From 97ac016b1dfbe1d603532a1bdb347882b7e22bc2 Mon Sep 17 00:00:00 2001
From: Isuru Fernando <isuruf@gmail.com>
Date: Thu, 3 Nov 2022 10:26:16 -0500
Subject: [PATCH] use opencl fma builtin
---
pyopencl/cl/pyopencl-complex.h | 6 +++---
test/test_clmath.py | 2 +-
2 files changed, 4 insertions(+), 4 deletions(-)
diff --git a/pyopencl/cl/pyopencl-complex.h b/pyopencl/cl/pyopencl-complex.h
index c33cbca0..892b5868 100644
--- a/pyopencl/cl/pyopencl-complex.h
+++ b/pyopencl/cl/pyopencl-complex.h
@@ -77,11 +77,11 @@
; \
} \
\
- inline TP TPROOT##_muladd(TP a, TP b, TP c) \
+ inline TP TPROOT##_fma(TP a, TP b, TP c) \
{ \
return TPROOT##_new( \
- (c.real + a.real*b.real) - a.imag*b.imag, \
- (c.imag + a.real*b.imag) + a.imag*b.real); \
+ fma(a.real, b.real, c.real) - a.imag*b.imag, \
+ fma(a.imag, b.real, fma(a.real, b.imag, c.imag))); \
} \
\
inline TP TPROOT##_mul(TP a, TP b) \
diff --git a/test/test_clmath.py b/test/test_clmath.py
index b640e0bb..55d14194 100644
--- a/test/test_clmath.py
+++ b/test/test_clmath.py
@@ -488,7 +488,7 @@ def test_complex_muladd(ctx_factory, dtype):
)
{{
int gid = get_global_id(0);
- res[gid] = c{real_type_name}_muladd(a[gid], b[gid], c[gid]);
+ res[gid] = c{real_type_name}_fma(a[gid], b[gid], c[gid]);
}}
""".format(real_type_name=real_type_name)
--
GitLab