i965: Have NIR lower flrp on pre-GEN6 vec4 backend

Previously we were doing the lowering by hand in vec4_visitor::emit_lrp.
By doing it in NIR, we have the opportunity for NIR to do additional
optimization of the expanded code.

This also enables optimizations added by the next commit.

shader-db results:

G4X / Ironlake
total instructions in shared programs: 4024401 -> 4016538 (-0.20%)
instructions in affected programs: 447686 -> 439823 (-1.76%)
helped: 2623
HURT: 0

total cycles in shared programs: 84375846 -> 84328296 (-0.06%)
cycles in affected programs: 16964960 -> 16917410 (-0.28%)
helped: 2556
HURT: 41

Unsurprisingly, no changes on later platforms.

v2: Formatting and comment changes suggested by Matt.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/mesa/drivers/dri/i965/brw_compiler.c b/src/mesa/drivers/dri/i965/brw_compiler.c
index 2f05a26e0e03103d87aec4cfdbb40bab0dab8d55..3da6aac2cbfbaf79b49a2d270e348fe907b8a4e8 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_compiler.c
+++ b/src/mesa/drivers/dri/i965/brw_compiler.c
@@ -107,6 +107,26 @@ static const struct nir_shader_compiler_options vector_nir_options = {
     */
    .fdot_replicates = true,
 
+   /* Prior to Gen6, there are no three source operations for SIMD4x2. */
+   .lower_flrp = true,
+
+   .lower_pack_snorm_2x16 = true,
+   .lower_pack_unorm_2x16 = true,
+   .lower_unpack_snorm_2x16 = true,
+   .lower_unpack_unorm_2x16 = true,
+   .lower_extract_byte = true,
+   .lower_extract_word = true,
+};
+
+static const struct nir_shader_compiler_options vector_nir_options_gen6 = {
+   COMMON_OPTIONS,
+
+   /* In the vec4 backend, our dpN instruction replicates its result to all the
+    * components of a vec4.  We would like NIR to give us replicated fdot
+    * instructions because it can optimize better for us.
+    */
+   .fdot_replicates = true,
+
    .lower_pack_snorm_2x16 = true,
    .lower_pack_unorm_2x16 = true,
    .lower_unpack_snorm_2x16 = true,
@@ -159,8 +179,12 @@ brw_compiler_create(void *mem_ctx, const struct brw_device_info *devinfo)
       if (devinfo->gen < 7)
          compiler->glsl_compiler_options[i].EmitNoIndirectSampler = true;
 
-      compiler->glsl_compiler_options[i].NirOptions =
-         is_scalar ? &scalar_nir_options : &vector_nir_options;
+      if (is_scalar) {
+         compiler->glsl_compiler_options[i].NirOptions = &scalar_nir_options;
+      } else {
+         compiler->glsl_compiler_options[i].NirOptions =
+            devinfo->gen < 6 ? &vector_nir_options : &vector_nir_options_gen6;
+      }
 
       compiler->glsl_compiler_options[i].LowerBufferInterfaceBlocks = true;
    }