From 1b5cf34835c624453eac56fcdeebd762d89ee2cc Mon Sep 17 00:00:00 2001
From: James Willis <james.s.willis@durham.ac.uk>
Date: Thu, 29 Jun 2017 13:49:41 +0100
Subject: [PATCH] Fixed bugs in AVX-512 instructions.
Conflicts:
src/vector.h
---
src/vector.h | 11 +++++++----
1 file changed, 7 insertions(+), 4 deletions(-)
diff --git a/src/vector.h b/src/vector.h
index 7d82b9f5c9..93b7968475 100644
--- a/src/vector.h
+++ b/src/vector.h
@@ -116,10 +116,13 @@
for (int i = 0; i < VEC_SIZE; i++) b += a.f[i]; \
}
#endif
-/* Calculates the number of set bits in the mask and adds the result to an int.
- */
-#define VEC_FORM_PACKED_MASK(mask, v_mask, pack) \
- pack += __builtin_popcount(mask);
+
+/* Do nothing in the case of AVX-512 as there are already
+ * instructions for left-packing.*/
+#define VEC_FORM_PACKED_MASK(mask, packed_mask) packed_mask = mask
+
+/* Finds the horizontal maximum of vector b and returns a float. */
+#define VEC_HMAX(a, b) b = _mm512_reduce_max_ps(a.v)
/* Performs a left-pack on a vector based upon a mask and returns the result. */
#define VEC_LEFT_PACK(a, mask, result) \
--
GitLab