[Clang][BuiltIn][AVX512]Adding intrinsics for cmp{ss|sd} instruction set

cfe/trunk/include/clang/Basic/BuiltinsX86.def

 TARGET_BUILTIN(__builtin_ia32_dbpsadbw128_mask, "V8sV16cV16cIiV8sUc","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_dbpsadbw256_mask, "V16sV32cV32cIiV16sUs","","avx512bw,avx512vl")
 TARGET_BUILTIN(__builtin_ia32_dbpsadbw512_mask, "V32sV64cV64cIiV32sUi","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_psadbw512, "V8LLiV64cV64c","","avx512bw")
 TARGET_BUILTIN(__builtin_ia32_compressdf512_mask, "V8dV8dV8dUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compressdi512_mask, "V8LLiV8LLiV8LLiUc","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compresssf512_mask, "V16fV16fV16fUs","","avx512f")
 TARGET_BUILTIN(__builtin_ia32_compresssi512_mask, "V16iV16iV16iUs","","avx512f")
-
+TARGET_BUILTIN(__builtin_ia32_cmpsd_mask, "UcV2dV2dIiUcIi","","avx512f")
+TARGET_BUILTIN(__builtin_ia32_cmpss_mask, "UcV4fV4fIiUcIi","","avx512f")
 
 #undef BUILTIN
 #undef TARGET_BUILTIN

cfe/trunk/lib/Headers/avx512fintrin.h

 _mm512_maskz_compress_epi32 (__mmask16 __U, __m512i __A)
 {
   return (__m512i) __builtin_ia32_compresssi512_mask ((__v16si) __A,
                   (__v16si)
                   _mm512_setzero_si512 (),
                   (__mmask16) __U);
 }
 
+#define _mm_cmp_round_ss_mask( __X, __Y, __P, __R) __extension__ ({ \
+__builtin_ia32_cmpss_mask ((__v4sf)( __X),\
+                 (__v4sf)( __Y), __P,\
+                 (__mmask8) -1, __R);\
+})
+
+#define _mm_mask_cmp_round_ss_mask( __M, __X, __Y, __P, __R) __extension__ ({ \
+__builtin_ia32_cmpss_mask ((__v4sf)( __X),\
+                 (__v4sf)( __Y), __P,\
+                 (__mmask8)( __M), __R);\
+})
+
+#define _mm_cmp_ss_mask( __X, __Y, __P) __extension__ ({ \
+__builtin_ia32_cmpss_mask ((__v4sf)( __X),\
+                 (__v4sf)( __Y),( __P),\
+                 (__mmask8) -1,\
+                 _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_mask_cmp_ss_mask( __M, __X, __Y, __P) __extension__ ({ \
+__builtin_ia32_cmpss_mask ((__v4sf)( __X),\
+                 (__v4sf)( __Y),( __P),\
+                 (__mmask8)( __M),\
+                 _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_cmp_round_sd_mask( __X, __Y, __P,__R) __extension__ ({ \
+__builtin_ia32_cmpsd_mask ((__v2df)( __X),\
+                 (__v2df)( __Y), __P,\
+                 (__mmask8) -1, __R);\
+})
+
+#define _mm_mask_cmp_round_sd_mask( __M, __X, __Y, __P, __R) __extension__ ({ \
+__builtin_ia32_cmpsd_mask ((__v2df)( __X),\
+                 (__v2df)( __Y), __P,\
+                 (__mmask8)( __M), __R);\
+})
+
+#define _mm_cmp_sd_mask( __X, __Y, __P) __extension__ ({ \
+__builtin_ia32_cmpsd_mask ((__v2df)( __X),\
+                 (__v2df)( __Y),( __P),\
+                 (__mmask8) -1,\
+                 _MM_FROUND_CUR_DIRECTION);\
+})
+
+#define _mm_mask_cmp_sd_mask( __M, __X, __Y, __P) __extension__ ({ \
+__builtin_ia32_cmpsd_mask ((__v2df)( __X),\
+                 (__v2df)( __Y),( __P),\
+                 (__mmask8)( __M),\
+                 _MM_FROUND_CUR_DIRECTION);\
+})
+
 #undef __DEFAULT_FN_ATTRS
 
 #endif // __AVX512FINTRIN_H

cfe/trunk/test/CodeGen/avx512f-builtins.c

__m512i test_mm512_mask_compress_epi32(__m512i __W, __mmask16 __U, __m512i __A) {
   return _mm512_mask_compress_epi32(__W, __U, __A); 
 }
 
 __m512i test_mm512_maskz_compress_epi32(__mmask16 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_compress_epi32
   // CHECK: @llvm.x86.avx512.mask.compress.d.512
   return _mm512_maskz_compress_epi32(__U, __A); 
 }
+
+__mmask8 test_mm_cmp_round_ss_mask(__m128 __X, __m128 __Y) {
+  // CHECK-LABEL: @test_mm_cmp_round_ss_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_cmp_round_ss_mask(__X, __Y, 5, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__mmask8 test_mm_mask_cmp_round_ss_mask(__mmask8 __M, __m128 __X, __m128 __Y) {
+  // CHECK-LABEL: @test_mm_mask_cmp_round_ss_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_mask_cmp_round_ss_mask(__M, __X, __Y, 5, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__mmask8 test_mm_cmp_ss_mask(__m128 __X, __m128 __Y) {
+  // CHECK-LABEL: @test_mm_cmp_ss_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_cmp_ss_mask(__X, __Y, 5); 
+}
+
+__mmask8 test_mm_mask_cmp_ss_mask(__mmask8 __M, __m128 __X, __m128 __Y) {
+  // CHECK-LABEL: @test_mm_mask_cmp_ss_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_mask_cmp_ss_mask(__M, __X, __Y, 5); 
+}
+
+__mmask8 test_mm_cmp_round_sd_mask(__m128d __X, __m128d __Y) {
+  // CHECK-LABEL: @test_mm_cmp_round_sd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_cmp_round_sd_mask(__X, __Y, 5, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__mmask8 test_mm_mask_cmp_round_sd_mask(__mmask8 __M, __m128d __X, __m128d __Y) {
+  // CHECK-LABEL: @test_mm_mask_cmp_round_sd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_mask_cmp_round_sd_mask(__M, __X, __Y, 5, _MM_FROUND_CUR_DIRECTION); 
+}
+
+__mmask8 test_mm_cmp_sd_mask(__m128d __X, __m128d __Y) {
+  // CHECK-LABEL: @test_mm_cmp_sd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_cmp_sd_mask(__X, __Y, 5); 
+}
+
+__mmask8 test_mm_mask_cmp_sd_mask(__mmask8 __M, __m128d __X, __m128d __Y) {
+  // CHECK-LABEL: @test_mm_mask_cmp_sd_mask
+  // CHECK: @llvm.x86.avx512.mask.cmp
+  return _mm_mask_cmp_sd_mask(__M, __X, __Y, 5); 
+}