Clang part of MMX rewrite (goes with 115243).

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@115244 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Headers/mmintrin.h

@@ -43,14 +43,13 @@ _mm_empty(void)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi32_si64(int __i)
 {
-    return (__m64)(__v2si){__i, 0};
+    return (__m64)__builtin_ia32_vec_init_v2si(__i, 0);
 }
 
 static __inline__ int __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi64_si32(__m64 __m)
 {
-    __v2si __mmx_var2 = (__v2si)__m;
-    return __mmx_var2[0];
+    return __builtin_ia32_vec_ext_v2si((__v2si)__m, 0);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -86,59 +85,55 @@ _mm_packs_pu16(__m64 __m1, __m64 __m2)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpackhi_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 4, 8+4, 5,
-                                          8+5, 6, 8+6, 7, 8+7);
+    return (__m64)__builtin_ia32_punpckhbw((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpackhi_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 2, 4+2, 3,
-                                          4+3);
+    return (__m64)__builtin_ia32_punpckhwd((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpackhi_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 1, 2+1);
+    return (__m64)__builtin_ia32_punpckhdq((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpacklo_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v8qi)__m1, (__v8qi)__m2, 0, 8+0, 1,
-                                          8+1, 2, 8+2, 3, 8+3);
+    return (__m64)__builtin_ia32_punpcklbw((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpacklo_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v4hi)__m1, (__v4hi)__m2, 0, 4+0, 1,
-                                          4+1);
+    return (__m64)__builtin_ia32_punpcklwd((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_unpacklo_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)__builtin_shufflevector((__v2si)__m1, (__v2si)__m2, 0, 2+0);
+    return (__m64)__builtin_ia32_punpckldq((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_add_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v8qi)__m1 + (__v8qi)__m2);
+    return (__m64)__builtin_ia32_paddb((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_add_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v4hi)__m1 + (__v4hi)__m2);
+    return (__m64)__builtin_ia32_paddw((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_add_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v2si)__m1 + (__v2si)__m2);
+    return (__m64)__builtin_ia32_paddd((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -168,19 +163,19 @@ _mm_adds_pu16(__m64 __m1, __m64 __m2)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_sub_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v8qi)__m1 - (__v8qi)__m2);
+    return (__m64)__builtin_ia32_psubb((__v8qi)__m1, (__v8qi)__m2);
 }
  
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_sub_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v4hi)__m1 - (__v4hi)__m2);
+    return (__m64)__builtin_ia32_psubw((__v4hi)__m1, (__v4hi)__m2);
 }
  
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_sub_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v2si)__m1 - (__v2si)__m2);
+    return (__m64)__builtin_ia32_psubd((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -222,7 +217,7 @@ _mm_mulhi_pi16(__m64 __m1, __m64 __m2)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_mullo_pi16(__m64 __m1, __m64 __m2) 
 {
-    return (__m64)((__v4hi)__m1 * (__v4hi)__m2);
+    return (__m64)__builtin_ia32_pmullw((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -252,13 +247,13 @@ _mm_slli_pi32(__m64 __m, int __count)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_sll_si64(__m64 __m, __m64 __count)
 {
-    return __builtin_ia32_psllq(__m, __count);
+    return (__m64)__builtin_ia32_psllq(__m, __count);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_slli_si64(__m64 __m, int __count)
 {
-    return __builtin_ia32_psllqi(__m, __count);    
+    return (__m64)__builtin_ia32_psllqi(__m, __count);    
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -318,67 +313,67 @@ _mm_srl_si64(__m64 __m, __m64 __count)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_srli_si64(__m64 __m, int __count)
 {
-    return __builtin_ia32_psrlqi(__m, __count);    
+    return (__m64)__builtin_ia32_psrlqi(__m, __count);    
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_and_si64(__m64 __m1, __m64 __m2)
 {
-    return __m1 & __m2;
+    return __builtin_ia32_pand(__m1, __m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_andnot_si64(__m64 __m1, __m64 __m2)
 {
-    return ~__m1 & __m2;
+    return __builtin_ia32_pandn(__m1, __m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_or_si64(__m64 __m1, __m64 __m2)
 {
-    return __m1 | __m2;
+    return __builtin_ia32_por(__m1, __m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_xor_si64(__m64 __m1, __m64 __m2)
 {
-    return __m1 ^ __m2;
+    return __builtin_ia32_pxor(__m1, __m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v8qi)__m1 == (__v8qi)__m2);
+    return (__m64)__builtin_ia32_pcmpeqb((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v4hi)__m1 == (__v4hi)__m2);
+    return (__m64)__builtin_ia32_pcmpeqw((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v2si)__m1 == (__v2si)__m2);
+    return (__m64)__builtin_ia32_pcmpeqd((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpgt_pi8(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v8qi)__m1 > (__v8qi)__m2);
+    return (__m64)__builtin_ia32_pcmpgtb((__v8qi)__m1, (__v8qi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpgt_pi16(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v4hi)__m1 > (__v4hi)__m2);
+    return (__m64)__builtin_ia32_pcmpgtw((__v4hi)__m1, (__v4hi)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpgt_pi32(__m64 __m1, __m64 __m2)
 {
-    return (__m64)((__v2si)__m1 > (__v2si)__m2);
+    return (__m64)__builtin_ia32_pcmpgtd((__v2si)__m1, (__v2si)__m2);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
@@ -390,57 +385,58 @@ _mm_setzero_si64(void)
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set_pi32(int __i1, int __i0)
 {
-    return (__m64)(__v2si){ __i0, __i1 };
+    return (__m64)__builtin_ia32_vec_init_v2si(__i0, __i1);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set_pi16(short __s3, short __s2, short __s1, short __s0)
 {
-    return (__m64)(__v4hi){ __s0, __s1, __s2, __s3 };    
+    return (__m64)__builtin_ia32_vec_init_v4hi(__s0, __s1, __s2, __s3);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
             char __b1, char __b0)
 {
-    return (__m64)(__v8qi){ __b0, __b1, __b2, __b3, __b4, __b5, __b6, __b7 };
+    return (__m64)__builtin_ia32_vec_init_v8qi(__b0, __b1, __b2, __b3,
+                                               __b4, __b5, __b6, __b7);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set1_pi32(int __i)
 {
-    return (__m64)(__v2si){ __i, __i };
+    return _mm_set_pi32(__i, __i);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
-_mm_set1_pi16(short __s)
+_mm_set1_pi16(short __w)
 {
-    return (__m64)(__v4hi){ __s, __s, __s, __s };
+    return _mm_set_pi16(__w, __w, __w, __w);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set1_pi8(char __b)
 {
-    return (__m64)(__v8qi){ __b, __b, __b, __b, __b, __b, __b, __b };
+    return _mm_set_pi8(__b, __b, __b, __b, __b, __b, __b, __b);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_setr_pi32(int __i1, int __i0)
 {
-    return (__m64)(__v2si){ __i1, __i0 };
+    return _mm_set_pi32(__i1, __i0);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
-_mm_setr_pi16(short __s3, short __s2, short __s1, short __s0)
+_mm_setr_pi16(short __w3, short __w2, short __w1, short __w0)
 {
-    return (__m64)(__v4hi){ __s3, __s2, __s1, __s0 };
+    return _mm_set_pi16(__w3, __w2, __w1, __w0);
 }
 
 static __inline__ __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_setr_pi8(char __b7, char __b6, char __b5, char __b4, char __b3, char __b2,
              char __b1, char __b0)
 {
-    return (__m64)(__v8qi){ __b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0 };
+    return _mm_set_pi8(__b7, __b6, __b5, __b4, __b3, __b2, __b1, __b0);
 }
 
 

test/CodeGen/mmx-shift-with-immediate.c

@@ -2,25 +2,25 @@
 #include <mmintrin.h>
 
 void shift(__m64 a, __m64 b, int c) {
-  // CHECK: <4 x i16> @llvm.x86.mmx.pslli.w(<4 x i16> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.pslli.w(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_slli_pi16(a, c);
-  // CHECK: <2 x i32> @llvm.x86.mmx.pslli.d(<2 x i32> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.pslli.d(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_slli_pi32(a, c);
   // FIXME: <1 x i64> @llvm.x86.mmx.pslli.q(<1 x i64> %{{.*}}, i32 {{.*}})
   // This is currently lowered into non-intrinsic instructions. This may not be
   // correct once the MMX reworking is finished.
   _mm_slli_si64(a, c);
 
-  // CHECK: <4 x i16> @llvm.x86.mmx.psrli.w(<4 x i16> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.psrli.w(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_srli_pi16(a, c);
-  // CHECK: <2 x i32> @llvm.x86.mmx.psrli.d(<2 x i32> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.psrli.d(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_srli_pi32(a, c);
   // FIXME: <1 x i64> @llvm.x86.mmx.psrli.q(<1 x i64> %{{.*}}, i32 {{.*}})
   // See above.
   _mm_srli_si64(a, c);
 
-  // CHECK: <4 x i16> @llvm.x86.mmx.psrai.w(<4 x i16> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.psrai.w(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_srai_pi16(a, c);
-  // CHECK: <2 x i32> @llvm.x86.mmx.psrai.d(<2 x i32> %{{.*}}, i32 {{.*}})
+  // CHECK: x86_mmx @llvm.x86.mmx.psrai.d(x86_mmx %{{.*}}, i32 {{.*}})
   _mm_srai_pi32(a, c);
 }