From b410108ca3b40ebba4a551a95b9a1159fac595f5 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@gmail.com>
Date: Wed, 18 Jan 2017 02:17:10 +0000
Subject: [PATCH] [AVX-512] Replace subvector broadcast builtins with
shufflevectors and selects.
Verified that the backend codegens this equally well.
git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@292329 91177308-0d34-0410-b5e6-96231b3b80d8
---
include/clang/Basic/BuiltinsX86.def | 12 ----
lib/Headers/avx512dqintrin.h | 95 +++++++++++++-------------
lib/Headers/avx512fintrin.h | 102 +++++++++++++---------------
lib/Headers/avx512vldqintrin.h | 42 ++++++------
lib/Headers/avx512vlintrin.h | 45 ++++++------
test/CodeGen/avx512dq-builtins.c | 81 ++++++++++++----------
test/CodeGen/avx512f-builtins.c | 80 ++++++++++++----------
test/CodeGen/avx512vl-builtins.c | 28 ++++----
test/CodeGen/avx512vldq-builtins.c | 40 ++++++-----
9 files changed, 260 insertions(+), 265 deletions(-)
diff --git a/include/clang/Basic/BuiltinsX86.def b/include/clang/Basic/BuiltinsX86.def
index e53992bbf50..62d4c33a9e1 100644
--- a/include/clang/Basic/BuiltinsX86.def
+++ b/include/clang/Basic/BuiltinsX86.def
@@ -1590,27 +1590,15 @@ TARGET_BUILTIN(__builtin_ia32_cvtq2mask128, "UcV2LLi","","avx512dq,avx512vl")
TARGET_BUILTIN(__builtin_ia32_cvtq2mask256, "UcV4LLi","","avx512dq,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcastmb512, "V8LLiUc","","avx512cd")
TARGET_BUILTIN(__builtin_ia32_broadcastmw512, "V16iUs","","avx512cd")
-TARGET_BUILTIN(__builtin_ia32_broadcastf32x4_512, "V16fV4fV16fUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_broadcastf64x4_512, "V8dV4dV8dUc","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_broadcasti32x4_512, "V16iV4iV16iUs","","avx512f")
-TARGET_BUILTIN(__builtin_ia32_broadcasti64x4_512, "V8LLiV4LLiV8LLiUc","","avx512f")
TARGET_BUILTIN(__builtin_ia32_broadcastmb128, "V2LLiUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcastmb256, "V4LLiUc","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcastmw128, "V4iUs","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcastmw256, "V8iUs","","avx512cd,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_512_mask, "V16fV4fV16fUs","","avx512dq")
-TARGET_BUILTIN(__builtin_ia32_broadcastf32x8_512_mask, "V16fV8fV16fUs","","avx512dq")
-TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_512_mask, "V8dV2dV8dUc","","avx512dq")
TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_512_mask, "V16iV4iV16iUs","","avx512dq")
-TARGET_BUILTIN(__builtin_ia32_broadcasti32x8_512_mask, "V16iV8iV16iUs","","avx512dq")
-TARGET_BUILTIN(__builtin_ia32_broadcasti64x2_512_mask, "V8LLiV2LLiV8LLiUc","","avx512dq")
TARGET_BUILTIN(__builtin_ia32_broadcastf32x2_256_mask, "V8fV4fV8fUc","","avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastf64x2_256_mask, "V4dV2dV4dUc","","avx512dq,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_128_mask, "V4iV4iV4iUc","","avx512dq,avx512vl")
TARGET_BUILTIN(__builtin_ia32_broadcasti32x2_256_mask, "V8iV4iV8iUc","","avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcasti64x2_256_mask, "V4LLiV2LLiV4LLiUc","","avx512dq,avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcastf32x4_256_mask, "V8fV4fV8fUc","","avx512vl")
-TARGET_BUILTIN(__builtin_ia32_broadcasti32x4_256_mask, "V8iV4iV8iUc","","avx512vl")
TARGET_BUILTIN(__builtin_ia32_pbroadcastw512_gpr_mask, "V32shV32sUi","","avx512bw")
TARGET_BUILTIN(__builtin_ia32_pbroadcastw256_gpr_mask, "V16shV16sUs","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_pbroadcastw128_gpr_mask, "V8ssV8sUc","","avx512bw,avx512vl")
diff --git a/lib/Headers/avx512dqintrin.h b/lib/Headers/avx512dqintrin.h
index ae44b98a949..4fd1add7735 100644
--- a/lib/Headers/avx512dqintrin.h
+++ b/lib/Headers/avx512dqintrin.h
@@ -995,51 +995,50 @@ _mm512_maskz_broadcast_f32x2 (__mmask16 __M, __m128 __A)
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_broadcast_f32x8 (__m256 __A)
+_mm512_broadcast_f32x8(__m256 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
- _mm512_undefined_ps(),
- (__mmask16) -1);
+ return (__m512)__builtin_shufflevector((__v8sf)__A, (__v8sf)__A,
+ 0, 1, 2, 3, 4, 5, 6, 7,
+ 0, 1, 2, 3, 4, 5, 6, 7);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f32x8 (__m512 __O, __mmask16 __M, __m256 __A)
+_mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
- (__v16sf)__O,
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask8)__M,
+ (__v16sf)_mm512_broadcast_f32x8(__A),
+ (__v16sf)__O);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f32x8 (__mmask16 __M, __m256 __A)
+_mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x8_512_mask ((__v8sf) __A,
- (__v16sf)_mm512_setzero_ps (),
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask8)__M,
+ (__v16sf)_mm512_broadcast_f32x8(__A),
+ (__v16sf)_mm512_setzero_ps());
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_broadcast_f64x2 (__m128d __A)
+_mm512_broadcast_f64x2(__m128d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
- (__v8df)_mm512_undefined_pd(),
- (__mmask8) -1);
+ return (__m512d)__builtin_shufflevector((__v2df)__A, (__v2df)__A,
+ 0, 1, 0, 1, 0, 1, 0, 1);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f64x2 (__m512d __O, __mmask8 __M, __m128d __A)
+_mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
- (__v8df)
- __O, __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x2(__A),
+ (__v8df)__O);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A)
+_mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x2_512_mask ((__v2df) __A,
- (__v8df)_mm512_setzero_ps (),
- __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x2(__A),
+ (__v8df)_mm512_setzero_pd());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
@@ -1067,52 +1066,50 @@ _mm512_maskz_broadcast_i32x2 (__mmask16 __M, __m128i __A)
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i32x8 (__m256i __A)
+_mm512_broadcast_i32x8(__m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
- (__v16si)_mm512_setzero_si512(),
- (__mmask16) -1);
+ return (__m512i)__builtin_shufflevector((__v8si)__A, (__v8si)__A,
+ 0, 1, 2, 3, 4, 5, 6, 7,
+ 0, 1, 2, 3, 4, 5, 6, 7);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i32x8 (__m512i __O, __mmask16 __M, __m256i __A)
+_mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
- (__v16si)__O,
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask8)__M,
+ (__v16si)_mm512_broadcast_i32x8(__A),
+ (__v16si)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i32x8 (__mmask16 __M, __m256i __A)
+_mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x8_512_mask ((__v8si) __A,
- (__v16si)
- _mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask8)__M,
+ (__v16si)_mm512_broadcast_i32x8(__A),
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i64x2 (__m128i __A)
+_mm512_broadcast_i64x2(__m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
- (__v8di)_mm512_setzero_si512(),
- (__mmask8) -1);
+ return (__m512i)__builtin_shufflevector((__v2di)__A, (__v2di)__A,
+ 0, 1, 0, 1, 0, 1, 0, 1);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i64x2 (__m512i __O, __mmask8 __M, __m128i __A)
+_mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
- (__v8di)
- __O, __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x2(__A),
+ (__v8di)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
+_mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x2_512_mask ((__v2di) __A,
- (__v8di)_mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x2(__A),
+ (__v8di)_mm512_setzero_si512());
}
#define _mm512_extractf32x8_ps(A, imm) __extension__ ({ \
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index e6a7217c896..7d9a52fdf23 100644
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -7278,107 +7278,97 @@ _mm_maskz_sqrt_ss (__mmask8 __U, __m128 __A, __m128 __B)
(__mmask8)(U), (int)(R)); })
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_broadcast_f32x4 (__m128 __A)
+_mm512_broadcast_f32x4(__m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf)
- _mm512_undefined_ps (),
- (__mmask16) -1);
+ return (__m512)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f32x4 (__m512 __O, __mmask16 __M, __m128 __A)
+_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, __m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf) __O,
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+ (__v16sf)_mm512_broadcast_f32x4(__A),
+ (__v16sf)__O);
}
static __inline__ __m512 __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f32x4 (__mmask16 __M, __m128 __A)
+_mm512_maskz_broadcast_f32x4(__mmask16 __M, __m128 __A)
{
- return (__m512) __builtin_ia32_broadcastf32x4_512 ((__v4sf) __A,
- (__v16sf)
- _mm512_setzero_ps (),
- __M);
+ return (__m512)__builtin_ia32_selectps_512((__mmask16)__M,
+ (__v16sf)_mm512_broadcast_f32x4(__A),
+ (__v16sf)_mm512_setzero_ps());
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_broadcast_f64x4 (__m256d __A)
+_mm512_broadcast_f64x4(__m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df)
- _mm512_undefined_pd (),
- (__mmask8) -1);
+ return (__m512d)__builtin_shufflevector((__v4df)__A, (__v4df)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_f64x4 (__m512d __O, __mmask8 __M, __m256d __A)
+_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, __m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df) __O,
- __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x4(__A),
+ (__v8df)__O);
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_f64x4 (__mmask8 __M, __m256d __A)
+_mm512_maskz_broadcast_f64x4(__mmask8 __M, __m256d __A)
{
- return (__m512d) __builtin_ia32_broadcastf64x4_512 ((__v4df) __A,
- (__v8df)
- _mm512_setzero_pd (),
- __M);
+ return (__m512d)__builtin_ia32_selectpd_512((__mmask8)__M,
+ (__v8df)_mm512_broadcast_f64x4(__A),
+ (__v8df)_mm512_setzero_pd());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i32x4 (__m128i __A)
+_mm512_broadcast_i32x4(__m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si)
- _mm512_undefined_epi32 (),
- (__mmask16) -1);
+ return (__m512i)__builtin_shufflevector((__v4si)__A, (__v4si)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i32x4 (__m512i __O, __mmask16 __M, __m128i __A)
+_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si) __O,
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+ (__v16si)_mm512_broadcast_i32x4(__A),
+ (__v16si)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i32x4 (__mmask16 __M, __m128i __A)
+_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i __A)
{
- return (__m512i) __builtin_ia32_broadcasti32x4_512 ((__v4si) __A,
- (__v16si)
- _mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectd_512((__mmask16)__M,
+ (__v16si)_mm512_broadcast_i32x4(__A),
+ (__v16si)_mm512_setzero_si512());
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_broadcast_i64x4 (__m256i __A)
+_mm512_broadcast_i64x4(__m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di)
- _mm512_undefined_epi32 (),
- (__mmask8) -1);
+ return (__m512i)__builtin_shufflevector((__v4di)__A, (__v4di)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_mask_broadcast_i64x4 (__m512i __O, __mmask8 __M, __m256i __A)
+_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di) __O,
- __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x4(__A),
+ (__v8di)__O);
}
static __inline__ __m512i __DEFAULT_FN_ATTRS
-_mm512_maskz_broadcast_i64x4 (__mmask8 __M, __m256i __A)
+_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i __A)
{
- return (__m512i) __builtin_ia32_broadcasti64x4_512 ((__v4di) __A,
- (__v8di)
- _mm512_setzero_si512 (),
- __M);
+ return (__m512i)__builtin_ia32_selectq_512((__mmask8)__M,
+ (__v8di)_mm512_broadcast_i64x4(__A),
+ (__v8di)_mm512_setzero_si512());
}
static __inline__ __m512d __DEFAULT_FN_ATTRS
diff --git a/lib/Headers/avx512vldqintrin.h b/lib/Headers/avx512vldqintrin.h
index cd9da437056..aecd7df34d0 100644
--- a/lib/Headers/avx512vldqintrin.h
+++ b/lib/Headers/avx512vldqintrin.h
@@ -1000,27 +1000,26 @@ _mm256_maskz_broadcast_f32x2 (__mmask8 __M, __m128 __A)
}
static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_broadcast_f64x2 (__m128d __A)
+_mm256_broadcast_f64x2(__m128d __A)
{
- return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df) __A,
- (__v4df)_mm256_undefined_pd(),
- (__mmask8) -1);
+ return (__m256d)__builtin_shufflevector((__v2df)__A, (__v2df)__A,
+ 0, 1, 0, 1);
}
static __inline__ __m256d __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_f64x2 (__m256d __O, __mmask8 __M, __m128d __A)
+_mm256_mask_broadcast_f64x2(__m256d __O, __mmask8 __M, __m128d __A)
{
- return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df) __A,
- (__v4df) __O,
- __M);
+ return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__M,
+ (__v4df)_mm256_broadcast_f64x2(__A),
+ (__v4df)__O);
}
static __inline__ __m256d __DEFAULT_FN_ATTRS
_mm256_maskz_broadcast_f64x2 (__mmask8 __M, __m128d __A)
{
- return (__m256d) __builtin_ia32_broadcastf64x2_256_mask ((__v2df) __A,
- (__v4df) _mm256_setzero_ps (),
- __M);
+ return (__m256d)__builtin_ia32_selectpd_256((__mmask8)__M,
+ (__v4df)_mm256_broadcast_f64x2(__A),
+ (__v4df)_mm256_setzero_pd());
}
static __inline__ __m128i __DEFAULT_FN_ATTRS
@@ -1072,27 +1071,26 @@ _mm256_maskz_broadcast_i32x2 (__mmask8 __M, __m128i __A)
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_broadcast_i64x2 (__m128i __A)
+_mm256_broadcast_i64x2(__m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di) __A,
- (__v4di)_mm256_undefined_si256(),
- (__mmask8) -1);
+ return (__m256i)__builtin_shufflevector((__v2di)__A, (__v2di)__A,
+ 0, 1, 0, 1);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_i64x2 (__m256i __O, __mmask8 __M, __m128i __A)
+_mm256_mask_broadcast_i64x2(__m256i __O, __mmask8 __M, __m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di) __A,
- (__v4di) __O,
- __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_broadcast_i64x2(__A),
+ (__v4di)__O);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
_mm256_maskz_broadcast_i64x2 (__mmask8 __M, __m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti64x2_256_mask ((__v2di) __A,
- (__v4di) _mm256_setzero_si256 (),
- __M);
+ return (__m256i)__builtin_ia32_selectq_256((__mmask8)__M,
+ (__v4di)_mm256_broadcast_i64x2(__A),
+ (__v4di)_mm256_setzero_si256());
}
#define _mm256_extractf64x2_pd(A, imm) __extension__ ({ \
diff --git a/lib/Headers/avx512vlintrin.h b/lib/Headers/avx512vlintrin.h
index f3744da6ab8..99bb050de4d 100644
--- a/lib/Headers/avx512vlintrin.h
+++ b/lib/Headers/avx512vlintrin.h
@@ -7189,52 +7189,49 @@ _mm256_maskz_rsqrt14_ps (__mmask8 __U, __m256 __A)
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_broadcast_f32x4 (__m128 __A)
+_mm256_broadcast_f32x4(__m128 __A)
{
- return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
- (__v8sf)_mm256_undefined_pd (),
- (__mmask8) -1);
+ return (__m256)__builtin_shufflevector((__v4sf)__A, (__v4sf)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_f32x4 (__m256 __O, __mmask8 __M, __m128 __A)
+_mm256_mask_broadcast_f32x4(__m256 __O, __mmask8 __M, __m128 __A)
{
- return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
- (__v8sf) __O,
- __M);
+ return (__m256)__builtin_ia32_selectps_256((__mmask8)__M,
+ (__v8sf)_mm256_broadcast_f32x4(__A),
+ (__v8sf)__O);
}
static __inline__ __m256 __DEFAULT_FN_ATTRS
_mm256_maskz_broadcast_f32x4 (__mmask8 __M, __m128 __A)
{
- return (__m256) __builtin_ia32_broadcastf32x4_256_mask ((__v4sf) __A,
- (__v8sf) _mm256_setzero_ps (),
- __M);
+ return (__m256)__builtin_ia32_selectps_256((__mmask8)__M,
+ (__v8sf)_mm256_broadcast_f32x4(__A),
+ (__v8sf)_mm256_setzero_ps());
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_broadcast_i32x4 (__m128i __A)
+_mm256_broadcast_i32x4(__m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si) __A,
- (__v8si)_mm256_undefined_si256 (),
- (__mmask8) -1);
+ return (__m256i)__builtin_shufflevector((__v4si)__A, (__v4si)__A,
+ 0, 1, 2, 3, 0, 1, 2, 3);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_mask_broadcast_i32x4 (__m256i __O, __mmask8 __M, __m128i __A)
+_mm256_mask_broadcast_i32x4(__m256i __O, __mmask8 __M, __m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si) __A,
- (__v8si)
- __O, __M);
+ return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+ (__v8si)_mm256_broadcast_i32x4(__A),
+ (__v8si)__O);
}
static __inline__ __m256i __DEFAULT_FN_ATTRS
-_mm256_maskz_broadcast_i32x4 (__mmask8 __M, __m128i __A)
+_mm256_maskz_broadcast_i32x4(__mmask8 __M, __m128i __A)
{
- return (__m256i) __builtin_ia32_broadcasti32x4_256_mask ((__v4si)
- __A,
- (__v8si) _mm256_setzero_si256 (),
- __M);
+ return (__m256i)__builtin_ia32_selectd_256((__mmask8)__M,
+ (__v8si)_mm256_broadcast_i32x4(__A),
+ (__v8si)_mm256_setzero_si256());
}
static __inline__ __m256d __DEFAULT_FN_ATTRS
diff --git a/test/CodeGen/avx512dq-builtins.c b/test/CodeGen/avx512dq-builtins.c
index f57433a3616..c0e1d648098 100644
--- a/test/CodeGen/avx512dq-builtins.c
+++ b/test/CodeGen/avx512dq-builtins.c
@@ -963,40 +963,44 @@ __m512 test_mm512_maskz_broadcast_f32x2(__mmask16 __M, __m128 __A) {
return _mm512_maskz_broadcast_f32x2(__M, __A);
}
-__m512 test_mm512_broadcast_f32x8(__m256 __A) {
+__m512 test_mm512_broadcast_f32x8(float const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_f32x8
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
- return _mm512_broadcast_f32x8(__A);
+ // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ return _mm512_broadcast_f32x8(_mm256_loadu_ps(__A));
}
-__m512 test_mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, __m256 __A) {
+__m512 test_mm512_mask_broadcast_f32x8(__m512 __O, __mmask16 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_f32x8
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
- return _mm512_mask_broadcast_f32x8(__O, __M, __A);
+ // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
+ return _mm512_mask_broadcast_f32x8(__O, __M, _mm256_loadu_ps(__A));
}
-__m512 test_mm512_maskz_broadcast_f32x8(__mmask16 __M, __m256 __A) {
+__m512 test_mm512_maskz_broadcast_f32x8(__mmask16 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_f32x8
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x8
- return _mm512_maskz_broadcast_f32x8(__M, __A);
+ // CHECK: shufflevector <8 x float> %{{.*}}, <8 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
+ return _mm512_maskz_broadcast_f32x8(__M, _mm256_loadu_ps(__A));
}
-__m512d test_mm512_broadcast_f64x2(__m128d __A) {
+__m512d test_mm512_broadcast_f64x2(double const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm512_broadcast_f64x2(__A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ return _mm512_broadcast_f64x2(_mm_loadu_pd(__A));
}
-__m512d test_mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, __m128d __A) {
+__m512d test_mm512_mask_broadcast_f64x2(__m512d __O, __mmask8 __M, double const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm512_mask_broadcast_f64x2(__O, __M, __A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
+ return _mm512_mask_broadcast_f64x2(__O, __M, _mm_loadu_pd(__A));
}
-__m512d test_mm512_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) {
+__m512d test_mm512_maskz_broadcast_f64x2(__mmask8 __M, double const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm512_maskz_broadcast_f64x2(__M, __A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
+ return _mm512_maskz_broadcast_f64x2(__M, _mm_loadu_pd(__A));
}
__m512i test_mm512_broadcast_i32x2(__m128i __A) {
@@ -1017,41 +1021,46 @@ __m512i test_mm512_maskz_broadcast_i32x2(__mmask16 __M, __m128i __A) {
return _mm512_maskz_broadcast_i32x2(__M, __A);
}
-__m512i test_mm512_broadcast_i32x8(__m256i __A) {
+__m512i test_mm512_broadcast_i32x8(__m256i const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_i32x8
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
- return _mm512_broadcast_i32x8(__A);
+ // CHECK: shufflevector <8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ return _mm512_broadcast_i32x8(_mm256_loadu_si256(__A));
}
-__m512i test_mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i __A) {
+__m512i test_mm512_mask_broadcast_i32x8(__m512i __O, __mmask16 __M, __m256i const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_i32x8
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
- return _mm512_mask_broadcast_i32x8(__O, __M, __A);
+ // CHECK: shufflevector <8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
+ return _mm512_mask_broadcast_i32x8(__O, __M, _mm256_loadu_si256(__A));
}
-__m512i test_mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i __A) {
+__m512i test_mm512_maskz_broadcast_i32x8(__mmask16 __M, __m256i const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_i32x8
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x8
- return _mm512_maskz_broadcast_i32x8(__M, __A);
+ // CHECK: shufflevector <8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
+ return _mm512_maskz_broadcast_i32x8(__M, _mm256_loadu_si256(__A));
}
-__m512i test_mm512_broadcast_i64x2(__m128i __A) {
+__m512i test_mm512_broadcast_i64x2(__m128i const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm512_broadcast_i64x2(__A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ return _mm512_broadcast_i64x2(_mm_loadu_si128(__A));
}
-__m512i test_mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i __A) {
+__m512i test_mm512_mask_broadcast_i64x2(__m512i __O, __mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm512_mask_broadcast_i64x2(__O, __M, __A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
+ return _mm512_mask_broadcast_i64x2(__O, __M, _mm_loadu_si128(__A));
}
-__m512i test_mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) {
+__m512i test_mm512_maskz_broadcast_i64x2(__mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm512_maskz_broadcast_i64x2(__M, __A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
+ return _mm512_maskz_broadcast_i64x2(__M, _mm_loadu_si128(__A));
}
+
__m256 test_mm512_extractf32x8_ps(__m512 __A) {
// CHECK-LABEL: @test_mm512_extractf32x8_ps
// CHECK: shufflevector <16 x float> %{{.*}}, <16 x float> undef, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index 760783af1ce..5b78dd2ffe7 100644
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -4647,76 +4647,84 @@ __m128 test_mm_maskz_sqrt_round_ss(__mmask8 __U, __m128 __A, __m128 __B){
return _mm_maskz_sqrt_round_ss(__U,__A,__B,_MM_FROUND_CUR_DIRECTION);
}
-__m512 test_mm512_broadcast_f32x4(__m128 __A) {
+__m512 test_mm512_broadcast_f32x4(float const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
- return _mm512_broadcast_f32x4(__A);
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ return _mm512_broadcast_f32x4(_mm_loadu_ps(__A));
}
-__m512 test_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, __m128 __A) {
+__m512 test_mm512_mask_broadcast_f32x4(__m512 __O, __mmask16 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
- return _mm512_mask_broadcast_f32x4(__O, __M, __A);
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
+ return _mm512_mask_broadcast_f32x4(__O, __M, _mm_loadu_ps(__A));
}
-__m512 test_mm512_maskz_broadcast_f32x4(__mmask16 __M, __m128 __A) {
+__m512 test_mm512_maskz_broadcast_f32x4(__mmask16 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
- return _mm512_maskz_broadcast_f32x4(__M, __A);
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
+ return _mm512_maskz_broadcast_f32x4(__M, _mm_loadu_ps(__A));
}
-__m512d test_mm512_broadcast_f64x4(__m256d __A) {
+__m512d test_mm512_broadcast_f64x4(float const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_f64x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
- return _mm512_broadcast_f64x4(__A);
+ // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ return _mm512_broadcast_f64x4(_mm256_loadu_ps(__A));
}
-__m512d test_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, __m256d __A) {
+__m512d test_mm512_mask_broadcast_f64x4(__m512d __O, __mmask8 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_f64x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
- return _mm512_mask_broadcast_f64x4(__O, __M, __A);
+ // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
+ return _mm512_mask_broadcast_f64x4(__O, __M, _mm256_loadu_ps(__A));
}
-__m512d test_mm512_maskz_broadcast_f64x4(__mmask8 __M, __m256d __A) {
+__m512d test_mm512_maskz_broadcast_f64x4(__mmask8 __M, float const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_f64x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x4
- return _mm512_maskz_broadcast_f64x4(__M, __A);
+ // CHECK: shufflevector <4 x double> %{{.*}}, <4 x double> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
+ return _mm512_maskz_broadcast_f64x4(__M, _mm256_loadu_ps(__A));
}
-__m512i test_mm512_broadcast_i32x4(__m128i __A) {
+__m512i test_mm512_broadcast_i32x4(__m128i const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm512_broadcast_i32x4(__A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ return _mm512_broadcast_i32x4(_mm_loadu_si128(__A));
}
-__m512i test_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i __A) {
+__m512i test_mm512_mask_broadcast_i32x4(__m512i __O, __mmask16 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm512_mask_broadcast_i32x4(__O, __M, __A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
+ return _mm512_mask_broadcast_i32x4(__O, __M, _mm_loadu_si128(__A));
}
-__m512i test_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i __A) {
+__m512i test_mm512_maskz_broadcast_i32x4(__mmask16 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm512_maskz_broadcast_i32x4(__M, __A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
+ return _mm512_maskz_broadcast_i32x4(__M, _mm_loadu_si128(__A));
}
-__m512i test_mm512_broadcast_i64x4(__m256i __A) {
+__m512i test_mm512_broadcast_i64x4(__m256i const* __A) {
// CHECK-LABEL: @test_mm512_broadcast_i64x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
- return _mm512_broadcast_i64x4(__A);
+ // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ return _mm512_broadcast_i64x4(_mm256_loadu_si256(__A));
}
-__m512i test_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i __A) {
+__m512i test_mm512_mask_broadcast_i64x4(__m512i __O, __mmask8 __M, __m256i const* __A) {
// CHECK-LABEL: @test_mm512_mask_broadcast_i64x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
- return _mm512_mask_broadcast_i64x4(__O, __M, __A);
+ // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
+ return _mm512_mask_broadcast_i64x4(__O, __M, _mm256_loadu_si256(__A));
}
-__m512i test_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i __A) {
+__m512i test_mm512_maskz_broadcast_i64x4(__mmask8 __M, __m256i const* __A) {
// CHECK-LABEL: @test_mm512_maskz_broadcast_i64x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x4
- return _mm512_maskz_broadcast_i64x4(__M, __A);
+ // CHECK: shufflevector <4 x i64> %{{.*}}, <4 x i64> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i64> %{{.*}}, <8 x i64> %{{.*}}
+ return _mm512_maskz_broadcast_i64x4(__M, _mm256_loadu_si256(__A));
}
__m512d test_mm512_broadcastsd_pd(__m128d __A) {
diff --git a/test/CodeGen/avx512vl-builtins.c b/test/CodeGen/avx512vl-builtins.c
index fe4ebe12ddd..49c7ba0d82d 100644
--- a/test/CodeGen/avx512vl-builtins.c
+++ b/test/CodeGen/avx512vl-builtins.c
@@ -5734,38 +5734,42 @@ __m256 test_mm256_maskz_rsqrt14_ps(__mmask8 __U, __m256 __A) {
__m256 test_mm256_broadcast_f32x4(__m128 __A) {
// CHECK-LABEL: @test_mm256_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
return _mm256_broadcast_f32x4(__A);
}
__m256 test_mm256_mask_broadcast_f32x4(__m256 __O, __mmask8 __M, __m128 __A) {
// CHECK-LABEL: @test_mm256_mask_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
return _mm256_mask_broadcast_f32x4(__O, __M, __A);
}
__m256 test_mm256_maskz_broadcast_f32x4(__mmask8 __M, __m128 __A) {
// CHECK-LABEL: @test_mm256_maskz_broadcast_f32x4
- // CHECK: @llvm.x86.avx512.mask.broadcastf32x4
+ // CHECK: shufflevector <4 x float> %{{.*}}, <4 x float> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
return _mm256_maskz_broadcast_f32x4(__M, __A);
}
-__m256i test_mm256_broadcast_i32x4(__m128i __A) {
+__m256i test_mm256_broadcast_i32x4(__m128i const* __A) {
// CHECK-LABEL: @test_mm256_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm256_broadcast_i32x4(__A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ return _mm256_broadcast_i32x4(_mm_loadu_si128(__A));
}
-__m256i test_mm256_mask_broadcast_i32x4(__m256i __O, __mmask8 __M, __m128i __A) {
+__m256i test_mm256_mask_broadcast_i32x4(__m256i __O, __mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm256_mask_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm256_mask_broadcast_i32x4(__O, __M, __A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
+ return _mm256_mask_broadcast_i32x4(__O, __M, _mm_loadu_si128(__A));
}
-__m256i test_mm256_maskz_broadcast_i32x4(__mmask8 __M, __m128i __A) {
+__m256i test_mm256_maskz_broadcast_i32x4(__mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm256_maskz_broadcast_i32x4
- // CHECK: @llvm.x86.avx512.mask.broadcasti32x4
- return _mm256_maskz_broadcast_i32x4(__M, __A);
+ // CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
+ // CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
+ return _mm256_maskz_broadcast_i32x4(__M, _mm_loadu_si128(__A));
}
__m256d test_mm256_mask_broadcastsd_pd(__m256d __O, __mmask8 __M, __m128d __A) {
diff --git a/test/CodeGen/avx512vldq-builtins.c b/test/CodeGen/avx512vldq-builtins.c
index 6834b6f46b2..fac61ae4715 100644
--- a/test/CodeGen/avx512vldq-builtins.c
+++ b/test/CodeGen/avx512vldq-builtins.c
@@ -918,22 +918,24 @@ __m256 test_mm256_maskz_broadcast_f32x2(__mmask8 __M, __m128 __A) {
return _mm256_maskz_broadcast_f32x2(__M, __A);
}
-__m256d test_mm256_broadcast_f64x2(__m128d __A) {
+__m256d test_mm256_broadcast_f64x2(double const* __A) {
// CHECK-LABEL: @test_mm256_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm256_broadcast_f64x2(__A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ return _mm256_broadcast_f64x2(_mm_loadu_pd(__A));
}
-__m256d test_mm256_mask_broadcast_f64x2(__m256d __O, __mmask8 __M, __m128d __A) {
+__m256d test_mm256_mask_broadcast_f64x2(__m256d __O, __mmask8 __M, double const* __A) {
// CHECK-LABEL: @test_mm256_mask_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm256_mask_broadcast_f64x2(__O, __M, __A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
+ return _mm256_mask_broadcast_f64x2(__O, __M, _mm_loadu_pd(__A));
}
-__m256d test_mm256_maskz_broadcast_f64x2(__mmask8 __M, __m128d __A) {
+__m256d test_mm256_maskz_broadcast_f64x2(__mmask8 __M, double const* __A) {
// CHECK-LABEL: @test_mm256_maskz_broadcast_f64x2
- // CHECK: @llvm.x86.avx512.mask.broadcastf64x2
- return _mm256_maskz_broadcast_f64x2(__M, __A);
+ // CHECK: shufflevector <2 x double> %{{.*}}, <2 x double> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <4 x i1> %{{.*}}, <4 x double> %{{.*}}, <4 x double> %{{.*}}
+ return _mm256_maskz_broadcast_f64x2(__M, _mm_loadu_pd(__A));
}
__m128i test_mm_broadcast_i32x2(__m128i __A) {
@@ -972,22 +974,24 @@ __m256i test_mm256_maskz_broadcast_i32x2(__mmask8 __M, __m128i __A) {
return _mm256_maskz_broadcast_i32x2(__M, __A);
}
-__m256i test_mm256_broadcast_i64x2(__m128i __A) {
+__m256i test_mm256_broadcast_i64x2(__m128i const* __A) {
// CHECK-LABEL: @test_mm256_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm256_broadcast_i64x2(__A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ return _mm256_broadcast_i64x2(_mm_loadu_si128(__A));
}
-__m256i test_mm256_mask_broadcast_i64x2(__m256i __O, __mmask8 __M, __m128i __A) {
+__m256i test_mm256_mask_broadcast_i64x2(__m256i __O, __mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm256_mask_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm256_mask_broadcast_i64x2(__O, __M, __A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
+ return _mm256_mask_broadcast_i64x2(__O, __M, _mm_loadu_si128(__A));
}
-__m256i test_mm256_maskz_broadcast_i64x2(__mmask8 __M, __m128i __A) {
+__m256i test_mm256_maskz_broadcast_i64x2(__mmask8 __M, __m128i const* __A) {
// CHECK-LABEL: @test_mm256_maskz_broadcast_i64x2
- // CHECK: @llvm.x86.avx512.mask.broadcasti64x2
- return _mm256_maskz_broadcast_i64x2(__M, __A);
+ // CHECK: shufflevector <2 x i64> %{{.*}}, <2 x i64> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+ // CHECK: select <4 x i1> %{{.*}}, <4 x i64> %{{.*}}, <4 x i64> %{{.*}}
+ return _mm256_maskz_broadcast_i64x2(__M, _mm_loadu_si128(__A));
}
__m128d test_mm256_extractf64x2_pd(__m256d __A) {
--
GitLab