summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMichael Zuckerman <Michael.zuckerman@intel.com>2016-03-03 08:55:20 +0000
committerMichael Zuckerman <Michael.zuckerman@intel.com>2016-03-03 08:55:20 +0000
commitabbe34bce65b74ad23b4304eb180691d8685f6a0 (patch)
treef701b324d619a929c08726be90b57084f1b19387
parentc4d054fa4ac32fae1973c8cebf17b6f4f7f58e02 (diff)
downloadbcm5719-llvm-abbe34bce65b74ad23b4304eb180691d8685f6a0.tar.gz
bcm5719-llvm-abbe34bce65b74ad23b4304eb180691d8685f6a0.zip
[Clang][AVX512][BUILTIN] Adding PSRL{W|WI}{128|256|512}
Differential Revision: http://reviews.llvm.org/D17754 llvm-svn: 262593
-rw-r--r--clang/include/clang/Basic/BuiltinsX86.def6
-rw-r--r--clang/lib/Headers/avx512bwintrin.h51
-rw-r--r--clang/lib/Headers/avx512vlbwintrin.h67
-rw-r--r--clang/test/CodeGen/avx512bw-builtins.c37
4 files changed, 161 insertions, 0 deletions
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
index bfe071a63ee..a872555abe6 100644
--- a/clang/include/clang/Basic/BuiltinsX86.def
+++ b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1691,6 +1691,12 @@ TARGET_BUILTIN(__builtin_ia32_psraw128_mask, "V8sV8sV8sV8sUc","","avx512bw,avx51
TARGET_BUILTIN(__builtin_ia32_psraw256_mask, "V16sV16sV8sV16sUs","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psrawi128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl")
TARGET_BUILTIN(__builtin_ia32_psrawi256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psrlw512_mask, "V32sV32sV8sV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psrlwi512_mask, "V32sV32sIiV32sUi","","avx512bw")
+TARGET_BUILTIN(__builtin_ia32_psrlw128_mask, "V8sV8sV8sV8sUc","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psrlw256_mask, "V16sV16sV8sV16sUs","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psrlwi128_mask, "V8sV8sIiV8sUc","","avx512bw,avx512vl")
+TARGET_BUILTIN(__builtin_ia32_psrlwi256_mask, "V16sV16sIiV16sUs","","avx512bw,avx512vl")
#undef BUILTIN
#undef TARGET_BUILTIN
diff --git a/clang/lib/Headers/avx512bwintrin.h b/clang/lib/Headers/avx512bwintrin.h
index 7f3f1bf4f02..17ca52ad4d0 100644
--- a/clang/lib/Headers/avx512bwintrin.h
+++ b/clang/lib/Headers/avx512bwintrin.h
@@ -1829,6 +1829,57 @@ __builtin_ia32_psrawi512_mask ((__v32hi)( __A),( __B),\
(__mmask32)( __U));\
})
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_srl_epi16 (__m512i __A, __m128i __B)
+{
+ return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
+ (__v8hi) __B,
+ (__v32hi)
+ _mm512_setzero_hi (),
+ (__mmask32) -1);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_mask_srl_epi16 (__m512i __W, __mmask32 __U, __m512i __A,
+ __m128i __B)
+{
+ return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
+ (__v8hi) __B,
+ (__v32hi) __W,
+ (__mmask32) __U);
+}
+
+static __inline__ __m512i __DEFAULT_FN_ATTRS
+_mm512_maskz_srl_epi16 (__mmask32 __U, __m512i __A, __m128i __B)
+{
+ return (__m512i) __builtin_ia32_psrlw512_mask ((__v32hi) __A,
+ (__v8hi) __B,
+ (__v32hi)
+ _mm512_setzero_hi (),
+ (__mmask32) __U);
+}
+
+#define _mm512_srli_epi16( __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\
+ (__v32hi)\
+ _mm512_setzero_hi (),\
+ (__mmask32) -1);\
+})
+
+#define _mm512_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\
+ (__v32hi)( __W),\
+ (__mmask32)( __U));\
+})
+
+#define _mm512_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi512_mask ((__v32hi)( __A),( __imm),\
+ (__v32hi)\
+ _mm512_setzero_hi (),\
+ (__mmask32)( __U));\
+})
+
#undef __DEFAULT_FN_ATTRS
#endif
diff --git a/clang/lib/Headers/avx512vlbwintrin.h b/clang/lib/Headers/avx512vlbwintrin.h
index a4b8ff9a068..0c213bf6e09 100644
--- a/clang/lib/Headers/avx512vlbwintrin.h
+++ b/clang/lib/Headers/avx512vlbwintrin.h
@@ -2781,6 +2781,73 @@ __builtin_ia32_psrawi256_mask ((__v16hi)( __A),( __imm),\
(__mmask16)( __U));\
})
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_mask_srl_epi16 (__m128i __W, __mmask8 __U, __m128i __A,
+ __m128i __B)
+{
+ return (__m128i) __builtin_ia32_psrlw128_mask ((__v8hi) __A,
+ (__v8hi) __B,
+ (__v8hi) __W,
+ (__mmask8) __U);
+}
+
+static __inline__ __m128i __DEFAULT_FN_ATTRS
+_mm_maskz_srl_epi16 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+ return (__m128i) __builtin_ia32_psrlw128_mask ((__v8hi) __A,
+ (__v8hi) __B,
+ (__v8hi)
+ _mm_setzero_si128 (),
+ (__mmask8) __U);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_mask_srl_epi16 (__m256i __W, __mmask16 __U, __m256i __A,
+ __m128i __B)
+{
+ return (__m256i) __builtin_ia32_psrlw256_mask ((__v16hi) __A,
+ (__v8hi) __B,
+ (__v16hi) __W,
+ (__mmask16) __U);
+}
+
+static __inline__ __m256i __DEFAULT_FN_ATTRS
+_mm256_maskz_srl_epi16 (__mmask16 __U, __m256i __A, __m128i __B)
+{
+ return (__m256i) __builtin_ia32_psrlw256_mask ((__v16hi) __A,
+ (__v8hi) __B,
+ (__v16hi)
+ _mm256_setzero_si256 (),
+ (__mmask16) __U);
+}
+
+#define _mm_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi128_mask ((__v8hi)( __A),( __imm),\
+ (__v8hi)( __W),\
+ (__mmask8)( __U));\
+})
+
+#define _mm_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi128_mask ((__v8hi)( __A),( __imm),\
+ (__v8hi)\
+ _mm_setzero_si128 (),\
+ (__mmask8)( __U));\
+})
+
+#define _mm256_mask_srli_epi16( __W, __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\
+ (__v16hi)( __W),\
+ (__mmask16)( __U));\
+})
+
+#define _mm256_maskz_srli_epi16( __U, __A, __imm) __extension__ ({ \
+__builtin_ia32_psrlwi256_mask ((__v16hi)( __A),( __imm),\
+ (__v16hi)\
+ _mm256_setzero_si256 (),\
+ (__mmask16)( __U));\
+})
+
+
#undef __DEFAULT_FN_ATTRS
#endif /* __AVX512VLBWINTRIN_H */
diff --git a/clang/test/CodeGen/avx512bw-builtins.c b/clang/test/CodeGen/avx512bw-builtins.c
index a1fafe4ce50..77f1007ce99 100644
--- a/clang/test/CodeGen/avx512bw-builtins.c
+++ b/clang/test/CodeGen/avx512bw-builtins.c
@@ -1239,5 +1239,42 @@ __m512i test_mm512_maskz_srai_epi16(__mmask32 __U, __m512i __A) {
return _mm512_maskz_srai_epi16(__U, __A, 5);
}
+__m512i test_mm512_srl_epi16(__m512i __A, __m128i __B) {
+ // CHECK-LABEL: @test_mm512_srl_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ return _mm512_srl_epi16(__A, __B);
+}
+
+__m512i test_mm512_mask_srl_epi16(__m512i __W, __mmask32 __U, __m512i __A, __m128i __B) {
+ // CHECK-LABEL: @test_mm512_mask_srl_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ return _mm512_mask_srl_epi16(__W, __U, __A, __B);
+}
+
+__m512i test_mm512_maskz_srl_epi16(__mmask32 __U, __m512i __A, __m128i __B) {
+ // CHECK-LABEL: @test_mm512_maskz_srl_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.w.512
+ return _mm512_maskz_srl_epi16(__U, __A, __B);
+}
+
+__m512i test_mm512_srli_epi16(__m512i __A) {
+ // CHECK-LABEL: @test_mm512_srli_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ return _mm512_srli_epi16(__A, 5);
+}
+
+__m512i test_mm512_mask_srli_epi16(__m512i __W, __mmask32 __U, __m512i __A) {
+ // CHECK-LABEL: @test_mm512_mask_srli_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ return _mm512_mask_srli_epi16(__W, __U, __A, 5);
+}
+
+__m512i test_mm512_maskz_srli_epi16(__mmask32 __U, __m512i __A) {
+ // CHECK-LABEL: @test_mm512_maskz_srli_epi16
+ // CHECK: @llvm.x86.avx512.mask.psrl.wi.512
+ return _mm512_maskz_srli_epi16(__U, __A, 5);
+}
+
+
OpenPOWER on IntegriCloud