[AVX512] add missing integer to float conversion

[AsafBadouh]

Summary

I used SSE intrinsics to implement some of the conversion.
for example:

#define _mm_cvtss_i32 _mm_cvtss_si32

Backend will generate avx512 asm in cases we compile with avx512f flag.

[igorb]

LGTM

cfe/trunk/lib/Headers/avx512fintrin.h

(__m128)__builtin_ia32_cvtsd2ss_round_mask((__v4sf)(__m128)(A), \
                                              (__mmask8)(U), (int)(R)); })
 
 #define _mm_maskz_cvt_roundsd_ss(U, A, B, R) __extension__ ({ \
   (__m128)__builtin_ia32_cvtsd2ss_round_mask((__v4sf)(__m128)(A), \
                                              (__v2df)(__m128d)(B), \
                                              (__v4sf)_mm_setzero_ps(), \
                                              (__mmask8)(U), (int)(R)); })
 
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_mask_cvtsd_ss (__m128 W, __mmask8 U, __m128 A, __m128d B)
+{
+  return __builtin_ia32_cvtsd2ss_round_mask ((__v4sf)(A),
+                                             (__v2df)(B),
+                                             (__v4sf)(W), 
+                                             (__mmask8)(U), _MM_FROUND_CUR_DIRECTION);
+}
+
+static __inline__ __m128 __DEFAULT_FN_ATTRS
+_mm_maskz_cvtsd_ss (__mmask8 U, __m128 A, __m128d B)
+{
+  return __builtin_ia32_cvtsd2ss_round_mask ((__v4sf)(A),
+                                             (__v2df)(B),
+                                             (__v4sf)_mm_setzero_ps(), 
+                                             (__mmask8)(U), _MM_FROUND_CUR_DIRECTION);
+}
+
+#define _mm_cvtss_i32 _mm_cvtss_si32
+#define _mm_cvtss_i64 _mm_cvtss_si64
+#define _mm_cvtsd_i32 _mm_cvtsd_si32
+#define _mm_cvtsd_i64 _mm_cvtsd_si64
+#define _mm_cvti32_sd _mm_cvtsi32_sd
+#define _mm_cvti64_sd _mm_cvtsi64_sd
+#define _mm_cvti32_ss _mm_cvtsi32_ss
+#define _mm_cvti64_ss _mm_cvtsi64_ss
+
 #define _mm_cvt_roundi64_sd(A, B, R) __extension__ ({ \
   (__m128d)__builtin_ia32_cvtsi2sd64((__v2df)(__m128d)(A), (long long)(B), \
                                      (int)(R)); })
 
 #define _mm_cvt_roundsi64_sd(A, B, R) __extension__ ({ \
   (__m128d)__builtin_ia32_cvtsi2sd64((__v2df)(__m128d)(A), (long long)(B), \
                                      (int)(R)); })
 

 
 #define _mm_maskz_cvt_roundss_sd(U, A, B, R) __extension__ ({ \
   (__m128d)__builtin_ia32_cvtss2sd_round_mask((__v2df)(__m128d)(A), \
                                               (__v4sf)(__m128)(B), \
                                               (__v2df)_mm_setzero_pd(), \
                                               (__mmask8)(U), (int)(R)); })
 
 static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_mask_cvtss_sd (__m128d W, __mmask8 U, __m128d A, __m128 B)
+{
+  return __builtin_ia32_cvtss2sd_round_mask((__v2df)(A),
+                                              (__v4sf)(B),
+                                              (__v2df)(W),
+                                              (__mmask8)(U), _MM_FROUND_CUR_DIRECTION); 
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
+_mm_maskz_cvtss_sd (__mmask8 U, __m128d A, __m128 B)
+{
+  return __builtin_ia32_cvtss2sd_round_mask((__v2df)(A),
+                                              (__v4sf)(B),
+                                              (__v2df)_mm_setzero_pd(), 
+                                              (__mmask8)(U), _MM_FROUND_CUR_DIRECTION); 
+}
+
+static __inline__ __m128d __DEFAULT_FN_ATTRS
 _mm_cvtu32_sd (__m128d __A, unsigned __B)
 {
   return (__m128d) __builtin_ia32_cvtusi2sd32 ((__v2df) __A, __B);
 }
 
 #define _mm_cvt_roundu64_sd(A, B, R) __extension__ ({ \
   (__m128d)__builtin_ia32_cvtusi2sd64((__v2df)(__m128d)(A), \
                                       (unsigned long long)(B), (int)(R)); })

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

__m512 test_mm512_setr_ps (float __A, float __B, float __C, float __D,
     //CHECK: insertelement{{.*}}i32 11
     //CHECK: insertelement{{.*}}i32 12
     //CHECK: insertelement{{.*}}i32 13
     //CHECK: insertelement{{.*}}i32 14
     //CHECK: insertelement{{.*}}i32 15
     return _mm512_setr_ps( __A, __B, __C, __D, __E, __F, __G, __H,
                           __I, __J, __K, __L, __M, __N, __O, __P);
 }
+
+int test_mm_cvtss_i32(__m128 A) {
+  // CHECK-LABEL: test_mm_cvtss_i32
+  // CHECK: call i32 @llvm.x86.sse.cvtss2si(<4 x float> %{{.*}})
+  return _mm_cvtss_i32(A);
+}
+
+long long test_mm_cvtss_i64(__m128 A) {
+  // CHECK-LABEL: test_mm_cvtss_i64
+  // CHECK: call i64 @llvm.x86.sse.cvtss2si64(<4 x float> %{{.*}})
+  return _mm_cvtss_i64(A);
+}
+
+__m128d test_mm_cvti32_sd(__m128d A, int B) {
+  // CHECK-LABEL: test_mm_cvti32_sd
+  // CHECK: sitofp i32 %{{.*}} to double
+  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  return _mm_cvti32_sd(A, B);
+}
+
+__m128d test_mm_cvti64_sd(__m128d A, long long B) {
+  // CHECK-LABEL: test_mm_cvti64_sd
+  // CHECK: sitofp i64 %{{.*}} to double
+  // CHECK: insertelement <2 x double> %{{.*}}, double %{{.*}}, i32 0
+  return _mm_cvti64_sd(A, B);
+}
+
+__m128 test_mm_cvti32_ss(__m128 A, int B) {
+  // CHECK-LABEL: test_mm_cvti32_ss
+  // CHECK: sitofp i32 %{{.*}} to float
+  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  return _mm_cvti32_ss(A, B);
+}
+
+__m128 test_mm_cvti64_ss(__m128 A, long long B) {
+  // CHECK-LABEL: test_mm_cvti64_ss
+  // CHECK: sitofp i64 %{{.*}} to float
+  // CHECK: insertelement <4 x float> %{{.*}}, float %{{.*}}, i32 0
+  return _mm_cvti64_ss(A, B);
+}
+
+int test_mm_cvtsd_i32(__m128d A) {
+  // CHECK-LABEL: test_mm_cvtsd_i32
+  // CHECK: call i32 @llvm.x86.sse2.cvtsd2si(<2 x double> %{{.*}})
+  return _mm_cvtsd_i32(A);
+}
+
+long long test_mm_cvtsd_i64(__m128d A) {
+  // CHECK-LABEL: test_mm_cvtsd_i64
+  // CHECK: call i64 @llvm.x86.sse2.cvtsd2si64(<2 x double> %{{.*}})
+  return _mm_cvtsd_i64(A);
+}
+
+__m128d test_mm_mask_cvtss_sd(__m128d __W, __mmask8 __U, __m128d __A, __m128 __B) {
+  // CHECK-LABEL: @test_mm_mask_cvtss_sd
+  // CHECK: @llvm.x86.avx512.mask.cvtss2sd.round
+  return _mm_mask_cvtss_sd(__W, __U, __A, __B); 
+}
+
+__m128d test_mm_maskz_cvtss_sd( __mmask8 __U, __m128d __A, __m128 __B) {
+  // CHECK-LABEL: @test_mm_maskz_cvtss_sd
+  // CHECK: @llvm.x86.avx512.mask.cvtss2sd.round
+  return _mm_maskz_cvtss_sd( __U, __A, __B); 
+}
+
+__m128 test_mm_mask_cvtsd_ss(__m128 __W, __mmask8 __U, __m128 __A, __m128d __B) {
+  // CHECK-LABEL: @test_mm_mask_cvtsd_ss
+  // CHECK: @llvm.x86.avx512.mask.cvtsd2ss.round
+  return _mm_mask_cvtsd_ss(__W, __U, __A, __B); 
+}
+
+__m128 test_mm_maskz_cvtsd_ss(__mmask8 __U, __m128 __A, __m128d __B) {
+  // CHECK-LABEL: @test_mm_maskz_cvtsd_ss
+  // CHECK: @llvm.x86.avx512.mask.cvtsd2ss.round
+  return _mm_maskz_cvtsd_ss(__U, __A, __B); 
+}
+