MSVC doesn't provide overload of operator+ fr vector type

serge-sans-paille · serge-sans-paille · commit b91a9f8e15de · 2021-04-08T15:44:49.000+02:00
diff --git a/include/xsimd/types/xsimd_avx512_double.hpp b/include/xsimd/types/xsimd_avx512_double.hpp
@@ -490,7 +490,7 @@ namespace xsimd
             {
                 __m256d tmp1 = _mm512_extractf64x4_pd(rhs, 1);
                 __m256d tmp2 = _mm512_extractf64x4_pd(rhs, 0);
-                __m256d res1 = tmp1 + tmp2;
+                __m256d res1 = _mm256_add_pd(tmp1, tmp2);
                 return xsimd::hadd(batch<double, 4>(res1));
             }
 
@@ -501,7 +501,7 @@ namespace xsimd
         {                                                                    \
             auto tmp1 = _mm512_shuffle_f64x2(a, b, _MM_SHUFFLE(1, 0, 1, 0)); \
             auto tmp2 = _mm512_shuffle_f64x2(a, b, _MM_SHUFFLE(3, 2, 3, 2)); \
-            res ## I = (tmp1 + tmp2);                                        \
+            res ## I = _mm512_add_pd(tmp1, tmp2);                                        \
         }                                                                    \
 
                 step1(1, row[0], row[2]);
@@ -514,12 +514,12 @@ namespace xsimd
                 batch<double, 8> tmp5 = _mm512_shuffle_f64x2(res1, res2, _MM_SHUFFLE(2, 0, 2, 0));
                 batch<double, 8> tmp6 = _mm512_shuffle_f64x2(res1, res2, _MM_SHUFFLE(3, 1, 3, 1));
 
-                batch<double, 8> resx1 = (tmp5 + tmp6);
+                batch<double, 8> resx1 = _mm512_add_pd(tmp5, tmp6);
 
                 batch<double, 8> tmp7 = _mm512_shuffle_f64x2(res3, res4, _MM_SHUFFLE(2, 0, 2, 0));
                 batch<double, 8> tmp8 = _mm512_shuffle_f64x2(res3, res4, _MM_SHUFFLE(3, 1, 3, 1));
 
-                batch<double, 8> resx2 = (tmp7 + tmp8);
+                batch<double, 8> resx2 = _mm512_add_pd(tmp7, tmp8);
 
                 batch<double, 8> tmpx = _mm512_shuffle_pd(resx1, resx2, 0b00000000);
                 batch<double, 8> tmpy = _mm512_shuffle_pd(resx1, resx2, 0b11111111);
diff --git a/include/xsimd/types/xsimd_avx512_float.hpp b/include/xsimd/types/xsimd_avx512_float.hpp
@@ -513,7 +513,7 @@ namespace xsimd
             {
                 __m256 tmp1 = _mm512_extractf32x8_ps(rhs, 1);
                 __m256 tmp2 = _mm512_extractf32x8_ps(rhs, 0);
-                __m256 res1 = tmp1 + tmp2;
+                __m256 res1 = _mm256_add_ps(tmp1, tmp2);
                 return xsimd::hadd(batch<float, 8>(res1));
             }
 
@@ -527,7 +527,7 @@ namespace xsimd
         {                                                                                      \
             auto tmp1 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(1, 0, 1, 0));                   \
             auto tmp2 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(3, 2, 3, 2));                   \
-            res ## I = tmp1 + tmp2;                                                            \
+            res ## I = _mm512_add_ps(tmp1, tmp2);                                              \
         }                                                                                      \
 
                 XSIMD_AVX512_HADDP_STEP1(0, row[0], row[2]);
@@ -551,17 +551,17 @@ namespace xsimd
             batch<float, 16> tmp1 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(2, 0, 2, 0));        \
             batch<float, 16> tmp2 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(3, 1, 3, 1));        \
                                                                                                 \
-            batch<float, 16> resx1 = tmp1 + tmp2;                                               \
+            batch<float, 16> resx1 = _mm512_add_ps(tmp1, tmp2);                                               \
                                                                                                 \
             batch<float, 16> tmp3 = _mm512_shuffle_f32x4(c, d, _MM_SHUFFLE(2, 0, 2, 0));        \
             batch<float, 16> tmp4 = _mm512_shuffle_f32x4(c, d, _MM_SHUFFLE(3, 1, 3, 1));        \
                                                                                                 \
-            batch<float, 16> resx2 = tmp3 + tmp4;                                               \
+            batch<float, 16> resx2 = _mm512_add_ps(tmp3, tmp4);                                               \
                                                                                                 \
             batch<float, 16> tmp5 = _mm512_shuffle_ps(resx1, resx2, _MM_SHUFFLE(2, 0, 2, 0));   \
             batch<float, 16> tmp6 = _mm512_shuffle_ps(resx1, resx2, _MM_SHUFFLE(3, 1, 3, 1));   \
                                                                                                 \
-            batch<float, 16> resx3 = tmp5 + tmp6;                                               \
+            batch<float, 16> resx3 = _mm512_add_ps(tmp5, tmp6);                                               \
                                                                                                 \
             halfx ## I  = _mm256_hadd_ps(_mm512_extractf32x8_ps(resx3, 0),                      \
                                          _mm512_extractf32x8_ps(resx3, 1));                     \
diff --git a/include/xsimd/types/xsimd_avx512_int32.hpp b/include/xsimd/types/xsimd_avx512_int32.hpp
@@ -236,7 +236,7 @@ namespace xsimd
                 // TODO Why not _mm512_reduce_add_...?
                 __m256i tmp1 = _mm512_extracti32x8_epi32(rhs, 0);
                 __m256i tmp2 = _mm512_extracti32x8_epi32(rhs, 1);
-                __m256i res1 = tmp1 + tmp2;
+                __m256i res1 = _mm256_add_epi32(tmp1, tmp2);
                 return xsimd::hadd(batch<int32_t, 8>(res1));
             }
 
diff --git a/include/xsimd/types/xsimd_avx512_int64.hpp b/include/xsimd/types/xsimd_avx512_int64.hpp
@@ -293,7 +293,7 @@ namespace xsimd
             {
                 __m256i tmp1 = _mm512_extracti32x8_epi32(rhs, 0);
                 __m256i tmp2 = _mm512_extracti32x8_epi32(rhs, 1);
-                __m256i res1 = tmp1 + tmp2;
+                __m256i res1 = _mm256_add_epi64(tmp1, tmp2);
                 return xsimd::hadd(batch<int64_t, 4>(res1));
             }
 

Original file line number	Diff line number	Diff line change
`@@ -513,7 +513,7 @@ namespace xsimd`
`513`	`513`	`{`
`514`	`514`	`__m256 tmp1 = _mm512_extractf32x8_ps(rhs, 1);`
`515`	`515`	`__m256 tmp2 = _mm512_extractf32x8_ps(rhs, 0);`
`516`		`- __m256 res1 = tmp1 + tmp2;`
	`516`	`+ __m256 res1 = _mm256_add_ps(tmp1, tmp2);`
`517`	`517`	`return xsimd::hadd(batch<float, 8>(res1));`
`518`	`518`	`}`
`519`	`519`
`@@ -527,7 +527,7 @@ namespace xsimd`
`527`	`527`	`{ \`
`528`	`528`	`auto tmp1 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(1, 0, 1, 0)); \`
`529`	`529`	`auto tmp2 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(3, 2, 3, 2)); \`
`530`		`- res ## I = tmp1 + tmp2; \`
	`530`	`+ res ## I = _mm512_add_ps(tmp1, tmp2); \`
`531`	`531`	`} \`
`532`	`532`
`533`	`533`	`XSIMD_AVX512_HADDP_STEP1(0, row[0], row[2]);`
`@@ -551,17 +551,17 @@ namespace xsimd`
`551`	`551`	`batch<float, 16> tmp1 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(2, 0, 2, 0)); \`
`552`	`552`	`batch<float, 16> tmp2 = _mm512_shuffle_f32x4(a, b, _MM_SHUFFLE(3, 1, 3, 1)); \`
`553`	`553`	`\`
`554`		`- batch<float, 16> resx1 = tmp1 + tmp2; \`
	`554`	`+ batch<float, 16> resx1 = _mm512_add_ps(tmp1, tmp2); \`
`555`	`555`	`\`
`556`	`556`	`batch<float, 16> tmp3 = _mm512_shuffle_f32x4(c, d, _MM_SHUFFLE(2, 0, 2, 0)); \`
`557`	`557`	`batch<float, 16> tmp4 = _mm512_shuffle_f32x4(c, d, _MM_SHUFFLE(3, 1, 3, 1)); \`
`558`	`558`	`\`
`559`		`- batch<float, 16> resx2 = tmp3 + tmp4; \`
	`559`	`+ batch<float, 16> resx2 = _mm512_add_ps(tmp3, tmp4); \`
`560`	`560`	`\`
`561`	`561`	`batch<float, 16> tmp5 = _mm512_shuffle_ps(resx1, resx2, _MM_SHUFFLE(2, 0, 2, 0)); \`
`562`	`562`	`batch<float, 16> tmp6 = _mm512_shuffle_ps(resx1, resx2, _MM_SHUFFLE(3, 1, 3, 1)); \`
`563`	`563`	`\`
`564`		`- batch<float, 16> resx3 = tmp5 + tmp6; \`
	`564`	`+ batch<float, 16> resx3 = _mm512_add_ps(tmp5, tmp6); \`
`565`	`565`	`\`
`566`	`566`	`halfx ## I = _mm256_hadd_ps(_mm512_extractf32x8_ps(resx3, 0), \`
`567`	`567`	`_mm512_extractf32x8_ps(resx3, 1)); \`
Original file line number	Diff line number	Diff line change
`@@ -236,7 +236,7 @@ namespace xsimd`
`236`	`236`	`// TODO Why not _mm512_reduce_add_...?`
`237`	`237`	`__m256i tmp1 = _mm512_extracti32x8_epi32(rhs, 0);`
`238`	`238`	`__m256i tmp2 = _mm512_extracti32x8_epi32(rhs, 1);`
`239`		`- __m256i res1 = tmp1 + tmp2;`
	`239`	`+ __m256i res1 = _mm256_add_epi32(tmp1, tmp2);`
`240`	`240`	`return xsimd::hadd(batch<int32_t, 8>(res1));`
`241`	`241`	`}`
`242`	`242`
Original file line number	Diff line number	Diff line change
`@@ -293,7 +293,7 @@ namespace xsimd`
`293`	`293`	`{`
`294`	`294`	`__m256i tmp1 = _mm512_extracti32x8_epi32(rhs, 0);`
`295`	`295`	`__m256i tmp2 = _mm512_extracti32x8_epi32(rhs, 1);`
`296`		`- __m256i res1 = tmp1 + tmp2;`
	`296`	`+ __m256i res1 = _mm256_add_epi64(tmp1, tmp2);`
`297`	`297`	`return xsimd::hadd(batch<int64_t, 4>(res1));`
`298`	`298`	`}`
`299`	`299`