diff options
| -rw-r--r-- | src/gallium/drivers/freedreno/ir3/ir3_nir.c | 43 |
1 files changed, 27 insertions, 16 deletions
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_nir.c index 58bef169ad0..023da3be292 100644 --- a/src/gallium/drivers/freedreno/ir3/ir3_nir.c +++ b/src/gallium/drivers/freedreno/ir3/ir3_nir.c @@ -82,6 +82,27 @@ ir3_key_lowers_nir(const struct ir3_shader_key *key) #define OPT_V(nir, pass, ...) NIR_PASS_V(nir, pass, ##__VA_ARGS__) +static void +ir3_optimize_loop(nir_shader *s) +{ + bool progress; + do { + progress = false; + + OPT_V(s, nir_lower_vars_to_ssa); + OPT_V(s, nir_lower_alu_to_scalar); + OPT_V(s, nir_lower_phis_to_scalar); + + progress |= OPT(s, nir_copy_prop); + progress |= OPT(s, nir_opt_dce); + progress |= OPT(s, nir_opt_cse); + progress |= OPT(s, ir3_nir_lower_if_else); + progress |= OPT(s, nir_opt_algebraic); + progress |= OPT(s, nir_opt_constant_folding); + + } while (progress); +} + struct nir_shader * ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s, const struct ir3_shader_key *key) @@ -89,7 +110,6 @@ ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s, struct nir_lower_tex_options tex_options = { .lower_rect = 0, }; - bool progress; if (key) { switch (shader->type) { @@ -145,24 +165,15 @@ ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s, } OPT_V(s, nir_lower_tex, &tex_options); - OPT_V(s, nir_lower_idiv); OPT_V(s, nir_lower_load_const_to_scalar); - do { - progress = false; - - OPT_V(s, nir_lower_vars_to_ssa); - OPT_V(s, nir_lower_alu_to_scalar); - OPT_V(s, nir_lower_phis_to_scalar); + ir3_optimize_loop(s); - progress |= OPT(s, nir_copy_prop); - progress |= OPT(s, nir_opt_dce); - progress |= OPT(s, nir_opt_cse); - progress |= OPT(s, ir3_nir_lower_if_else); - progress |= OPT(s, nir_opt_algebraic); - progress |= OPT(s, nir_opt_constant_folding); - - } while (progress); + /* do idiv lowering after first opt loop to give a chance for + * divide by immed power-of-two to be caught first: + */ + if (OPT(s, nir_lower_idiv)) + ir3_optimize_loop(s); OPT_V(s, nir_remove_dead_variables, nir_var_local); |