pengfei added a comment.
LGTM.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
FreddyYe wrote:
> pengfei wrote:
> > FreddyYe wrote:
> > > pengfei wrote:
> > > >
FreddyYe marked an inline comment as done.
FreddyYe added inline comments.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
pengfei wrote:
> FreddyYe wrote:
> >
pengfei added inline comments.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
FreddyYe wrote:
> pengfei wrote:
> > pengfei wrote:
> > > RKSimon wrote:
> > > > Is
FreddyYe marked an inline comment as done.
FreddyYe added inline comments.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
pengfei wrote:
> pengfei wrote:
> >
pengfei added inline comments.
Comment at: clang/include/clang/Driver/Options.td:4595-4596
def mno_avxvnniint8 : Flag<["-"], "mno-avxvnniint8">,
Group;
+def mavxneconvert : Flag<["-"], "mavxneconvert">, Group;
+def mno_avxneconvert : Flag<["-"], "mno-avxneconvert">,
Group;
pengfei added inline comments.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
RKSimon wrote:
> Is there no way for __attribute__ to allow different attribute
RKSimon added inline comments.
Comment at: clang/lib/Headers/avx512vlbf16intrin.h:164
+#define _mm_cvtneps_pbh(A) \
+ ((__m128bh)__builtin_ia32_vcvtneps2bf16128((__v4sf)(A)))
Is there no way for __attribute__ to allow different attribute permutations?
Also,
FreddyYe added inline comments.
Comment at: llvm/test/CodeGen/X86/avxneconvert-intrinsics.ll:4
+; RUN: llc < %s -O0 -verify-machineinstrs -mtriple=i686-unknown-unknown
--show-mc-encoding -mattr=+avxneconvert | FileCheck %s --check-prefixes=X86
+
+define <4 x float>
RKSimon added a comment.
Possibly rename the x86-64-* test files to *-64 (and *-32 equivalent) so that
the 32/64 bit files are closer together for tracking (and to help avoid bitrot).
Comment at: clang/lib/Headers/immintrin.h:257
+/* FIXME: Change these When _Float16 type
pengfei added inline comments.
Comment at: clang/include/clang/Basic/BuiltinsX86.def:2106-2107
+TARGET_BUILTIN(__builtin_ia32_vcvtneoph2ps256, "V8fV16xC*", "nV:256:",
"avxneconvert")
+TARGET_BUILTIN(__builtin_ia32_vcvtneps2bf16128, "V8sV4f", "nV:128:",
"avxneconvert")
pengfei added inline comments.
Comment at: clang/test/CodeGen/X86/avxneconvert-builtins.c:2
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown
-target-feature +avx2 -target-feature +avxneconvert \
+// RUN: -target-feature +avx512fp16 -emit-llvm -o - -Wall
pengfei added inline comments.
Comment at: clang/lib/Headers/immintrin.h:257
+/* FIXME: Change these When _Float16 type is supported */
+#if defined(__AVXNECONVERT__) && defined(__AVX512FP16__)
FreddyYe wrote:
> pengfei wrote:
> > craig.topper wrote:
> > > Is
RKSimon added a comment.
merge att/intel test coverage files and rename the 32/64 bit files so that they
are close together in the file lists
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D135930/new/
https://reviews.llvm.org/D135930
RKSimon added inline comments.
Comment at: llvm/test/MC/X86/avx-ne-convert-att.s:1
+// RUN: llvm-mc -triple i686-unknown-unknown --show-encoding %s | FileCheck %s
+
merge the att + intel test files and use --check-prefixes to test both
Repository:
rG LLVM
FreddyYe marked 5 inline comments as done.
FreddyYe added a comment.
THX for reviews!
Comment at: clang/lib/Headers/immintrin.h:257
+/* FIXME: Change these When _Float16 type is supported */
+#if defined(__AVXNECONVERT__) && defined(__AVX512FP16__)
pengfei
RKSimon added inline comments.
Comment at: clang/test/CodeGen/X86/avxneconvert-builtins.c:2
+// RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown
-target-feature +avx2 -target-feature +avxneconvert \
+// RUN: -target-feature +avx512fp16 -emit-llvm -o - -Wall
pengfei added inline comments.
Comment at: clang/lib/Headers/avxneconvertintrin.h:47
+static __inline__ __m128 __DEFAULT_FN_ATTRS128
+_mm_cvtneebf16_ps(const __m128bh *__A) {
+ return (__m128)__builtin_ia32_vcvtneebf162ps128((const __v8hi *)__A);
I think the
pengfei added inline comments.
Comment at: clang/lib/Basic/Targets/X86.cpp:781
+Builder.defineMacro("__AVXNECONVERT__");
+ Builder.defineMacro("__AVXNECONVERT_SUPPORTED__");
if (HasAVXVNNI)
LuoYuanke wrote:
> Do we need it here?
We don't need it.
craig.topper added inline comments.
Comment at: clang/lib/Headers/immintrin.h:257
+/* FIXME: Change these When _Float16 type is supported */
+#if defined(__AVXNECONVERT__) && defined(__AVX512FP16__)
Is this FIXME still relevant? Don't we support _Float16 with
LuoYuanke added inline comments.
Comment at: clang/lib/Basic/Targets/X86.cpp:781
+Builder.defineMacro("__AVXNECONVERT__");
+ Builder.defineMacro("__AVXNECONVERT_SUPPORTED__");
if (HasAVXVNNI)
Do we need it here?
Repository:
rG LLVM Github Monorepo
20 matches
Mail list logo
