From 51e537c9fa4d10bc5b065a60095bf2d85080d3c5 Mon Sep 17 00:00:00 2001 From: Alyssa Rosenzweig Date: Fri, 6 Mar 2020 16:29:35 -0500 Subject: [PATCH] pan/bi: Implement load_const In the laziest possible way... We can just emit worst case moves which DCE will eat for breakfast anyway, and inline constants on instructions where that is supported directly. This approach eliminates a lot of nasty corner cases that Midgard's crazy cache scheme has hit, at the expense of slightly more work for DCE (but it's only a single iteration of an O(N) pass that has to run anyway..) Part-of: --- src/panfrost/bifrost/bifrost_compile.c | 31 +++++++++++++++++++++++++------ 1 file changed, 25 insertions(+), 6 deletions(-) diff --git a/src/panfrost/bifrost/bifrost_compile.c b/src/panfrost/bifrost/bifrost_compile.c index 3aedcd0..a0b9d03 100644 --- a/src/panfrost/bifrost/bifrost_compile.c +++ b/src/panfrost/bifrost/bifrost_compile.c @@ -222,14 +222,33 @@ emit_intrinsic(bi_context *ctx, nir_intrinsic_instr *instr) } static void +emit_load_const(bi_context *ctx, nir_load_const_instr *instr) +{ + /* Make sure we've been lowered */ + assert(instr->def.num_components == 1); + + bi_instruction move = { + .type = BI_MOV, + .dest = bir_ssa_index(&instr->def), + .dest_type = instr->def.bit_size | nir_type_uint, + .src = { + BIR_INDEX_CONSTANT + }, + .constant = { + .u64 = nir_const_value_as_uint(instr->value[0], instr->def.bit_size) + } + }; + + bi_emit(ctx, move); +} + +static void emit_instr(bi_context *ctx, struct nir_instr *instr) { switch (instr->type) { -#if 0 case nir_instr_type_load_const: emit_load_const(ctx, nir_instr_as_load_const(instr)); break; -#endif case nir_instr_type_intrinsic: emit_intrinsic(ctx, nir_instr_as_intrinsic(instr)); @@ -494,6 +513,8 @@ bi_optimize_nir(nir_shader *nir) }; NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options); + NIR_PASS(progress, nir, nir_lower_alu_to_scalar, NULL, NULL); + NIR_PASS(progress, nir, nir_lower_load_const_to_scalar); do { progress = false; @@ -538,6 +559,8 @@ bi_optimize_nir(nir_shader *nir) } while (progress); NIR_PASS(progress, nir, nir_opt_algebraic_late); + NIR_PASS(progress, nir, nir_lower_alu_to_scalar, NULL, NULL); + NIR_PASS(progress, nir, nir_lower_load_const_to_scalar); /* Take us out of SSA */ NIR_PASS(progress, nir, nir_lower_locals_to_regs); @@ -571,10 +594,6 @@ bifrost_compile_shader_nir(nir_shader *nir, bifrost_program *program, unsigned p NIR_PASS_V(nir, nir_lower_io, nir_var_all, glsl_type_size, 0); NIR_PASS_V(nir, nir_lower_ssbo); - /* We have to lower ALU to scalar ourselves since viewport - * transformations produce vector ops */ - NIR_PASS_V(nir, nir_lower_alu_to_scalar, NULL, NULL); - bi_optimize_nir(nir); nir_print_shader(nir, stdout); -- 2.7.4