diff options
Diffstat (limited to 'src/compiler')
| -rw-r--r-- | src/compiler/nir/nir.h | 2 | ||||
| -rw-r--r-- | src/compiler/nir/nir_opt_peephole_select.c | 39 |
2 files changed, 32 insertions, 9 deletions
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h index 2bbfb3c6b15..54f9c64a3a3 100644 --- a/src/compiler/nir/nir.h +++ b/src/compiler/nir/nir.h @@ -3198,7 +3198,7 @@ bool nir_opt_move_comparisons(nir_shader *shader); bool nir_opt_move_load_ubo(nir_shader *shader); bool nir_opt_peephole_select(nir_shader *shader, unsigned limit, - bool indirect_load_ok); + bool indirect_load_ok, bool expensive_alu_ok); bool nir_opt_remove_phis_impl(nir_function_impl *impl); bool nir_opt_remove_phis(nir_shader *shader); diff --git a/src/compiler/nir/nir_opt_peephole_select.c b/src/compiler/nir/nir_opt_peephole_select.c index 6808d3eda6c..b27cdd28a99 100644 --- a/src/compiler/nir/nir_opt_peephole_select.c +++ b/src/compiler/nir/nir_opt_peephole_select.c @@ -59,7 +59,8 @@ static bool block_check_for_allowed_instrs(nir_block *block, unsigned *count, - bool alu_ok, bool indirect_load_ok) + bool alu_ok, bool indirect_load_ok, + bool expensive_alu_ok) { nir_foreach_instr(instr, block) { switch (instr->type) { @@ -117,6 +118,25 @@ block_check_for_allowed_instrs(nir_block *block, unsigned *count, case nir_op_vec3: case nir_op_vec4: break; + + case nir_op_fcos: + case nir_op_fdiv: + case nir_op_fexp2: + case nir_op_flog2: + case nir_op_fmod: + case nir_op_fpow: + case nir_op_frcp: + case nir_op_frem: + case nir_op_frsq: + case nir_op_fsin: + case nir_op_idiv: + case nir_op_irem: + case nir_op_udiv: + if (!alu_ok || !expensive_alu_ok) + return false; + + break; + default: if (!alu_ok) { /* It must be a move-like operation. */ @@ -160,7 +180,8 @@ block_check_for_allowed_instrs(nir_block *block, unsigned *count, static bool nir_opt_peephole_select_block(nir_block *block, nir_shader *shader, - unsigned limit, bool indirect_load_ok) + unsigned limit, bool indirect_load_ok, + bool expensive_alu_ok) { if (nir_cf_node_is_first(&block->cf_node)) return false; @@ -181,9 +202,9 @@ nir_opt_peephole_select_block(nir_block *block, nir_shader *shader, /* ... and those blocks must only contain "allowed" instructions. */ unsigned count = 0; if (!block_check_for_allowed_instrs(then_block, &count, limit != 0, - indirect_load_ok) || + indirect_load_ok, expensive_alu_ok) || !block_check_for_allowed_instrs(else_block, &count, limit != 0, - indirect_load_ok)) + indirect_load_ok, expensive_alu_ok)) return false; if (count > limit) @@ -250,14 +271,15 @@ nir_opt_peephole_select_block(nir_block *block, nir_shader *shader, static bool nir_opt_peephole_select_impl(nir_function_impl *impl, unsigned limit, - bool indirect_load_ok) + bool indirect_load_ok, bool expensive_alu_ok) { nir_shader *shader = impl->function->shader; bool progress = false; nir_foreach_block_safe(block, impl) { progress |= nir_opt_peephole_select_block(block, shader, limit, - indirect_load_ok); + indirect_load_ok, + expensive_alu_ok); } if (progress) @@ -268,14 +290,15 @@ nir_opt_peephole_select_impl(nir_function_impl *impl, unsigned limit, bool nir_opt_peephole_select(nir_shader *shader, unsigned limit, - bool indirect_load_ok) + bool indirect_load_ok, bool expensive_alu_ok) { bool progress = false; nir_foreach_function(function, shader) { if (function->impl) progress |= nir_opt_peephole_select_impl(function->impl, limit, - indirect_load_ok); + indirect_load_ok, + expensive_alu_ok); } return progress; |