Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in / Register
Toggle navigation
S
simde
Project overview
Project overview
Details
Activity
Releases
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Issues
0
Issues
0
List
Boards
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Analytics
Analytics
CI / CD
Repository
Value Stream
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
Libraries
simde
Commits
02d81577
Commit
02d81577
authored
Jan 24, 2025
by
Michael R. Crusoe
Committed by
Michael R. Crusoe
Jan 25, 2025
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
arm neon FCMLA with 16-bit floats, requires the FP16 feature
parent
8caaee79
Changes
4
Show whitespace changes
Inline
Side-by-side
Showing
4 changed files
with
36 additions
and
36 deletions
+36
-36
arm/neon/cmla_lane.h
arm/neon/cmla_lane.h
+8
-8
arm/neon/cmla_rot180_lane.h
arm/neon/cmla_rot180_lane.h
+13
-13
arm/neon/cmla_rot270_lane.h
arm/neon/cmla_rot270_lane.h
+7
-7
arm/neon/cmla_rot90_lane.h
arm/neon/cmla_rot90_lane.h
+8
-8
No files found.
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,
return
simde_vcvt_f16_f32
(
simde_float32x4_from_private
(
r_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_lane_f16
...
...
@@ -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_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_laneq_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_lane_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#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
return
simde_vcvt_f16_f32
(
simde_float32x4_from_private
(
r_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot180_lane_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot180_lane_f16
...
...
@@ -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_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot180_laneq_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot180_laneq_f16
...
...
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
return
simde_vcvt_f16_f32
(
simde_float32x4_from_private
(
r_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot270_lane_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot270_lane_f16
...
...
@@ -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_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot270_laneq_f16
...
...
@@ -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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#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
return
simde_vcvt_f16_f32
(
simde_float32x4_from_private
(
r_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot90_lane_f16
...
...
@@ -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_
));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmla_rot90_laneq_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot90_lane_f16
...
...
@@ -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
)));
#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(__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)
#endif
#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(__clang__) || SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0))))
#undef vcmlaq_rot90_laneq_f16
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment
