Reset MMX state after argsort/argselect

sterrettm2 · sterrettm2 · commit bd4c6d32cc67 · 2024-11-08T09:55:05.000-08:00
diff --git a/src/xss-common-argsort.h b/src/xss-common-argsort.h
@@ -575,6 +575,11 @@ X86_SIMD_SORT_INLINE void xss_argsort(T *arr,
 
         if (descending) { std::reverse(arg, arg + arrsize); }
     }
+
+#ifdef __MMX__
+    // Workaround for compiler bug generating MMX instructions without emms
+    _mm_empty();
+#endif
 }
 
 template <typename T>
@@ -632,6 +637,11 @@ X86_SIMD_SORT_INLINE void xss_argselect(T *arr,
         argselect_<vectype, argtype>(
                 arr, arg, k, 0, arrsize - 1, 2 * (arrsize_t)log2(arrsize));
     }
+
+#ifdef __MMX__
+    // Workaround for compiler bug generating MMX instructions without emms
+    _mm_empty();
+#endif
 }
 
 template <typename T>

Original file line number	Diff line number	Diff line change
`@@ -575,6 +575,11 @@ X86_SIMD_SORT_INLINE void xss_argsort(T *arr,`
`575`	`575`
`576`	`576`	`if (descending) { std::reverse(arg, arg + arrsize); }`
`577`	`577`	`}`
	`578`	`+`
	`579`	`+#ifdef __MMX__`
	`580`	`+ // Workaround for compiler bug generating MMX instructions without emms`
	`581`	`+ _mm_empty();`
	`582`	`+#endif`
`578`	`583`	`}`
`579`	`584`
`580`	`585`	`template <typename T>`
`@@ -632,6 +637,11 @@ X86_SIMD_SORT_INLINE void xss_argselect(T *arr,`
`632`	`637`	`argselect_<vectype, argtype>(`
`633`	`638`	`arr, arg, k, 0, arrsize - 1, 2 * (arrsize_t)log2(arrsize));`
`634`	`639`	`}`
	`640`	`+`
	`641`	`+#ifdef __MMX__`
	`642`	`+ // Workaround for compiler bug generating MMX instructions without emms`
	`643`	`+ _mm_empty();`
	`644`	`+#endif`
`635`	`645`	`}`
`636`	`646`
`637`	`647`	`template <typename T>`