trying to fix avx build failure on nix compiler

dhewm · Nov 18, 2024 · d0fc68c · d0fc68c
1 parent 8dcbdc2
commit d0fc68c
Show file tree

Hide file tree

Showing 6 changed files with 7,808 additions and 7,506 deletions.
diff --git a/neo/idlib/math/Simd_AVX.cpp b/neo/idlib/math/Simd_AVX.cpp
@@ -35,6 +35,10 @@ If you have questions concerning this license or the applicable additional terms
 //
 //===============================================================
 
+// Revelator: these work whether in gcc clang or msvc in x86 or x64 (no inline assembly used)
+#if defined(_MSC_VER) && ( defined(_M_X64) || defined(_M_IX86) ) || \
+	defined(__GNUC__) && ( defined(__i386__) || defined (__x86_64__) ) && defined(__AVX__)
+
 #include <immintrin.h>
 
 #include "idlib/geometry/DrawVert.h"
@@ -122,3 +126,5 @@ void VPCALL idSIMD_AVX::CullByFrustum2( idDrawVert *verts, const int numVerts, c
 	}
 	_mm256_zeroupper();
 }
+
+#endif  /* _MSC_VER */
diff --git a/neo/idlib/math/Simd_AVX.h b/neo/idlib/math/Simd_AVX.h
@@ -28,10 +28,15 @@
 
 class idSIMD_AVX : public idSIMD_SSE3 {
 public:
-	// Revelator: these work whether in gcc clang or msvc x86 or x64 (no inline assembly used)
+// Revelator: these work whether in gcc clang or msvc x86 or x64 (no inline assembly used)
+#if defined(_MSC_VER) && ( defined(_M_X64) || defined(_M_IX86) ) || \
+	defined(__GNUC__) && ( defined(__i386__) || defined (__x86_64__) ) && defined(__AVX__)
+
 	virtual const char *VPCALL GetName( void ) const;
 	virtual void VPCALL CullByFrustum( idDrawVert *verts, const int numVerts, const idPlane frustum[6], byte *pointCull, float epsilon );
 	virtual void VPCALL CullByFrustum2( idDrawVert *verts, const int numVerts, const idPlane frustum[6], unsigned short *pointCull, float epsilon );
+
+#endif  /* _MSC_VER */
 };
 
 #endif /* !__MATH_SIMD_AVX_H__ */
diff --git a/neo/idlib/math/Simd_AVX2.cpp b/neo/idlib/math/Simd_AVX2.cpp
@@ -26,7 +26,7 @@ If you have questions concerning this license or the applicable additional terms
 ===========================================================================
 */
 
-#include "sys/platform.h"
+#include "sys/platform.h"
 #include "Simd_AVX2.h"
 
 //===============================================================
@@ -35,6 +35,10 @@ If you have questions concerning this license or the applicable additional terms
 //
 //===============================================================
 
+// Revelator: these work whether in gcc clang or msvc in x86 or x64 (no inline assembly used)
+#if defined(_MSC_VER) && ( defined(_M_X64) || defined(_M_IX86) ) || \
+	defined(__GNUC__) && ( defined(__i386__) || defined (__x86_64__) ) && defined(__AVX2__)
+
 #include <immintrin.h>
 
 #include "idlib/geometry/DrawVert.h"
@@ -62,20 +66,17 @@ void VPCALL idSIMD_AVX2::CullByFrustum( idDrawVert *verts, const int numVerts, c
 	const __m256 fC = _mm256_set_ps( 0, 0, frustum[5][2], frustum[4][2], frustum[3][2], frustum[2][2], frustum[1][2], frustum[0][2] );
 	const __m256 fD = _mm256_set_ps( 0, 0, frustum[5][3], frustum[4][3], frustum[3][3], frustum[2][3], frustum[1][3], frustum[0][3] );
 	const __m256 eps = _mm256_set1_ps( epsilon );
-	const byte mask6 = (1 << 6) - 1;
+	const byte mask6 = ( 1 << 6 ) - 1;
 
 	for ( int j = 0; j < numVerts; j++ ) {
 		idVec3 &vec = verts[j].xyz;
 		__m256 vX = _mm256_set1_ps( vec.x );
 		__m256 vY = _mm256_set1_ps( vec.y );
 		__m256 vZ = _mm256_set1_ps( vec.z );
-		__m256 d = _mm256_fmadd_ps( fA, vX,
-			_mm256_fmadd_ps( fB, vY,
-				_mm256_fmadd_ps( fC, vZ, fD )
-			) 
-		);
+		__m256 d = _mm256_fmadd_ps( fA, vX, _mm256_fmadd_ps( fB, vY,
+		                                    _mm256_fmadd_ps( fC, vZ, fD ) ) );
 		int mask_lo = _mm256_movemask_ps( _mm256_cmp_ps( d, eps, _CMP_LT_OQ ) );
-		pointCull[j] = (byte)mask_lo & mask6;
+		pointCull[j] = ( byte )mask_lo & mask6;
 	}
 	_mm256_zeroupper();
 }
@@ -92,21 +93,20 @@ void VPCALL idSIMD_AVX2::CullByFrustum2( idDrawVert *verts, const int numVerts,
 	const __m256 fD = _mm256_set_ps( 0, 0, frustum[5][3], frustum[4][3], frustum[3][3], frustum[2][3], frustum[1][3], frustum[0][3] );
 	const __m256 eps = _mm256_set1_ps( epsilon );
 	static const __m256 epsM = _mm256_set1_ps( -epsilon );
-	const short mask6 = (1 << 6) - 1;
+	const short mask6 = ( 1 << 6 ) - 1;
 
 	for ( int j = 0; j < numVerts; j++ ) {
 		idVec3 &vec = verts[j].xyz;
 		__m256 vX = _mm256_set1_ps( vec.x );
 		__m256 vY = _mm256_set1_ps( vec.y );
 		__m256 vZ = _mm256_set1_ps( vec.z );
-		__m256 d = _mm256_fmadd_ps( fA, vX,
-			_mm256_fmadd_ps( fB, vY,
-				_mm256_fmadd_ps( fC, vZ, fD )
-			) 
-		);
+		__m256 d = _mm256_fmadd_ps( fA, vX, _mm256_fmadd_ps( fB, vY, 
+											_mm256_fmadd_ps( fC, vZ, fD ) ) );
 		int mask_lo = _mm256_movemask_ps( _mm256_cmp_ps( d, eps, _CMP_LT_OQ ) );
 		int mask_hi = _mm256_movemask_ps( _mm256_cmp_ps( d, eps, _CMP_GT_OQ ) );
-		pointCull[j] = (unsigned short)(mask_lo & mask6 | (mask_hi & mask6) << 6);
+		pointCull[j] = ( unsigned short )( mask_lo & mask6 | ( mask_hi & mask6 ) << 6 );
 	}
 	_mm256_zeroupper();
 }
+
+#endif
diff --git a/neo/idlib/math/Simd_AVX2.h b/neo/idlib/math/Simd_AVX2.h
@@ -28,10 +28,15 @@
 
 class idSIMD_AVX2 : public idSIMD_AVX {
 public:
-	// Revelator: these work whether gcc clang or msvc in x86 or x64 (no inline assembly used)
+// Revelator: these work whether in gcc clang or msvc in x86 or x64 (no inline assembly used)
+#if defined(_MSC_VER) && ( defined(_M_X64) || defined(_M_IX86) ) || \
+	defined(__GNUC__) && ( defined(__i386__) || defined (__x86_64__) ) && defined(__AVX2__)
+
 	virtual const char *VPCALL GetName( void ) const;
 	virtual void VPCALL CullByFrustum( idDrawVert *verts, const int numVerts, const idPlane frustum[6], byte *pointCull, float epsilon );
 	virtual void VPCALL CullByFrustum2( idDrawVert *verts, const int numVerts, const idPlane frustum[6], unsigned short *pointCull, float epsilon );
+
+#endif  /* _MSC_VER */
 };
 
 #endif /* !__MATH_SIMD_AVX2_H__ */