Signed-off-by: Rhys Perry <[email protected]>
---
src/amd/common/ac_nir_to_llvm.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 939b8eb13de..8bfc63958ca 100644
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -889,7 +889,9 @@ static void visit_alu(struct ac_nir_context *ctx, const
nir_alu_instr *instr)
src[0] = LLVMBuildFPTrunc(ctx->ac.builder, src[0],
ctx->ac.f32, "");
LLVMValueRef param[2] = { src[0], ctx->ac.f32_0 };
result = ac_build_cvt_pkrtz_f16(&ctx->ac, param);
- result = LLVMBuildExtractElement(ctx->ac.builder, result,
ctx->ac.i32_0, "");
+ // generates better code than an extractelement with slp
vectorization
+ result = LLVMBuildBitCast(ctx->ac.builder, result, ctx->ac.i32,
"");
+ result = LLVMBuildTrunc(ctx->ac.builder, result, ctx->ac.i16,
"");
break;
case nir_op_f2f16_rtne:
case nir_op_f2f16:
--
2.20.1
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev