BUG, SIMD: Fix invalid value encountered in cos/sin on aarch64 & ppc64le

numpy · Dec 19, 2022 · 0398998 · 0398998
1 parent 6bbc09d
commit 0398998
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/numpy/core/src/common/simd/neon/operators.h b/numpy/core/src/common/simd/neon/operators.h
@@ -248,7 +248,7 @@ NPY_FINLINE npyv_b32 npyv_cmpgtq_f32(npyv_f32 a, npyv_f32 b)
 {
     npyv_f32 max = vmaxq_f32(a, b);
     npyv_b32 nnan = vceqq_f32(max, max);
-    return vandq_u32(nnan, vceqq_f32(max, b));
+    return vbicq_u32(nnan, vceqq_f32(max, b));
 }
 #define npyv_cmpleq_f32(A, B) npyv_cmpgeq_f32(B, A)
 #define npyv_cmpltq_f32(A, B) npyv_cmpgtq_f32(B, A)

diff --git a/numpy/core/src/umath/loops_trigonometric.dispatch.c.src b/numpy/core/src/umath/loops_trigonometric.dispatch.c.src
@@ -124,7 +124,7 @@ simd_sincos_f32(const float *src, npy_intp ssrc, float *dst, npy_intp sdst,
         } else {
             x_in = npyv_loadn_tillz_f32(src, ssrc, len);
         }
-        npyv_b32 simd_mask = npyv_cmple_f32(npyv_abs_f32(x_in), max_cody);
+        npyv_b32 simd_mask = npyv_cmpleq_f32(npyv_abs_f32(x_in), max_cody);
         npy_uint64 simd_maski = npyv_tobits_b32(simd_mask);
         /*
          * For elements outside of this range, Cody-Waite's range reduction