[Mesa-dev] [PATCH] i965/nir: Don't emit a copy after a texture instruction if it's not needed

[Jason Ekstrand]

Shader-db results on HSW:
total instructions in shared programs: 4112172 -> 4111347 (-0.02%)
instructions in affected programs:     18265 -> 17440 (-4.52%)
helped:                                137
HURT:                                  0
GAINED:                                0
LOST:                                  0
---
 src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 17 ++++++++++++++---
 1 file changed, 14 insertions(+), 3 deletions(-)

diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
index 42c50cb..625a5ae 100644
--- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
+++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
@@ -1895,9 +1895,20 @@ fs_visitor::nir_emit_texture(nir_tex_instr *instr)
                 is_cube_array, is_rect, sampler, sampler_reg, texunit);
 
    fs_reg dest = get_nir_dest(instr->dest);
-   dest.type = this->result.type;
-   unsigned num_components = nir_tex_instr_dest_size(instr);
-   emit_percomp(MOV(dest, this->result), (1 << num_components) - 1);
+
+   fs_inst *inst = (fs_inst *) this->instructions.get_tail();
+   if (inst->is_tex() && alloc.sizes[inst->dst.reg] <= alloc.sizes[dest.reg]) {
+      // The last instruction is the actual texture operation and we have
+      // enough space in the NIR dest register.  Just smash the destination
+      // with the register from the NIR dest.
+      assert(inst->dst.file == GRF && dest.file == GRF);
+      assert(dest.reg_offset == 0);
+      inst->dst.reg = dest.reg;
+   } else {
+      dest.type = this->result.type;
+      unsigned num_components = nir_tex_instr_dest_size(instr);
+      emit_percomp(MOV(dest, this->result), (1 << num_components) - 1);
+   }
 }
 
 void
-- 
2.3.2