NeroReflex/mesa
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

freedreno: Use the NIR lowering for isign.

Browse source 
I think this will save an instruction and hopefully not increase any other
costs (possibly the immediate -1 and 1?), but I haven't actually tested.

Reviewed-by: Kristian H. Kristensen <hoegsberg@chromium.org>
This commit is contained in:
Eric Anholt 2019-02-06 13:32:21 -08:00
parent 8f3694e1ab
commit 338d399fd0
2 changed files with 1 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/freedreno/ir3/ir3_compiler_nir.c
View file

@ -563,20 +563,6 @@ emit_alu(struct ir3_context *ctx, nir_alu_instr *alu)
case nir_op_ishr:
dst[0] = ir3_ASHR_B(b, src[0], 0, src[1], 0);
break;
case nir_op_isign: {
/* maybe this would be sane to lower in nir.. */
struct ir3_instruction *neg, *pos;
neg = ir3_CMPS_S(b, src[0], 0, create_immed(b, 0), 0);
neg->cat2.condition = IR3_COND_LT;
pos = ir3_CMPS_S(b, src[0], 0, create_immed(b, 0), 0);
pos->cat2.condition = IR3_COND_GT;
dst[0] = ir3_SUB_U(b, pos, 0, neg, 0);
break;
}
case nir_op_isub:
dst[0] = ir3_SUB_U(b, src[0], 0, src[1], 0);
break;

1
src/freedreno/ir3/ir3_nir.c
View file

@ -40,6 +40,7 @@ static const nir_shader_compiler_options options = {
.lower_fmod32 = true,
.lower_fmod64 = true,
.lower_fdiv = true,
.lower_isign = true,
.lower_ldexp = true,
.fuse_ffma = true,
.native_integers = true,