From f268e0f83643f2b55db9eb331bc3a1d7808210a5 Mon Sep 17 00:00:00 2001
From: Megvii Engine Team <megengine@megvii.com>
Date: Mon, 22 Jun 2020 19:55:27 +0800
Subject: [PATCH] fix(dnn/x86): fix x86 compile in gcc 5.4

GitOrigin-RevId: b8683b1ad2afba5119f434b1e0b231ee0918db16
---
 .../x86/matrix_mul/int8/kernel_avx2_4x16x2.h  | 23 +++++++++++++++----
 1 file changed, 19 insertions(+), 4 deletions(-)

diff --git a/dnn/src/x86/matrix_mul/int8/kernel_avx2_4x16x2.h b/dnn/src/x86/matrix_mul/int8/kernel_avx2_4x16x2.h
index 01565354..ff32e2d3 100644
--- a/dnn/src/x86/matrix_mul/int8/kernel_avx2_4x16x2.h
+++ b/dnn/src/x86/matrix_mul/int8/kernel_avx2_4x16x2.h
@@ -17,13 +17,21 @@
 #include "src/common/utils.h"
 #include "src/x86/matrix_mul/common/common.h"
 
+#define DNN_AVX2_TARGET
+#if !defined(__clang__)
+//! bypass gcc bug https://bugs.launchpad.net/ubuntu/+source/gcc-5/+bug/1642109
+#pragma GCC target("avx2")
+#else
+#undef DNN_AVX2_TARGET
+#define DNN_AVX2_TARGET MEGDNN_ATTRIBUTE_TARGET("avx2")
+#endif
 namespace megdnn {
 namespace x86 {
 
 namespace matmul_avx2_4x16x2 {
+
 template <typename CType>
-MEGDNN_ATTRIBUTE_TARGET("avx2")
-void store_overflow(void* ptr, __m256i a);
+DNN_AVX2_TARGET void store_overflow(void* ptr, __m256i a);
 
 template <>
 void store_overflow<int16_t>(void* ptr, __m256i a) {
@@ -33,13 +41,14 @@ void store_overflow<int16_t>(void* ptr, __m256i a) {
     a = _mm256_permutevar8x32_epi32(a, idx);
     _mm_storeu_si128((__m128i*)ptr, _mm256_extractf128_si256(a, 0));
 }
+
 template <>
 void store_overflow<int32_t>(void* ptr, __m256i a) {
     _mm256_storeu_si256((__m256i*)(ptr), a);
 }
+
 template <typename CType>
-MEGDNN_ATTRIBUTE_TARGET("avx2")
-void store_overflow(void* ptr, __m256i a, int remain);
+DNN_AVX2_TARGET void store_overflow(void* ptr, __m256i a, int remain);
 
 template <>
 void store_overflow<int16_t>(void* ptr, __m256i a, int remain) {
@@ -51,6 +60,7 @@ void store_overflow<int16_t>(void* ptr, __m256i a, int remain) {
     _mm_maskmoveu_si128(_mm256_extractf128_si256(a, 0), mask,
                         reinterpret_cast<char*>(ptr));
 }
+
 template <>
 void store_overflow<int32_t>(void* ptr, __m256i a, int remain) {
     __m256i mask = _m256_continue_mask(remain);
@@ -870,4 +880,9 @@ static inline void gemm_s8s8s32_avx2_4x16x2_pack_at(dt_int16* out,
 }  // namespace x86
 }  // namespace megdnn
 
+#if !defined(__clang__)
+#pragma GCC reset_options
+#endif
+#undef DNN_AVX2_TARGET
+
 // vim: syntax=cpp.doxygen
\ No newline at end of file