[PowerPC] [Clang] Port SSE4.1-compatible insert intrinsics

[qiucf]

Summary

This patch adds three intrinsics compatible to x86's SSE 4.1 on PowerPC target, with tests:

• _mm_insert_epi8
• _mm_insert_epi32
• _mm_insert_epi64

The intrinsics implementation is contributed by Paul Clarke.

[qiucf]

Ping...

[jsji]

LGTM. Thanks.

clang/lib/Headers/ppc_wrappers/smmintrin.h

/* This header is distributed to simplify porting x86_64 code that
    makes explicit use of Intel intrinsics to powerp64/powerpc64le.
 
    It is the user's responsibility to determine if the results are
    acceptable and make additional changes as necessary.
 
    Note that much code that uses Intel intrinsics can be rewritten in
    standard C or GNU C extensions, which are more portable and better
    optimized across multiple targets.  */
 #error                                                                         \
     "Please read comment above.  Use -DNO_WARN_X86_INTRINSICS to disable this error."
 #endif
 
 #ifndef SMMINTRIN_H_
 #define SMMINTRIN_H_
 
 #if defined(__linux__) && defined(__ppc64__)
 

 extern __inline __m128i
     __attribute__((__gnu_inline__, __always_inline__, __artificial__))
     _mm_blendv_epi8(__m128i __A, __m128i __B, __m128i __mask) {
   const __v16qu __seven = vec_splats((unsigned char)0x07);
   __v16qu __lmask = vec_sra((__v16qu)__mask, __seven);
   return (__m128i)vec_sel((__v16qu)__A, (__v16qu)__B, __lmask);
 }
 
+extern __inline __m128i
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+    _mm_insert_epi8(__m128i const __A, int const __D, int const __N) {
+  __v16qi result = (__v16qi)__A;
+  result[__N & 0xf] = __D;
+  return (__m128i)result;
+}
+
+extern __inline __m128i
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+    _mm_insert_epi32(__m128i const __A, int const __D, int const __N) {
+  __v4si result = (__v4si)__A;
+  result[__N & 3] = __D;
+  return (__m128i)result;
+}
+
+extern __inline __m128i
+    __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+    _mm_insert_epi64(__m128i const __A, long long const __D, int const __N) {
+  __v2di result = (__v2di)__A;
+  result[__N & 1] = __D;
+  return (__m128i)result;
+}
+
 #else
 #include_next <smmintrin.h>
 #endif /* defined(__linux__) && defined(__ppc64__) */
 
 #endif /* _SMMINTRIN_H_ */

clang/test/CodeGen/ppc-smmintrin.c

 // CHECK-NEXT: [[REG75:[0-9a-zA-Z_%.]+]] = load <2 x i64>, <2 x i64>* [[REG65]], align 16
 // CHECK-NEXT: [[REG76:[0-9a-zA-Z_%.]+]] = bitcast <2 x i64> [[REG75]] to <16 x i8>
 // CHECK-NEXT: [[REG77:[0-9a-zA-Z_%.]+]] = load <2 x i64>, <2 x i64>* [[REG66]], align 16
 // CHECK-NEXT: [[REG78:[0-9a-zA-Z_%.]+]] = bitcast <2 x i64> [[REG77]] to <16 x i8>
 // CHECK-NEXT: [[REG79:[0-9a-zA-Z_%.]+]] = load <16 x i8>, <16 x i8>* [[REG74]], align 16
 // CHECK-NEXT: [[REG80:[0-9a-zA-Z_%.]+]] = call <16 x i8> @vec_sel(unsigned char vector[16], unsigned char vector[16], unsigned char vector[16])(<16 x i8> [[REG76]], <16 x i8> [[REG78]], <16 x i8> [[REG79]])
 // CHECK-NEXT: [[REG81:[0-9a-zA-Z_%.]+]] = bitcast <16 x i8> [[REG80]] to <2 x i64>
 // CHECK-NEXT: ret <2 x i64> [[REG81]]
+
+void __attribute__((noinline))
+test_insert() {
+  _mm_insert_epi8(m1, 1, 0);
+  _mm_insert_epi32(m1, 1, 0);
+  _mm_insert_epi64(m1, 0xFFFFFFFF1L, 0);
+}
+
+// CHECK-LABEL: @test_insert
+
+// CHECK: define available_externally <2 x i64> @_mm_insert_epi8(<2 x i64> {{[0-9a-zA-Z_%.]+}}, i32 signext {{[0-9a-zA-Z_%.]+}}, i32 signext {{[0-9a-zA-Z_%.]+}})
+// CHECK: %{{[0-9a-zA-Z_.]+}} = bitcast <2 x i64> %{{[0-9a-zA-Z_.]+}} to <16 x i8>
+// CHECK: %[[R0:[0-9a-zA-Z_.]+]] = trunc i32 %{{[0-9a-zA-Z_.]+}} to i8
+// CHECK: %[[R1:[0-9a-zA-Z_.]+]] = and i32 %{{[0-9a-zA-Z_.]+}}, 15
+// CHECK: %{{[0-9a-zA-Z_.]+}} = insertelement <16 x i8> %{{[0-9a-zA-Z_.]+}}, i8 %[[R0]], i32 %[[R1]]
+// CHECK: %[[R2:[0-9a-zA-Z_.]+]] = bitcast <16 x i8> %{{[0-9a-zA-Z_.]+}} to <2 x i64>
+// CHECK: ret <2 x i64> %[[R2]]
+
+// CHECK: define available_externally <2 x i64> @_mm_insert_epi32(<2 x i64> {{[0-9a-zA-Z_%.]+}}, i32 signext {{[0-9a-zA-Z_%.]+}}, i32 signext {{[0-9a-zA-Z_%.]+}})
+// CHECK: %{{[0-9a-zA-Z_.]+}} = bitcast <2 x i64> %{{[0-9a-zA-Z_.]+}} to <4 x i32>
+// CHECK: %[[R0:[0-9a-zA-Z_.]+]] = and i32 %{{[0-9a-zA-Z_.]+}}, 3
+// CHECK: %{{[0-9a-zA-Z_.]+}} = insertelement <4 x i32> %{{[0-9a-zA-Z_.]+}}, i32 %{{[0-9a-zA-Z_.]+}}, i32 %[[R0]]
+// CHECK: %[[R1:[0-9a-zA-Z_.]+]] = bitcast <4 x i32> %{{[0-9a-zA-Z_.]+}} to <2 x i64>
+// CHECK: ret <2 x i64> %[[R1]]
+
+// CHECK: define available_externally <2 x i64> @_mm_insert_epi64(<2 x i64> {{[0-9a-zA-Z_%.]+}}, i64 {{[0-9a-zA-Z_%.]+}}, i32 signext {{[0-9a-zA-Z_%.]+}})
+// CHECK: %[[R0:[0-9a-zA-Z_.]+]] = and i32 %{{[0-9a-zA-Z_.]+}}, 1
+// CHECK: %{{[0-9a-zA-Z_.]+}} = insertelement <2 x i64> %{{[0-9a-zA-Z_.]+}}, i64 %{{[0-9a-zA-Z_.]+}}, i32 %[[R0:[0-9a-zA-Z_.]+]]
+// CHECK: ret <2 x i64> %{{[0-9a-zA-Z_.]+}}