Change use of removed intrinsic

A recent LLVM change removed the __builtin_ia32_vcvtph2ps256 intrinsic
because it can be implemented using __builtin_shufflevector and/or
__builtin_convertvector. This CL changes skia to use convertvector as
to match.

Bug: chromium:438445382
Change-Id: I2ed16cd507d7feeafba4c082b3840cbdd6368dd9
Reviewed-on: https://skia-review.googlesource.com/c/skcms/+/1038217
Reviewed-by: Ben Wagner <bungeman@google.com>
Commit-Queue: Ben Wagner <bungeman@google.com>
Auto-Submit: Devon Loehr <dloehr@google.com>

src/Transform_inl.h

diff --git a/src/Transform_inl.h b/src/Transform_inl.h
index bc10e7d..99faf25 100644
--- a/src/Transform_inl.h
+++ b/src/Transform_inl.h
@@ -156,8 +156,8 @@
 #elif defined(USING_AVX512F)
     return (F)_mm512_cvtph_ps((__m256i)half);
 #elif defined(USING_AVX_F16C)
-    typedef int16_t __attribute__((vector_size(16))) I16;
-    return __builtin_ia32_vcvtph2ps256((I16)half);
+    typedef _Float16 __attribute__((vector_size(16))) F16;
+    return __builtin_convertvector((F16)half, F);
 #else
     U32 wide = cast<U32>(half);
     // A half is 1-5-10 sign-exponent-mantissa, with 15 exponent bias.