ericxu233 wrote:
> > > @phoebewang Are you OK with this? An alternative is we just allow exact
> > > rounding (which is what we'll end up doing for the SSE2 style conversions
> > > like CVTPD2DQ).
> >
> >
> > Can we check for strictfp mode? I think we can assume `NearestTiesToEven`
> > in no
ericxu233 wrote:
@RKSimon Ping for status. I've fixed this last review comment. Would this be
ready to merge :)?
https://github.com/llvm/llvm-project/pull/162295
___
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/ma
@@ -56,14 +87,40 @@ TEST_CONSTEXPR(match_m256(
1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 0.5f, -2.0f, 0.0f
));
-__m128i test_mm_cvtps_ph(__m128 a) {
- // CHECK-LABEL: test_mm_cvtps_ph
- // CHECK: call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %{{.*}}, i32 0)
- return _mm_cvtps_
https://github.com/ericxu233 updated
https://github.com/llvm/llvm-project/pull/162295
>From eec4aebc63665d0b2fd4e07d42bd62e268d1f69f Mon Sep 17 00:00:00 2001
From: ericxu233
Date: Sun, 5 Oct 2025 01:41:40 -0400
Subject: [PATCH 1/4] [X86][Clang] Allow constexpr evaluation of F16C CVTPS2PH
intri
ericxu233 wrote:
> I just meant the AVX512F/VL variants vcvtps2ph_mask / vcvtps2ph256_mask /
> __builtin_ia32_vcvtps2ph512_mask - those should just require additional
> handling for the getNumArgs() == 4 masked variant.
>
> But I'm happy for this just to focus on the F16C cases if you prefer t
