avx512: naive implementation of fpclass

simde-math.h

@@ -434,6 +434,91 @@ simde_math_fpclassify(double v) {
   #endif
 }
 
+#define SIMDE_MATH_FP_QNAN      0x01
+#define SIMDE_MATH_FP_PZERO     0x02
+#define SIMDE_MATH_FP_NZERO     0x04
+#define SIMDE_MATH_FP_PINF      0x08
+#define SIMDE_MATH_FP_NINF      0x10
+#define SIMDE_MATH_FP_DENORMAL  0x20
+#define SIMDE_MATH_FP_NEGATIVE  0x40
+#define SIMDE_MATH_FP_SNAN      0x80
+
+static HEDLEY_INLINE
+int8_t
+simde_math_fpclassf(float v, const int imm8) {
+  union {
+    float f;
+    uint32_t u;
+  } fu;
+  fu.f = v;
+  uint32_t bits = fu.u;
+  int8_t NegNum = (bits >> 31) & 1;
+  uint32_t const ExpMask =  0x3F800000; // [30:23]
+  uint32_t const MantMask = 0x007FFFFF; // [22:0]
+  int8_t ExpAllOnes = ((bits & ExpMask) == ExpMask);
+  int8_t ExpAllZeros = ((bits & ExpMask) == 0);
+  int8_t MantAllZeros = ((bits & MantMask) == 0);
+  int8_t ZeroNumber = ExpAllZeros && MantAllZeros;
+  int8_t SignalingBit = (bits >> 22) & 1;
+
+  int8_t result = 0;
+  int8_t qNaN_res = ExpAllOnes && !MantAllZeros && SignalingBit;
+  int8_t Pzero_res = !NegNum && ExpAllZeros && MantAllZeros;
+  int8_t Nzero_res = NegNum && ExpAllZeros && MantAllZeros;
+  int8_t Pinf_res = !NegNum && ExpAllOnes && MantAllZeros;
+  int8_t Ninf_res = NegNum && ExpAllOnes && MantAllZeros;
+  int8_t Denorm_res = ExpAllZeros && !MantAllZeros;
+  int8_t FinNeg_res = NegNum && !ExpAllOnes && !ZeroNumber;
+  int8_t sNaN_res = ExpAllOnes && !MantAllZeros && !SignalingBit;
+  result = ((((imm8 >> 0) & 1) && qNaN_res)   || \
+            (((imm8 >> 1) & 1) && Pzero_res)  || \
+            (((imm8 >> 2) & 1) && Nzero_res)  || \
+            (((imm8 >> 3) & 1) && Pinf_res)   || \
+            (((imm8 >> 4) & 1) && Ninf_res)   || \
+            (((imm8 >> 5) & 1) && Denorm_res) || \
+            (((imm8 >> 6) & 1) && FinNeg_res) || \
+            (((imm8 >> 7) & 1) && sNaN_res));
+  return result;
+}
+
+static HEDLEY_INLINE
+int8_t
+simde_math_fpclass(double v, const int imm8) {
+  union {
+    double d;
+    uint64_t u;
+  } du;
+  du.d = v;
+  uint64_t bits = du.u;
+  int8_t NegNum = (bits >> 63) & 1;
+  uint64_t const ExpMask =  0x3FF0000000000000; // [62:52]
+  uint64_t const MantMask = 0x000FFFFFFFFFFFFF; // [51:0]
+  int8_t ExpAllOnes = ((bits & ExpMask) == ExpMask);
+  int8_t ExpAllZeros = ((bits & ExpMask) == 0);
+  int8_t MantAllZeros = ((bits & MantMask) == 0);
+  int8_t ZeroNumber = ExpAllZeros && MantAllZeros;
+  int8_t SignalingBit = (bits >> 51) & 1;
+
+  int8_t result = 0;
+  int8_t qNaN_res = ExpAllOnes && !MantAllZeros && SignalingBit;
+  int8_t Pzero_res = !NegNum && ExpAllZeros && MantAllZeros;
+  int8_t Nzero_res = NegNum && ExpAllZeros && MantAllZeros;
+  int8_t Pinf_res = !NegNum && ExpAllOnes && MantAllZeros;
+  int8_t Ninf_res = NegNum && ExpAllOnes && MantAllZeros;
+  int8_t Denorm_res = ExpAllZeros && !MantAllZeros;
+  int8_t FinNeg_res = NegNum && !ExpAllOnes && !ZeroNumber;
+  int8_t sNaN_res = ExpAllOnes && !MantAllZeros && !SignalingBit;
+  result = ((((imm8 >> 0) & 1) && qNaN_res)   || \
+            (((imm8 >> 1) & 1) && Pzero_res)  || \
+            (((imm8 >> 2) & 1) && Nzero_res)  || \
+            (((imm8 >> 3) & 1) && Pinf_res)   || \
+            (((imm8 >> 4) & 1) && Ninf_res)   || \
+            (((imm8 >> 5) & 1) && Denorm_res) || \
+            (((imm8 >> 6) & 1) && FinNeg_res) || \
+            (((imm8 >> 7) & 1) && sNaN_res));
+  return result;
+}
+
 /*** Manipulation functions ***/
 
 #if !defined(simde_math_nextafter)

x86/avx512.h

@@ -71,6 +71,7 @@
 #include "avx512/fmsub.h"
 #include "avx512/fnmadd.h"
 #include "avx512/fnmsub.h"
+#include "avx512/fpclass.h"
 #include "avx512/insert.h"
 #include "avx512/kshift.h"
 #include "avx512/knot.h"

x86/avx512/fpclass.h

+/* SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy,
+ * modify, merge, publish, distribute, sublicense, and/or sell copies
+ * of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Copyright:
+ *   2023      Michael R. Crusoe <crusoe@debian.org>
+ */
+
+#if !defined(SIMDE_X86_AVX512_FPCLASS_H)
+#define SIMDE_X86_AVX512_FPCLASS_H
+
+#include "types.h"
+
+HEDLEY_DIAGNOSTIC_PUSH
+SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
+SIMDE_BEGIN_DECLS_
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__mmask8
+simde_mm256_fpclass_ps_mask(simde__m256 a, int imm8)
+  SIMDE_REQUIRE_CONSTANT_RANGE(imm8, 0, 0x88) {
+  simde__mmask8 r = 0;
+  simde__m256_private a_ = simde__m256_to_private(a);
+
+  for (size_t i = 0 ; i < (sizeof(a_.f32) / sizeof(a_.f32[0])) ; i++) {
+    r |= simde_math_fpclassf(a_.f32[i], imm8) ? (UINT8_C(1) << i) : 0;
+  }
+  return r;
+}
+#if defined(SIMDE_X86_AVX512DQ_NATIVE) && defined(SIMDE_X86_AVX512VL_NATIVE)
+#  define simde_mm256_fpclass_ps_mask(a, imm8) _mm256_fpclass_ps_mask((a), (imm8))
+#endif
+#if defined(SIMDE_X86_AVX512DQ_ENABLE_NATIVE_ALIASES) && defined(SIMDE_X86_AVX512VL_ENABLE_NATIVE_ALIASES)
+#  undef _mm256_fpclass_ps_mask
+#  define _mm256_fpclass_ps_mask(a, imm8) simde_mm256_fpclass_ps_mask((a), (imm8))
+#endif
+
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__mmask8
+simde_mm512_fpclass_pd_mask(simde__m512d a, int imm8)
+  SIMDE_REQUIRE_CONSTANT_RANGE(imm8, 0, 0x88) {
+  simde__mmask8 r = 0;
+  simde__m512d_private a_ = simde__m512d_to_private(a);
+
+  for (size_t i = 0 ; i < (sizeof(a_.f64) / sizeof(a_.f64[0])) ; i++) {
+    r |= simde_math_fpclass(a_.f64[i], imm8) ? (UINT8_C(1) << i) : 0;
+  }
+  return r;
+}
+#if defined(SIMDE_X86_AVX512DQ_NATIVE)
+#  define simde_mm512_fpclass_pd_mask(a, imm8) _mm512_fpclass_pd_mask((a), (imm8))
+#endif
+#if defined(SIMDE_X86_AVX512DQ_ENABLE_NATIVE_ALIASES)
+#  undef _mm512_fpclass_pd_mask
+#  define _mm512_fpclass_pd_mask(a, imm8) simde_mm512_fpclass_pd_mask((a), (imm8))
+#endif
+
+SIMDE_END_DECLS_
+HEDLEY_DIAGNOSTIC_POP
+
+#endif /* !defined(SIMDE_X86_AVX512_FPCLASS_H) */