summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--numpy/core/src/_simd/_simd.dispatch.c.src4
-rw-r--r--numpy/core/src/common/simd/avx2/math.h4
-rw-r--r--numpy/core/src/common/simd/avx512/math.h4
-rw-r--r--numpy/core/src/common/simd/neon/math.h28
-rw-r--r--numpy/core/src/common/simd/sse/math.h26
-rw-r--r--numpy/core/src/common/simd/vsx/math.h4
-rw-r--r--numpy/core/tests/test_simd.py7
7 files changed, 74 insertions, 3 deletions
diff --git a/numpy/core/src/_simd/_simd.dispatch.c.src b/numpy/core/src/_simd/_simd.dispatch.c.src
index 84de9a059..003ef7ffd 100644
--- a/numpy/core/src/_simd/_simd.dispatch.c.src
+++ b/numpy/core/src/_simd/_simd.dispatch.c.src
@@ -381,7 +381,7 @@ SIMD_IMPL_INTRIN_1(sumup_@sfx@, @esfx@, v@sfx@)
***************************/
#if @fp_only@
/**begin repeat1
- * #intrin = sqrt, recip, abs, square, ceil, trunc#
+ * #intrin = sqrt, recip, abs, square, ceil, trunc, floor#
*/
SIMD_IMPL_INTRIN_1(@intrin@_@sfx@, v@sfx@, v@sfx@)
/**end repeat1**/
@@ -615,7 +615,7 @@ SIMD_INTRIN_DEF(sumup_@sfx@)
***************************/
#if @fp_only@
/**begin repeat1
- * #intrin = sqrt, recip, abs, square, ceil, trunc#
+ * #intrin = sqrt, recip, abs, square, ceil, trunc, floor#
*/
SIMD_INTRIN_DEF(@intrin@_@sfx@)
/**end repeat1**/
diff --git a/numpy/core/src/common/simd/avx2/math.h b/numpy/core/src/common/simd/avx2/math.h
index ec15e50e1..78608d51b 100644
--- a/numpy/core/src/common/simd/avx2/math.h
+++ b/numpy/core/src/common/simd/avx2/math.h
@@ -113,4 +113,8 @@ NPY_FINLINE npyv_s64 npyv_min_s64(npyv_s64 a, npyv_s64 b)
#define npyv_trunc_f32(A) _mm256_round_ps(A, _MM_FROUND_TO_ZERO)
#define npyv_trunc_f64(A) _mm256_round_pd(A, _MM_FROUND_TO_ZERO)
+// floor
+#define npyv_floor_f32 _mm256_floor_ps
+#define npyv_floor_f64 _mm256_floor_pd
+
#endif // _NPY_SIMD_AVX2_MATH_H
diff --git a/numpy/core/src/common/simd/avx512/math.h b/numpy/core/src/common/simd/avx512/math.h
index f30e50ad0..0d6e17993 100644
--- a/numpy/core/src/common/simd/avx512/math.h
+++ b/numpy/core/src/common/simd/avx512/math.h
@@ -120,4 +120,8 @@ NPY_FINLINE npyv_f64 npyv_minp_f64(npyv_f64 a, npyv_f64 b)
#define npyv_trunc_f32(A) _mm512_roundscale_ps(A, _MM_FROUND_TO_ZERO)
#define npyv_trunc_f64(A) _mm512_roundscale_pd(A, _MM_FROUND_TO_ZERO)
+// floor
+#define npyv_floor_f32(A) _mm512_roundscale_ps(A, _MM_FROUND_TO_NEG_INF)
+#define npyv_floor_f64(A) _mm512_roundscale_pd(A, _MM_FROUND_TO_NEG_INF)
+
#endif // _NPY_SIMD_AVX512_MATH_H
diff --git a/numpy/core/src/common/simd/neon/math.h b/numpy/core/src/common/simd/neon/math.h
index 19e5cd846..8c4788b3d 100644
--- a/numpy/core/src/common/simd/neon/math.h
+++ b/numpy/core/src/common/simd/neon/math.h
@@ -223,4 +223,32 @@ NPY_FINLINE npyv_s64 npyv_min_s64(npyv_s64 a, npyv_s64 b)
#define npyv_trunc_f64 vrndq_f64
#endif // NPY_SIMD_F64
+// floor
+#ifdef NPY_HAVE_ASIMD
+ #define npyv_floor_f32 vrndmq_f32
+#else
+ NPY_FINLINE npyv_f32 npyv_floor_f32(npyv_f32 a)
+ {
+ const npyv_s32 szero = vreinterpretq_s32_f32(vdupq_n_f32(-0.0f));
+ const npyv_u32 one = vreinterpretq_u32_f32(vdupq_n_f32(1.0f));
+ const npyv_s32 max_int = vdupq_n_s32(0x7fffffff);
+
+ npyv_s32 roundi = vcvtq_s32_f32(a);
+ npyv_f32 round = vcvtq_f32_s32(roundi);
+ npyv_f32 floor = vsubq_f32(round, vreinterpretq_f32_u32(
+ vandq_u32(vcgtq_f32(round, a), one)
+ ));
+
+ npyv_u32 nnan = npyv_notnan_f32(a);
+ npyv_u32 overflow = vorrq_u32(
+ vceqq_s32(roundi, szero), vceqq_s32(roundi, max_int)
+ );
+
+ return vbslq_f32(vbicq_u32(nnan, overflow), floor, a);
+ }
+#endif // NPY_HAVE_ASIMD
+#if NPY_SIMD_F64
+ #define npyv_floor_f64 vrndmq_f64
+#endif // NPY_SIMD_F64
+
#endif // _NPY_SIMD_NEON_MATH_H
diff --git a/numpy/core/src/common/simd/sse/math.h b/numpy/core/src/common/simd/sse/math.h
index 5daf7711e..117d39fb5 100644
--- a/numpy/core/src/common/simd/sse/math.h
+++ b/numpy/core/src/common/simd/sse/math.h
@@ -202,4 +202,30 @@ NPY_FINLINE npyv_s64 npyv_min_s64(npyv_s64 a, npyv_s64 b)
}
#endif
+// floor
+#ifdef NPY_HAVE_SSE41
+ #define npyv_floor_f32 _mm_floor_ps
+ #define npyv_floor_f64 _mm_floor_pd
+#else
+ NPY_FINLINE npyv_f32 npyv_floor_f32(npyv_f32 a)
+ {
+ const npyv_f32 szero = _mm_set1_ps(-0.0f);
+ const npyv_f32 one = _mm_set1_ps(1.0f);
+ npyv_s32 roundi = _mm_cvttps_epi32(a);
+ npyv_f32 round = _mm_cvtepi32_ps(roundi);
+ npyv_f32 floor = _mm_sub_ps(round, _mm_and_ps(_mm_cmpgt_ps(round, a), one));
+ return npyv_select_f32(_mm_cmpeq_epi32(roundi, _mm_castps_si128(szero)), a, floor);
+ }
+ NPY_FINLINE npyv_f64 npyv_floor_f64(npyv_f64 a)
+ {
+ const npyv_f64 szero = _mm_set1_pd(-0.0);
+ const npyv_f64 one = _mm_set1_pd(1.0);
+ const npyv_f64 two_power_52 = _mm_set1_pd(0x10000000000000);
+ npyv_f64 sign_two52 = _mm_or_pd(two_power_52, _mm_and_pd(a, szero));
+ // round by add magic number 2^52
+ npyv_f64 round = _mm_sub_pd(_mm_add_pd(a, sign_two52), sign_two52);
+ return _mm_sub_pd(round, _mm_and_pd(_mm_cmpgt_pd(round, a), one));
+ }
+#endif // NPY_HAVE_SSE41
+
#endif // _NPY_SIMD_SSE_MATH_H
diff --git a/numpy/core/src/common/simd/vsx/math.h b/numpy/core/src/common/simd/vsx/math.h
index d138cae8a..94f1233f1 100644
--- a/numpy/core/src/common/simd/vsx/math.h
+++ b/numpy/core/src/common/simd/vsx/math.h
@@ -77,4 +77,8 @@ NPY_FINLINE npyv_f64 npyv_square_f64(npyv_f64 a)
#define npyv_trunc_f32 vec_trunc
#define npyv_trunc_f64 vec_trunc
+// floor
+#define npyv_floor_f32 vec_floor
+#define npyv_floor_f64 vec_floor
+
#endif // _NPY_SIMD_VSX_MATH_H
diff --git a/numpy/core/tests/test_simd.py b/numpy/core/tests/test_simd.py
index 12a67c44d..548d89574 100644
--- a/numpy/core/tests/test_simd.py
+++ b/numpy/core/tests/test_simd.py
@@ -331,12 +331,13 @@ class _SIMD_FP(_Test_Utility):
assert square == data_square
@pytest.mark.parametrize("intrin, func", [("self.ceil", math.ceil),
- ("self.trunc", math.trunc)])
+ ("self.trunc", math.trunc), ("self.floor", math.floor)])
def test_rounding(self, intrin, func):
"""
Test intrinsics:
npyv_ceil_##SFX
npyv_trunc_##SFX
+ npyv_floor##SFX
"""
intrin_name = intrin
intrin = eval(intrin)
@@ -360,6 +361,10 @@ class _SIMD_FP(_Test_Utility):
_round = self._to_unsigned(intrin(self.setall(w)))
data_round = self._to_unsigned(self.setall(-0.0))
assert _round == data_round
+ if "floor" in intrin_name:
+ _round = self._to_unsigned(intrin(self.setall(-0.0)))
+ data_round = self._to_unsigned(self.setall(-0.0))
+ assert _round == data_round
def test_max(self):
"""
View Lorry Controller Status