From c6bb38c324a36d539b99c327ec7dcf1b0122a448 Mon Sep 17 00:00:00 2001
From: Michael Kuron <mkuron@icp.uni-stuttgart.de>
Date: Mon, 2 Sep 2019 10:15:35 +0200
Subject: [PATCH] AES-NI: Use FMA if available
---
pystencils/include/aesni_rand.h | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/pystencils/include/aesni_rand.h b/pystencils/include/aesni_rand.h
index 3a708314d..f182518c0 100644
--- a/pystencils/include/aesni_rand.h
+++ b/pystencils/include/aesni_rand.h
@@ -8,6 +8,9 @@
#include <immintrin.h> // AVX*
#else
#include <smmintrin.h> // SSE4
+#ifdef __FMA__
+#include <immintrin.h> // FMA
+#endif
#endif
#include <cstdint>
@@ -84,8 +87,12 @@ QUALIFIERS void aesni_double2(uint32 ctr0, uint32 ctr1, uint32 ctr2, uint32 ctr3
// convert uint64 to double
__m128d rs = _my_cvtepu64_pd(z);
// calculate rs * TWOPOW53_INV_DOUBLE + (TWOPOW53_INV_DOUBLE/2.0)
+#ifdef __FMA__
+ rs = _mm_fmadd_pd(rs, _mm_set_pd1(TWOPOW53_INV_DOUBLE), _mm_set_pd1(TWOPOW53_INV_DOUBLE/2.0));
+#else
rs = _mm_mul_pd(rs, _mm_set_pd1(TWOPOW53_INV_DOUBLE));
rs = _mm_add_pd(rs, _mm_set_pd1(TWOPOW53_INV_DOUBLE/2.0));
+#endif
// store result
double rr[2];
@@ -107,8 +114,12 @@ QUALIFIERS void aesni_float4(uint32 ctr0, uint32 ctr1, uint32 ctr2, uint32 ctr3,
// convert uint32 to float
__m128 rs = _my_cvtepu32_ps(c128);
// calculate rs * TWOPOW32_INV_FLOAT + (TWOPOW32_INV_FLOAT/2.0f)
+#ifdef __FMA__
+ rs = _mm_fmadd_ps(rs, _mm_set_ps1(TWOPOW32_INV_FLOAT), _mm_set_ps1(TWOPOW32_INV_FLOAT/2.0f));
+#else
rs = _mm_mul_ps(rs, _mm_set_ps1(TWOPOW32_INV_FLOAT));
rs = _mm_add_ps(rs, _mm_set_ps1(TWOPOW32_INV_FLOAT/2.0f));
+#endif
// store result
float r[4];
--
GitLab