New Neon versions

* Neon version for complex number Mat x Mat

* Neon version for Mat x Vec

* Neon version for Matrix transpose

Author: christophe0606

.github/workflows/runneontest.yaml

@@ -105,9 +105,13 @@ jobs:
              Testing/build/result.html
              Testing/build/result_transform_neon.html
              Testing/build/result_binaryf32_neon.html
+             Testing/build/result_binary_complexf32_neon.html
              Testing/build/result_binaryf16_neon.html
+             Testing/build/result_binary_complexf16_neon.html
              Testing/build/result_binaryq31_neon.html
+             Testing/build/result_binary_complexq31_neon.html
              Testing/build/result_binaryq15_neon.html
+             Testing/build/result_binary_complexq15_neon.html
              Testing/build/result_binaryq7_neon.html
              Testing/build/result_f16.html
 
@@ -120,9 +124,13 @@ jobs:
           test "$(grep "FAILED" result.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_transform_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_binaryf32_neon.html | wc -l)" -eq 0
+          test "$(grep "FAILED" result_binary_complexf32_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_binaryf16_neon.html | wc -l)" -eq 0
+          test "$(grep "FAILED" result_binary_complexf16_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_binaryq31_neon.html | wc -l)" -eq 0
+          test "$(grep "FAILED" result_binary_complexq31_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_binaryq15_neon.html | wc -l)" -eq 0
+          test "$(grep "FAILED" result_binary_complexq15_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_binaryq7_neon.html | wc -l)" -eq 0
           test "$(grep "FAILED" result_f16.html | wc -l)" -eq 0
 

Examples/cmsis_build/examples_ac6.cbuild-set.yml

@@ -0,0 +1,4 @@
+cbuild-set:
+  generated-by: vscode-cmsis-csolution version 1.54.0
+  contexts:
+    - context: bayes.Release+VHT-Corstone-310

Examples/cmsis_build/projects/.clangd

@@ -0,0 +1,3 @@
+CompileFlags:
+  CompilationDatabase: >-
+    /Volumes/Ext/Embedded/CMSIS-DSP/Examples/cmsis_build/out/bayes/VHT-Corstone-310/Release

Examples/cmsis_build/projects/bayes.Release+VHT-Corstone-310.cprj

@@ -0,0 +1,25 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no" ?>
+<cprj schemaVersion="2.0.0" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="CPRJ.xsd">
+  <created timestamp="2025-06-06T09:14:58" tool="csolution 2.4.0"/>
+
+  <info isLayer="false">
+    <description>Automatically generated project</description>
+  </info>
+
+  <packages/>
+
+  <compilers>
+    <compiler name="AC6" version="6.18.0"/>
+  </compilers>
+
+  <target>
+    <output elf="bayes.axf" intdir="../tmp/bayes/VHT-Corstone-310/Release" name="bayes" outdir="../out/bayes/VHT-Corstone-310/Release" rtedir="RTE" type="exe"/>
+    <options debug="on"/>
+    <asflags add="-masm=auto" compiler="AC6"/>
+    <cflags add="-DSEMIHOSTING -Wsign-compare -Wdouble-promotion -DNDEBUG -Wall -Wextra -Werror -std=c11 -Ofast -ffast-math -Wno-packed -Wno-missing-variable-declarations -Wno-missing-prototypes -Wno-missing-noreturn -Wno-sign-conversion -Wno-nonportable-include-path -Wno-reserved-id-macro -Wno-unused-macros -Wno-documentation-unknown-command -Wno-documentation -Wno-license-management -Wno-parentheses-equality -Wno-reserved-identifier" compiler="AC6"/>
+    <cxxflags add="-fno-rtti -DNDEBUG -Wall -Wextra -std=c++11 -Ofast -ffast-math" compiler="AC6"/>
+    <ldflags add="--entry=Reset_Handler --info=summarysizes --info=sizes --info=totals --info=unused --info=veneers" compiler="AC6"/>
+    <includes>../../../Include;../../../PrivateInclude</includes>
+  </target>
+</cprj>
+

Include/arm_math_memory.h

@@ -42,6 +42,14 @@
 #define ARM_MATH_L3_CACHE_SIZE (512*1024)
 #endif
 
+#if defined(ARM_MATH_NEON)
+
+
+extern char ARM_CACHE_PACKEDB[];
+extern char ARM_CACHE_PACKEDA[];
+// PACKEDC contains the accumulators
+extern char ARM_CACHE_PACKEDC[];
+#endif
 
 #ifdef   __cplusplus
 extern "C"

Include/arm_math_types.h

@@ -88,19 +88,23 @@ extern "C"
 #define __STATIC_INLINE static __inline
 #define __ALIGNED(x) __declspec(align(x))
 #define __WEAK
+#define SECTION_NOINIT
 #elif defined ( __APPLE_CC__ )
 #include <stdint.h>
 #define  __ALIGNED(x) __attribute__((aligned(x)))
 #define __STATIC_FORCEINLINE static inline __attribute__((always_inline)) 
 #define __STATIC_INLINE static inline
 #define __WEAK
+#define SECTION_NOINIT
 #elif defined (__GNUC_PYTHON__)
 #include <stdint.h>
 #define  __ALIGNED(x) __attribute__((aligned(x)))
 #define __STATIC_FORCEINLINE static inline __attribute__((always_inline)) 
 #define __STATIC_INLINE static inline
 #define __WEAK
+#define SECTION_NOINIT __attribute__((section(".noinit")))
 #else
+#define SECTION_NOINIT
 #include "cmsis_compiler.h"
 #endif
 
@@ -120,8 +124,9 @@ extern "C"
  * 
  * ARM_MATH_NEON is used to enable the Neon variants. 
  * When Neon variants are enabled, the DSP extension are disabled
+ * 
  */
-#if !defined(__GNUC_PYTHON__) && !defined(ARM_MATH_NEON) && !defined(ARM_MATH_NEON_EXPERIMENTAL)
+#if (!defined(__GNUC_PYTHON__) && !defined(ARM_MATH_NEON) && !defined(ARM_MATH_NEON_EXPERIMENTAL)) 
 #if (defined (__ARM_FEATURE_DSP) && (__ARM_FEATURE_DSP == 1))
   #define ARM_MATH_DSP                   1
 #endif
@@ -438,7 +443,7 @@ extern "C"
 
 #endif
 
-#if defined(ARM_MATH_NEON) || (defined(ARM_MATH_MVEF)  && !defined(ARM_MATH_AUTOVECTORIZE)) /* floating point vector*/
+#if defined(ARM_MATH_NEON) || (defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)) /* floating point vector*/
 
   /**
    * @brief 32-bit floating-point 128-bit vector type
@@ -466,7 +471,7 @@ extern "C"
 
 #endif
 
-#if defined(ARM_MATH_NEON)
+#if defined(ARM_MATH_NEON) 
   /**
    * @brief 32-bit fractional 64-bit vector data type in 1.31 format.
    */

Include/arm_neon_tables.h

@@ -39,7 +39,7 @@ extern "C"
 
 
 
-#if defined(ARM_MATH_NEON) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if defined(ARM_MATH_NEON)
 
 
 #define ARM_NEON_RFFT_TWIDDLES_16_F32_LEN 24
@@ -456,7 +456,7 @@ extern const uint32_t arm_neon_factors_4096_q15[ARM_NEON_FACTORS_4096_Q15_LEN];
 #define ARM_NE10_OFFSET_BACKWARD_TWID_4096_Q15 4092
 
 
-#endif /* defined(ARM_MATH_NEON) && !defined(ARM_MATH_AUTOVECTORIZE) */
+#endif /* defined(ARM_MATH_NEON) */
 
 
 

Include/arm_neon_tables_f16.h

@@ -39,7 +39,7 @@ extern "C"
 
 
 
-#if defined(ARM_MATH_NEON_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if defined(ARM_MATH_NEON_FLOAT16) && defined(ARM_FLOAT16_SUPPORTED)
 
 
 #define ARM_NEON_TWIDDLES_16_F16_LEN 24
@@ -206,7 +206,7 @@ extern const float16_t arm_neon_rfft_super_twiddles_neon_4096_f16[ARM_NEON_RFFT_
 
 
 
-#endif /* defined(ARM_MATH_NEON_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) */
+#endif /* defined(ARM_MATH_NEON_FLOAT16) */
 
 
 

Include/arm_vec_math.h

@@ -299,7 +299,7 @@ __STATIC_INLINE f32x4_t vpowq_f32(
 #if (defined(ARM_MATH_MVEI) || defined(ARM_MATH_HELIUM)) && !defined(ARM_MATH_AUTOVECTORIZE)
 #endif /* (defined(ARM_MATH_MVEI) || defined(ARM_MATH_HELIUM)) */
 
-#if (defined(ARM_MATH_NEON) || defined(ARM_MATH_NEON_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if defined(ARM_MATH_NEON) || defined(ARM_MATH_NEON_EXPERIMENTAL)
 
 #include "NEMath.h"
 /**
@@ -358,7 +358,7 @@ __STATIC_INLINE int32x4_t __arm_vec_sqrt_q31_neon(int32x4_t vec)
   return(vcvtq_n_s32_f32(temp,31));
 }
 
-#endif /*  (defined(ARM_MATH_NEON) || defined(ARM_MATH_NEON_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE) */
+#endif /*  defined(ARM_MATH_NEON) || defined(ARM_MATH_NEON_EXPERIMENTAL)*/
 
 #ifdef   __cplusplus
 }

Include/dsp/filtering_functions.h

@@ -1255,7 +1255,7 @@ arm_status arm_fir_decimate_init_f32(
         uint32_t blockSize);
 
 
-#if defined(ARM_MATH_NEON) || defined(DOXYGEN)
+#if (defined(ARM_MATH_NEON) || defined(DOXYGEN))
 /**
   @brief         Compute new coefficient arrays for use in vectorized filter (Neon only).
   @param[in]     numStages         number of 2nd order stages in the filter.