nir/lower_vec_to_movs: Only coalesce if the vec had a SSA destination

Otherwise we may end up trying to coalesce in a case such as

ssa_1 = fadd r1, r2
r3.x = fneg(r2);
r3 = vec4(ssa_1, ssa_1.y, ...)

and that would cause us to move the writes to r3 from the vec to the
fadd which would re-order them with respect to the write from the fneg.
In order to solve this, we just don't coalesce if the destination of the
vec is not SSA.  We could try to get clever and still coalesce if there
are no writes to the destination of the vec between the vec and the ALU
source.  However, since registers only come from phi webs and indirects,
the chances of having a vec with a register destination that is actually
coalescable into its source is very slim.

Shader-db results on Haswell:

    total instructions in shared programs: 13657906 -> 13659101 (<.01%)
    instructions in affected programs: 149291 -> 150486 (0.80%)
    helped: 0
    HURT: 592

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105440
Fixes: 2458ea95c56 "nir/lower_vec_to_movs: Coalesce movs on-the-fly when possible"
Reported-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Tested-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/compiler/nir/nir_lower_vec_to_movs.c b/src/compiler/nir/nir_lower_vec_to_movs.c
index 711ddd3..8b24376 100644 (file)
--- a/src/compiler/nir/nir_lower_vec_to_movs.c
+++ b/src/compiler/nir/nir_lower_vec_to_movs.c
@@ -230,6 +230,7 @@ lower_vec_to_movs_block(nir_block *block, nir_function_impl *impl)
          continue; /* The loop */
       }
 
+      bool vec_had_ssa_dest = vec->dest.dest.is_ssa;
       if (vec->dest.dest.is_ssa) {
          /* Since we insert multiple MOVs, we have a register destination. */
          nir_register *reg = nir_local_reg_create(impl);
@@ -263,7 +264,11 @@ lower_vec_to_movs_block(nir_block *block, nir_function_impl *impl)
          if (!(vec->dest.write_mask & (1 << i)))
             continue;
 
-         if (!(finished_write_mask & (1 << i)))
+         /* Coalescing moves the register writes from the vec up to the ALU
+          * instruction in the source.  We can only do this if the original
+          * vecN had an SSA destination.
+          */
+         if (vec_had_ssa_dest && !(finished_write_mask & (1 << i)))
             finished_write_mask |= try_coalesce(vec, i);
 
          if (!(finished_write_mask & (1 << i)))