compiler/brw_nir.c
View File

@@ -597,8 +597,7 @@ brw_nir_optimize(nir_shader *nir, const struct brw_compiler *compiler,
if (lower_flrp != 0) {
if (OPT(nir_lower_flrp,
lower_flrp,
false /* always_precise */,
compiler->devinfo->gen >= 6)) {
false /* always_precise */)) {
OPT(nir_opt_constant_folding);
}

3
src/mesa/state_tracker/st_glsl_to_nir.cpp
View File

@@ -305,8 +305,7 @@ st_nir_opts(nir_shader *nir)
NIR_PASS(lower_flrp_progress, nir, nir_lower_flrp,
lower_flrp,
false /* always_precise */,
nir->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, nir,
nir_opt_constant_folding);

3
src/panfrost/bifrost/bifrost_compile.c
View File

@@ -1272,8 +1272,7 @@ bi_optimize_nir(nir_shader *nir)
nir,
nir_lower_flrp,
lower_flrp,
false /* always_precise */,
nir->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, nir,
nir_opt_constant_folding);

3
src/panfrost/midgard/midgard_compile.c
View File

@@ -536,8 +536,7 @@ optimise_nir(nir_shader *nir, unsigned quirks, bool is_blend)
nir,
nir_lower_flrp,
lower_flrp,
false /* always_precise */,
nir->options->lower_ffma);
false /* always_precise */);
if (lower_flrp_progress) {
NIR_PASS(progress, nir,
nir_opt_constant_folding);