Skip to content

S
simde
Commit 02d81577 authored by Michael R. Crusoe's avatar Michael R. Crusoe Committed by Michael R. Crusoe
Browse files
Options

arm neon FCMLA with 16-bit floats, requires the FP16 feature

parent 8caaee79
Hide whitespace changes
Inline Side-by-side
Showing with 36 additions and 36 deletions
arm/neon/cmla_lane.h
View file @ 02d81577
...@@ -72,13 +72,13 @@ simde_float16x4_t simde_vcmla_lane_f16(simde_float16x4_t r, simde_float16x4_t a, ...@@ -72,13 +72,13 @@ simde_float16x4_t simde_vcmla_lane_f16(simde_float16x4_t r, simde_float16x4_t a,
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_lane_f16(r, a, b, lane) vcmla_lane_f16(r, a, b, lane) #define simde_vcmla_lane_f16(r, a, b, lane) vcmla_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_lane_f16 #undef vcmla_lane_f16
...@@ -161,13 +161,13 @@ simde_float16x4_t simde_vcmla_laneq_f16(simde_float16x4_t r, simde_float16x4_t a ...@@ -161,13 +161,13 @@ simde_float16x4_t simde_vcmla_laneq_f16(simde_float16x4_t r, simde_float16x4_t a
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_laneq_f16(r, a, b, lane) vcmla_laneq_f16(r, a, b, lane) #define simde_vcmla_laneq_f16(r, a, b, lane) vcmla_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_laneq_f16 #undef vcmla_laneq_f16
...@@ -258,13 +258,13 @@ simde_float16x8_t simde_vcmlaq_lane_f16(simde_float16x8_t r, simde_float16x8_t a ...@@ -258,13 +258,13 @@ simde_float16x8_t simde_vcmlaq_lane_f16(simde_float16x8_t r, simde_float16x8_t a
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_lane_f16(r, a, b, lane) vcmlaq_lane_f16(r, a, b, lane) #define simde_vcmlaq_lane_f16(r, a, b, lane) vcmlaq_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_lane_f16 #undef vcmlaq_lane_f16
...@@ -356,13 +356,13 @@ simde_float16x8_t simde_vcmlaq_laneq_f16(simde_float16x8_t r, simde_float16x8_t ...@@ -356,13 +356,13 @@ simde_float16x8_t simde_vcmlaq_laneq_f16(simde_float16x8_t r, simde_float16x8_t
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_laneq_f16(r, a, b, lane) vcmlaq_laneq_f16(r, a, b, lane) #define simde_vcmlaq_laneq_f16(r, a, b, lane) vcmlaq_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_laneq_f16 #undef vcmlaq_laneq_f16
......
arm/neon/cmla_rot180_lane.h
View file @ 02d81577
...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot180_lane_f16(simde_float16x4_t r, simde_float16 ...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot180_lane_f16(simde_float16x4_t r, simde_float16
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot180_lane_f16(r, a, b, lane) vcmla_rot180_lane_f16(r, a, b, lane) #define simde_vcmla_rot180_lane_f16(r, a, b, lane) vcmla_rot180_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot180_lane_f16 #undef vcmla_rot180_lane_f16
...@@ -176,13 +176,13 @@ simde_float16x8_t simde_vcmlaq_rot180_lane_f16(simde_float16x8_t r, simde_float1 ...@@ -176,13 +176,13 @@ simde_float16x8_t simde_vcmlaq_rot180_lane_f16(simde_float16x8_t r, simde_float1
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot180_lane_f16(r, a, b, lane) vcmlaq_rot180_lane_f16(r, a, b, lane) #define simde_vcmlaq_rot180_lane_f16(r, a, b, lane) vcmlaq_rot180_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot180_lane_f16 #undef vcmlaq_rot180_lane_f16
...@@ -269,13 +269,13 @@ simde_float16x4_t simde_vcmla_rot180_laneq_f16(simde_float16x4_t r, simde_float1 ...@@ -269,13 +269,13 @@ simde_float16x4_t simde_vcmla_rot180_laneq_f16(simde_float16x4_t r, simde_float1
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot180_laneq_f16(r, a, b, lane) vcmla_rot180_laneq_f16(r, a, b, lane) #define simde_vcmla_rot180_laneq_f16(r, a, b, lane) vcmla_rot180_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot180_laneq_f16 #undef vcmla_rot180_laneq_f16
...@@ -310,8 +310,8 @@ simde_float32x2_t simde_vcmla_rot180_laneq_f32(simde_float32x2_t r, simde_float3 ...@@ -310,8 +310,8 @@ simde_float32x2_t simde_vcmla_rot180_laneq_f32(simde_float32x2_t r, simde_float3
#endif #endif
return simde_float32x2_from_private(r_); return simde_float32x2_from_private(r_);
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot180_laneq_f32(r, a, b, lane) vcmla_rot180_laneq_f32(r, a, b, lane) #define simde_vcmla_rot180_laneq_f32(r, a, b, lane) vcmla_rot180_laneq_f32(r, a, b, lane)
#endif #endif
...@@ -369,13 +369,13 @@ simde_float16x8_t simde_vcmlaq_rot180_laneq_f16(simde_float16x8_t r, simde_float ...@@ -369,13 +369,13 @@ simde_float16x8_t simde_vcmlaq_rot180_laneq_f16(simde_float16x8_t r, simde_float
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot180_laneq_f16(r, a, b, lane) vcmlaq_rot180_laneq_f16(r, a, b, lane) #define simde_vcmlaq_rot180_laneq_f16(r, a, b, lane) vcmlaq_rot180_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot180_laneq_f16 #undef vcmlaq_rot180_laneq_f16
...@@ -412,8 +412,8 @@ simde_float32x4_t simde_vcmlaq_rot180_laneq_f32(simde_float32x4_t r, simde_float ...@@ -412,8 +412,8 @@ simde_float32x4_t simde_vcmlaq_rot180_laneq_f32(simde_float32x4_t r, simde_float
#endif #endif
return simde_float32x4_from_private(r_); return simde_float32x4_from_private(r_);
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot180_laneq_f32(r, a, b, lane) vcmlaq_rot180_laneq_f32(r, a, b, lane) #define simde_vcmlaq_rot180_laneq_f32(r, a, b, lane) vcmlaq_rot180_laneq_f32(r, a, b, lane)
#endif #endif
......
arm/neon/cmla_rot270_lane.h
View file @ 02d81577
...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot270_lane_f16(simde_float16x4_t r, simde_float16 ...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot270_lane_f16(simde_float16x4_t r, simde_float16
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot270_lane_f16(r, a, b, lane) vcmla_rot270_lane_f16(r, a, b, lane) #define simde_vcmla_rot270_lane_f16(r, a, b, lane) vcmla_rot270_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot270_lane_f16 #undef vcmla_rot270_lane_f16
...@@ -176,13 +176,13 @@ simde_float16x8_t simde_vcmlaq_rot270_lane_f16(simde_float16x8_t r, simde_float1 ...@@ -176,13 +176,13 @@ simde_float16x8_t simde_vcmlaq_rot270_lane_f16(simde_float16x8_t r, simde_float1
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot270_lane_f16(r, a, b, lane) vcmlaq_rot270_lane_f16(r, a, b, lane) #define simde_vcmlaq_rot270_lane_f16(r, a, b, lane) vcmlaq_rot270_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot270_lane_f16 #undef vcmlaq_rot270_lane_f16
...@@ -269,13 +269,13 @@ simde_float16x4_t simde_vcmla_rot270_laneq_f16(simde_float16x4_t r, simde_float1 ...@@ -269,13 +269,13 @@ simde_float16x4_t simde_vcmla_rot270_laneq_f16(simde_float16x4_t r, simde_float1
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot270_laneq_f16(r, a, b, lane) vcmla_rot270_laneq_f16(r, a, b, lane) #define simde_vcmla_rot270_laneq_f16(r, a, b, lane) vcmla_rot270_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot270_laneq_f16 #undef vcmla_rot270_laneq_f16
...@@ -375,7 +375,7 @@ simde_float16x8_t simde_vcmlaq_rot270_laneq_f16(simde_float16x8_t r, simde_float ...@@ -375,7 +375,7 @@ simde_float16x8_t simde_vcmlaq_rot270_laneq_f16(simde_float16x8_t r, simde_float
#define simde_vcmlaq_rot270_laneq_f16(r, a, b, lane) vcmlaq_rot270_laneq_f16(r, a, b, lane) #define simde_vcmlaq_rot270_laneq_f16(r, a, b, lane) vcmlaq_rot270_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot270_laneq_f16 #undef vcmlaq_rot270_laneq_f16
......
arm/neon/cmla_rot90_lane.h
View file @ 02d81577
...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot90_lane_f16(simde_float16x4_t r, simde_float16x ...@@ -76,13 +76,13 @@ simde_float16x4_t simde_vcmla_rot90_lane_f16(simde_float16x4_t r, simde_float16x
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot90_lane_f16(r, a, b, lane) vcmla_rot90_lane_f16(r, a, b, lane) #define simde_vcmla_rot90_lane_f16(r, a, b, lane) vcmla_rot90_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot90_lane_f16 #undef vcmla_rot90_lane_f16
...@@ -167,13 +167,13 @@ simde_float16x4_t simde_vcmla_rot90_laneq_f16(simde_float16x4_t r, simde_float16 ...@@ -167,13 +167,13 @@ simde_float16x4_t simde_vcmla_rot90_laneq_f16(simde_float16x4_t r, simde_float16
return simde_vcvt_f16_f32(simde_float32x4_from_private(r_)); return simde_vcvt_f16_f32(simde_float32x4_from_private(r_));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmla_rot90_laneq_f16(r, a, b, lane) vcmla_rot90_laneq_f16(r, a, b, lane) #define simde_vcmla_rot90_laneq_f16(r, a, b, lane) vcmla_rot90_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot90_laneq_f16 #undef vcmla_rot90_laneq_f16
...@@ -268,13 +268,13 @@ simde_float16x8_t simde_vcmlaq_rot90_lane_f16(simde_float16x8_t r, simde_float16 ...@@ -268,13 +268,13 @@ simde_float16x8_t simde_vcmlaq_rot90_lane_f16(simde_float16x8_t r, simde_float16
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot90_lane_f16(r, a, b, lane) vcmlaq_rot90_lane_f16(r, a, b, lane) #define simde_vcmlaq_rot90_lane_f16(r, a, b, lane) vcmlaq_rot90_lane_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot90_lane_f16 #undef vcmlaq_rot90_lane_f16
...@@ -370,13 +370,13 @@ simde_float16x8_t simde_vcmlaq_rot90_laneq_f16(simde_float16x8_t r, simde_float1 ...@@ -370,13 +370,13 @@ simde_float16x8_t simde_vcmlaq_rot90_laneq_f16(simde_float16x8_t r, simde_float1
simde_vcvt_f16_f32(simde_float32x4_from_private(r_high))); simde_vcvt_f16_f32(simde_float32x4_from_private(r_high)));
#endif #endif
} }
#if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && \ #if defined(SIMDE_ARM_NEON_A32V8_NATIVE) && SIMDE_ARCH_ARM_CHECK(8, 3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9, 0, 0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0)) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12, 0, 0))
#define simde_vcmlaq_rot90_laneq_f16(r, a, b, lane) vcmlaq_rot90_laneq_f16(r, a, b, lane) #define simde_vcmlaq_rot90_laneq_f16(r, a, b, lane) vcmlaq_rot90_laneq_f16(r, a, b, lane)
#endif #endif
#if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \ #if defined(SIMDE_ARM_NEON_A32V8_ENABLE_NATIVE_ALIASES) || (defined(SIMDE_ENABLE_NATIVE_ALIASES) && \
!(SIMDE_ARCH_ARM_CHECK(8,3) && \ !(SIMDE_ARCH_ARM_CHECK(8,3) && defined(SIMDE_ARM_NEON_FP16) && \
(!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \ (!defined(HEDLEY_GCC_VERSION) || HEDLEY_GCC_VERSION_CHECK(9,0,0)) && \
(!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0)))) (!defined(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot90_laneq_f16 #undef vcmlaq_rot90_laneq_f16
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment