x86 avx512f: added fmaddsub implementation (#1246)

meson.build

@@ -356,6 +356,7 @@ simde_avx512_families = [
   'fixupimm_round',
   'flushsubnormal',
   'fmadd',
+  'fmaddsub',
   'fmsub',
   'fnmadd',
   'fnmsub',

simde/x86/avx512.h

@@ -69,6 +69,7 @@
 #include "avx512/fixupimm_round.h"
 #include "avx512/flushsubnormal.h"
 #include "avx512/fmadd.h"
+#include "avx512/fmaddsub.h"
 #include "avx512/fmsub.h"
 #include "avx512/fnmadd.h"
 #include "avx512/fnmsub.h"

simde/x86/avx512/fmaddsub.h

@@ -0,0 +1,91 @@
+#if !defined(SIMDE_X86_AVX512_FMADDSUB_H)
+#define SIMDE_X86_AVX512_FMADDSUB_H
+
+#include "types.h"
+#include "../fma.h"
+#include "mul.h"
+
+HEDLEY_DIAGNOSTIC_PUSH
+SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
+SIMDE_BEGIN_DECLS_
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__m512d
+simde_x_mm512_addsub_pd (simde__m512d a, simde__m512d b) {
+  //mm512_addsub_pd does not exist, but we define it for utility purposes (only with simde_x prefix, no native alias)
+  simde__m512d_private
+    r_,
+    a_ = simde__m512d_to_private(a),
+    b_ = simde__m512d_to_private(b);
+
+  #if SIMDE_NATURAL_VECTOR_SIZE_LE(256)
+    r_.m256d[0] = simde_mm256_addsub_pd(a_.m256d[0], b_.m256d[0]);
+    r_.m256d[1] = simde_mm256_addsub_pd(a_.m256d[1], b_.m256d[1]);
+  #else
+    SIMDE_VECTORIZE
+    for (size_t i = 0 ; i < (sizeof(r_.f64) / sizeof(r_.f64[0])) ; i += 2) {
+      r_.f64[  i  ] = a_.f64[  i  ] - b_.f64[  i  ];
+      r_.f64[i + 1] = a_.f64[i + 1] + b_.f64[i + 1];
+    }
+  #endif
+
+  return simde__m512d_from_private(r_);
+}
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__m512
+simde_x_mm512_addsub_ps (simde__m512 a, simde__m512 b) {
+  //mm512_addsub_ps does not exist, but we define it for utility purposes (only with simde_x prefix, no native alias)
+  simde__m512_private
+    r_,
+    a_ = simde__m512_to_private(a),
+    b_ = simde__m512_to_private(b);
+
+  #if SIMDE_NATURAL_VECTOR_SIZE_LE(256)
+    r_.m256[0] = simde_mm256_addsub_ps(a_.m256[0], b_.m256[0]);
+    r_.m256[1] = simde_mm256_addsub_ps(a_.m256[1], b_.m256[1]);
+  #else
+    SIMDE_VECTORIZE
+    for (size_t i = 0 ; i < (sizeof(r_.f32) / sizeof(r_.f32[0])) ; i += 2) {
+      r_.f32[  i  ] = a_.f32[  i  ] - b_.f32[  i  ];
+      r_.f32[i + 1] = a_.f32[i + 1] + b_.f32[i + 1];
+    }
+  #endif
+
+  return simde__m512_from_private(r_);
+}
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__m512d
+simde_mm512_fmaddsub_pd (simde__m512d a, simde__m512d b, simde__m512d c) {
+  #if defined(SIMDE_X86_AVX512F_NATIVE)
+    return _mm512_fmaddsub_pd(a, b, c);
+  #else
+    return simde_x_mm512_addsub_pd(simde_mm512_mul_pd(a, b), c);
+  #endif
+}
+#if defined(SIMDE_X86_AVX512F_ENABLE_NATIVE_ALIASES)
+  #undef _mm512_fmaddsub_pd
+  #define _mm512_fmaddsub_pd(a, b, c) simde_mm512_fmaddsub_pd(a, b, c)
+#endif
+
+
+SIMDE_FUNCTION_ATTRIBUTES
+simde__m512
+simde_mm512_fmaddsub_ps (simde__m512 a, simde__m512 b, simde__m512 c) {
+  #if defined(SIMDE_X86_AVX512F_NATIVE)
+    return _mm512_fmaddsub_ps(a, b, c);
+  #else
+    return simde_x_mm512_addsub_ps(simde_mm512_mul_ps(a, b), c);
+  #endif
+}
+#if defined(SIMDE_X86_AVX512F_ENABLE_NATIVE_ALIASES)
+  #undef _mm512_fmaddsub_ps
+  #define _mm512_fmaddsub_ps(a, b, c) simde_mm512_fmaddsub_ps(a, b, c)
+#endif
+
+
+SIMDE_END_DECLS_
+HEDLEY_DIAGNOSTIC_POP
+
+#endif /* !defined(SIMDE_X86_AVX512_FMADDSUB_H) */