fix unexpected abs error on powerpc vsx (#5498)

Tencent · Jun 10, 2024 · 102e989 · 102e989
1 parent 19ea54f
commit 102e989
Show file tree

Hide file tree

Showing 4 changed files with 9 additions and 16 deletions.
diff --git a/.github/workflows/linux-ppc64-cpu-gcc.yml b/.github/workflows/linux-ppc64-cpu-gcc.yml
@@ -9,6 +9,7 @@ on:
     - 'cmake/**'
     - 'src/*'
     - 'src/layer/*'
+    - 'src/layer/x86/*'
     - 'tests/**'
   pull_request:
     branches: [master]
@@ -19,6 +20,7 @@ on:
     - 'cmake/**'
     - 'src/*'
     - 'src/layer/*'
+    - 'src/layer/x86/*'
     - 'tests/**'
 concurrency:
   group: linux-ppc64-cpu-gcc-${{ github.ref }}

diff --git a/src/layer/x86/avx512_mathfun.h b/src/layer/x86/avx512_mathfun.h
@@ -849,11 +849,8 @@ static NCNN_FORCEINLINE __m512 atan2512_ps(__m512 y, __m512 x)
 
 static NCNN_FORCEINLINE __m512 abs512_ps(__m512 x)
 {
-    // Use negative zero as the sign bit mask.
-    const __m512 magic_negative_zero = _mm512_set1_ps(-0.0f);
-
-    // return (!magic_negative_zero && x);
-    return _mm512_andnot_ps(magic_negative_zero, x);
+    const __m512 abs_mask = _mm512_castsi512_ps(_mm512_set1_epi32(0x7fffffff));
+    return _mm512_and_ps(abs_mask, x);
 }
 
 #endif // AVX512_MATHFUN_H
diff --git a/src/layer/x86/avx_mathfun.h b/src/layer/x86/avx_mathfun.h
@@ -1080,11 +1080,8 @@ static NCNN_FORCEINLINE __m256 atan2256_ps(__m256 y, __m256 x)
 
 static NCNN_FORCEINLINE __m256 abs256_ps(__m256 x)
 {
-    // Use negative zero as the sign bit mask.
-    const __m256 magic_negative_zero = _mm256_set1_ps(-0.0f);
-
-    // return (!magic_negative_zero && x);
-    return _mm256_andnot_ps(magic_negative_zero, x);
+    const __m256 abs_mask = _mm256_castsi256_ps(_mm256_set1_epi32(0x7fffffff));
+    return _mm256_and_ps(abs_mask, x);
 }
 
 #endif // AVX_MATHFUN_H
diff --git a/src/layer/x86/sse_mathfun.h b/src/layer/x86/sse_mathfun.h
@@ -1148,13 +1148,10 @@ static NCNN_FORCEINLINE __m128 atan2_ps(__m128 y, __m128 x)
                _mm_andnot_ps(normal_mode, special_result));
 }
 
-static NCNN_FORCEINLINE __m128 abs_ps(__m128 inputs)
+static NCNN_FORCEINLINE __m128 abs_ps(__m128 x)
 {
-    // Use negative zero as the sign bit mask.
-    const __m128 magic_negative_zero = _mm_set_ps1(-0.0f);
-
-    // return (!magic_negative_zero && x);
-    return _mm_andnot_ps(magic_negative_zero, inputs);
+    const __m128 abs_mask = _mm_castsi128_ps(_mm_set1_epi32(0x7fffffff));
+    return _mm_and_ps(abs_mask, x);
 }
 
 #endif // SSE_MATHFUN_H