Disable FMA intrinsics in the srot kernel when the compiler is PGI/NVIDIA

martin-frbg · web-flow · commit 114eb159a4b0 · 2020-12-19T22:15:58.000+01:00
diff --git a/kernel/x86_64/srot.c b/kernel/x86_64/srot.c
@@ -13,7 +13,7 @@ static void srot_kernel(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT c, FLOAT s)
 {
     BLASLONG i = 0;
     
-#if V_SIMD && (defined(HAVE_FMA3) || V_SIMD > 128)
+#if V_SIMD && !defined(C_PGI) && (defined(HAVE_FMA3) || V_SIMD > 128)
     const int vstep = v_nlanes_f32;
     const int unrollx4 = n & (-vstep * 4);
     const int unrollx = n & -vstep;

Original file line number	Diff line number	Diff line change
`@@ -13,7 +13,7 @@ static void srot_kernel(BLASLONG n, FLOAT x, FLOAT y, FLOAT c, FLOAT s)`
`13`	`13`	`{`
`14`	`14`	`BLASLONG i = 0;`
`15`	`15`
`16`		`-#if V_SIMD && (defined(HAVE_FMA3) \|\| V_SIMD > 128)`
	`16`	`+#if V_SIMD && !defined(C_PGI) && (defined(HAVE_FMA3) \|\| V_SIMD > 128)`
`17`	`17`	`const int vstep = v_nlanes_f32;`
`18`	`18`	`const int unrollx4 = n & (-vstep * 4);`
`19`	`19`	`const int unrollx = n & -vstep;`