From 536c8cbcd3326fc7207f8a2c166c380135924589 Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Tue, 5 Jan 2016 05:09:46 -0800 Subject: [PATCH] nir: Add a lower_fdiv option, turn fdiv into fmul/frcp. The nir_opt_algebraic rule (('fadd', ('flog2', a), ('fneg', ('flog2', b))), ('flog2', ('fdiv', a, b))), can produce new fdiv operations, which need to be lowered on i965, as we don't actually implement fdiv. (Normally, we handle this in GLSL IR's lower_instructions pass, but in the above case we introduce an fdiv after that point. So, make NIR do it for us.) Signed-off-by: Kenneth Graunke Reviewed-by: Jason Ekstrand Reviewed-by: Matt Turner Cc: mesa-stable@lists.freedesktop.org (cherry picked from commit 7295f4fcc2b2dd1bc6a8d1d834774b8152a029cf) --- src/glsl/nir/nir.h | 1 + src/glsl/nir/nir_opt_algebraic.py | 1 + src/mesa/drivers/dri/i965/brw_shader.cpp | 1 + 3 files changed, 3 insertions(+) diff --git a/src/glsl/nir/nir.h b/src/glsl/nir/nir.h index 087b4537c09..09caf34c830 100644 --- a/src/glsl/nir/nir.h +++ b/src/glsl/nir/nir.h @@ -1435,6 +1435,7 @@ typedef struct nir_function { exec_list_get_head(&(func)->overload_list), node) typedef struct nir_shader_compiler_options { + bool lower_fdiv; bool lower_ffma; bool lower_flrp; bool lower_fpow; diff --git a/src/glsl/nir/nir_opt_algebraic.py b/src/glsl/nir/nir_opt_algebraic.py index 6aa8b1f6cae..fb36acbbecb 100644 --- a/src/glsl/nir/nir_opt_algebraic.py +++ b/src/glsl/nir/nir_opt_algebraic.py @@ -179,6 +179,7 @@ optimizations = [ (('fmul', ('fexp2', a), ('fexp2', b)), ('fexp2', ('fadd', a, b))), # Division and reciprocal (('fdiv', 1.0, a), ('frcp', a)), + (('fdiv', a, b), ('fmul', a, ('frcp', b)), 'options->lower_fdiv'), (('frcp', ('frcp', a)), a), (('frcp', ('fsqrt', a)), ('frsq', a)), (('fsqrt', a), ('frcp', ('frsq', a)), 'options->lower_fsqrt'), diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp index 7a236cd9880..3dc57e4b9be 100644 --- a/src/mesa/drivers/dri/i965/brw_shader.cpp +++ b/src/mesa/drivers/dri/i965/brw_shader.cpp @@ -94,6 +94,7 @@ brw_compiler_create(void *mem_ctx, const struct brw_device_info *devinfo) nir_shader_compiler_options *nir_options = rzalloc(compiler, nir_shader_compiler_options); nir_options->native_integers = true; + nir_options->lower_fdiv = true; /* In order to help allow for better CSE at the NIR level we tell NIR * to split all ffma instructions during opt_algebraic and we then * re-combine them as a later step.