Fixes to build with MSVC on Windows on Arm (#278)

* Correct build issues with MSVC compiler for Windows on Arm when compiling Neon code

Author: christophe0606

.github/workflows/runwinarm.yml

@@ -0,0 +1,34 @@
+name: Neon tests
+on:
+  workflow_dispatch:
+  push:
+    branches: [winarm]
+
+permissions: 
+  actions: read
+  security-events: write
+
+jobs:
+   CI_test_run: 
+    runs-on: windows-11-arm
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v5
+
+      - name: Setup Python 3.12
+        uses: actions/setup-python@v6
+        with:
+          python-version: '3.12'
+
+      - name: Install standard tools
+        run: |
+          choco install gow
+          refreshenv
+
+      - name: Prepare framework
+        run: |
+          cmd /c '"C:\Program Files\Microsoft Visual Studio\2022\Enterprise\Common7\Tools\VsDevCmd.bat" -arch=arm64 -host_arch=arm64 && C:\a\CMSIS-DSP\CMSIS-DSP\Testing\winarm_test\test.bat'
+         
+
+    

Include/arm_math_types.h

@@ -89,22 +89,30 @@ extern "C"
 #define __ALIGNED(x) __declspec(align(x))
 #define __WEAK
 #define SECTION_NOINIT
+#define NO_INLINE __declspec(noinline)
+
+
 #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
+#define NO_INLINE __attribute__ ((noinline))
 #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")))
+#define NO_INLINE __attribute__ ((noinline))
+
 #else
 #define SECTION_NOINIT
+#define NO_INLINE __attribute__ ((noinline))
+
 #include "cmsis_compiler.h"
 #endif
 

Include/arm_neon_tables.h

@@ -47,7 +47,8 @@ extern const float32_t arm_neon_rfft_twiddles_16_f32[ARM_NEON_RFFT_TWIDDLES_16_F
 #define ARM_NEON_RFFT_FACTORS_16_F32_LEN 4
 extern const uint32_t arm_neon_rfft_factors_16_f32[ARM_NEON_RFFT_FACTORS_16_F32_LEN];
 #define ARM_NEON_RFFT_TWIDDLES_NEON_16_F32_LEN 0
-extern const float32_t arm_neon_rfft_twiddles_neon_16_f32[ARM_NEON_RFFT_TWIDDLES_NEON_16_F32_LEN];
+extern const float32_t *arm_neon_rfft_twiddles_neon_16_f32;
+
 #define ARM_NEON_RFFT_FACTORS_NEON_16_F32_LEN 4
 extern const uint32_t arm_neon_rfft_factors_neon_16_f32[ARM_NEON_RFFT_FACTORS_NEON_16_F32_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_16_F32_LEN 24
@@ -60,7 +61,8 @@ extern const float32_t arm_neon_rfft_twiddles_32_f32[ARM_NEON_RFFT_TWIDDLES_32_F
 #define ARM_NEON_RFFT_FACTORS_32_F32_LEN 4
 extern const uint32_t arm_neon_rfft_factors_32_f32[ARM_NEON_RFFT_FACTORS_32_F32_LEN];
 #define ARM_NEON_RFFT_TWIDDLES_NEON_32_F32_LEN 0
-extern const float32_t arm_neon_rfft_twiddles_neon_32_f32[ARM_NEON_RFFT_TWIDDLES_NEON_32_F32_LEN];
+extern const float32_t *arm_neon_rfft_twiddles_neon_32_f32;
+
 #define ARM_NEON_RFFT_FACTORS_NEON_32_F32_LEN 4
 extern const uint32_t arm_neon_rfft_factors_neon_32_f32[ARM_NEON_RFFT_FACTORS_NEON_32_F32_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_32_F32_LEN 24
@@ -160,7 +162,8 @@ extern const float32_t arm_neon_rfft_super_twiddles_neon_4096_f32[ARM_NEON_RFFT_
 #define ARM_NE10_OFFSET_BACKWARD_TWID_NEON_4096_F32 1020
 
 #define ARM_NEON_RFFT_TWIDDLES_16_Q31_LEN 0
-extern const q31_t arm_neon_rfft_twiddles_16_q31[ARM_NEON_RFFT_TWIDDLES_16_Q31_LEN];
+extern const q31_t *arm_neon_rfft_twiddles_16_q31;
+
 #define ARM_NEON_RFFT_FACTORS_16_Q31_LEN 4
 extern const uint32_t arm_neon_rfft_factors_16_q31[ARM_NEON_RFFT_FACTORS_16_Q31_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_16_Q31_LEN 8
@@ -240,7 +243,8 @@ extern const q31_t arm_neon_rfft_super_twiddles_neon_8192_q31[ARM_NEON_RFFT_SUPE
 #define ARM_NE10_OFFSET_BACKWARD_TWID_RFFT_8192_Q31 4092
 
 #define ARM_NEON_RFFT_TWIDDLES_16_Q15_LEN 0
-extern const q15_t arm_neon_rfft_twiddles_16_q15[ARM_NEON_RFFT_TWIDDLES_16_Q15_LEN];
+extern const q15_t *arm_neon_rfft_twiddles_16_q15;
+
 #define ARM_NEON_RFFT_FACTORS_16_Q15_LEN 4
 extern const uint32_t arm_neon_rfft_factors_16_q15[ARM_NEON_RFFT_FACTORS_16_Q15_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_16_Q15_LEN 8

Include/arm_neon_tables_f16.h

@@ -92,7 +92,8 @@ extern const float16_t arm_neon_rfft_twiddles_16_f16[ARM_NEON_RFFT_TWIDDLES_16_F
 #define ARM_NEON_RFFT_FACTORS_16_F16_LEN 4
 extern const uint32_t arm_neon_rfft_factors_16_f16[ARM_NEON_RFFT_FACTORS_16_F16_LEN];
 #define ARM_NEON_RFFT_TWIDDLES_NEON_16_F16_LEN 0
-extern const float16_t arm_neon_rfft_twiddles_neon_16_f16[ARM_NEON_RFFT_TWIDDLES_NEON_16_F16_LEN];
+extern const float16_t *arm_neon_rfft_twiddles_neon_16_f16;
+
 #define ARM_NEON_RFFT_FACTORS_NEON_16_F16_LEN 4
 extern const uint32_t arm_neon_rfft_factors_neon_16_f16[ARM_NEON_RFFT_FACTORS_NEON_16_F16_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_16_F16_LEN 24
@@ -105,7 +106,8 @@ extern const float16_t arm_neon_rfft_twiddles_32_f16[ARM_NEON_RFFT_TWIDDLES_32_F
 #define ARM_NEON_RFFT_FACTORS_32_F16_LEN 4
 extern const uint32_t arm_neon_rfft_factors_32_f16[ARM_NEON_RFFT_FACTORS_32_F16_LEN];
 #define ARM_NEON_RFFT_TWIDDLES_NEON_32_F16_LEN 0
-extern const float16_t arm_neon_rfft_twiddles_neon_32_f16[ARM_NEON_RFFT_TWIDDLES_NEON_32_F16_LEN];
+extern const float16_t *arm_neon_rfft_twiddles_neon_32_f16;
+
 #define ARM_NEON_RFFT_FACTORS_NEON_32_F16_LEN 4
 extern const uint32_t arm_neon_rfft_factors_neon_32_f16[ARM_NEON_RFFT_FACTORS_NEON_32_F16_LEN];
 #define ARM_NEON_RFFT_SUPER_TWIDDLES_NEON_32_F16_LEN 24
@@ -206,7 +208,7 @@ extern const float16_t arm_neon_rfft_super_twiddles_neon_4096_f16[ARM_NEON_RFFT_
 
 
 
-#endif /* defined(ARM_MATH_NEON_FLOAT16) */
+#endif /* defined(ARM_MATH_NEON_FLOAT16) && defined(ARM_FLOAT16_SUPPORTED) */
 
 
 

Include/dsp/none.h

@@ -72,7 +72,7 @@ compiler file in Core or Core_A would not make sense.
         count += 1U;
         mask = mask >> 1U;
       }
-      return count;
+      return ((uint8_t)count);
     }
 
   __STATIC_FORCEINLINE int32_t __SSAT(int32_t val, uint32_t sat)

Ne10/CMSIS_NE10_fft.neonintrinsic.h

@@ -37,13 +37,13 @@
 #include <arm_neon.h>
 
 #define NE10_CPX_ADD_NEON_F32(Z,A,B) do {           \
-    Z.val[0] = A.val[0] + B.val[0];    \
-    Z.val[1] = A.val[1] + B.val[1];    \
+    Z.val[0] = vaddq_f32(A.val[0], B.val[0]);    \
+    Z.val[1] = vaddq_f32(A.val[1], B.val[1]);    \
 } while (0);
 
 #define NE10_CPX_SUB_NEON_F32(Z,A,B) do {           \
-    Z.val[0] = A.val[0] - B.val[0];    \
-    Z.val[1] = A.val[1] - B.val[1];    \
+    Z.val[0] = vsubq_f32(A.val[0] , B.val[0]);    \
+    Z.val[1] = vsubq_f32(A.val[1] , B.val[1]);    \
 } while (0);
 
 #define NE10_CPX_MUL_NEON_F32(Z,A,B) do {           \
@@ -111,16 +111,10 @@
 #define VDUPQ_N_F32(VAR) { VAR, VAR, VAR, VAR }
 
 #define CONST_TW_81   0.70710678f
-#define CONST_TW_81N -0.70710678f
-
-static const float32x4_t Q_TW_81    = VDUPQ_N_F32(CONST_TW_81 );
-static const float32x4_t Q_TW_81N   = VDUPQ_N_F32(CONST_TW_81N);
+#define CONST_TW_81N (-0.70710678f)
 
 #define DIV_TW81   1.4142136f
-#define DIV_TW81N -1.4142136f
-
-static const float32x4_t DIV_TW81_NEON  = VDUPQ_N_F32(DIV_TW81);
-static const float32x4_t DIV_TW81N_NEON = VDUPQ_N_F32(DIV_TW81N);
+#define DIV_TW81N (-1.4142136f)
 
 #define NE10_RADIX8x4_R2C_NEON_KERNEL_S1(Q_OUT,Q_IN) do {   \
         Q_OUT ## 0 = vaddq_f32 (Q_IN ## 0, Q_IN ## 4);      \
@@ -131,8 +125,8 @@ static const float32x4_t DIV_TW81N_NEON = VDUPQ_N_F32(DIV_TW81N);
         Q_OUT ## 5 = vsubq_f32 (Q_IN ## 2, Q_IN ## 6);      \
         Q_OUT ## 6 = vaddq_f32 (Q_IN ## 3, Q_IN ## 7);      \
         Q_OUT ## 7 = vsubq_f32 (Q_IN ## 3, Q_IN ## 7);      \
-        Q_OUT ## 3 = vmulq_f32 (Q_OUT ## 3, Q_TW_81 );      \
-        Q_OUT ## 7 = vmulq_f32 (Q_OUT ## 7, Q_TW_81N);      \
+        Q_OUT ## 3 = vmulq_n_f32 (Q_OUT ## 3, CONST_TW_81 );      \
+        Q_OUT ## 7 = vmulq_n_f32 (Q_OUT ## 7, CONST_TW_81N);      \
 } while(0);
 
 #define NE10_RADIX8x4_R2C_NEON_KERNEL_S2(Q_OUT,Q_IN) do {   \
@@ -172,8 +166,8 @@ static const float32x4_t DIV_TW81N_NEON = VDUPQ_N_F32(DIV_TW81N);
 } while (0);
 
 #define NE10_RADIX8x4_C2R_NEON_KERNEL_S2(Q_OUT,Q_IN) do {   \
-        Q_IN ## 3 = vmulq_f32(Q_IN ## 3,DIV_TW81_NEON);     \
-        Q_IN ## 7 = vmulq_f32(Q_IN ## 7,DIV_TW81N_NEON);    \
+        Q_IN ## 3 = vmulq_n_f32(Q_IN ## 3,DIV_TW81);     \
+        Q_IN ## 7 = vmulq_n_f32(Q_IN ## 7,DIV_TW81N);    \
         Q_OUT ## 0 = vaddq_f32(Q_IN ## 0, Q_IN ## 1);       \
         Q_OUT ## 4 = vsubq_f32(Q_IN ## 0, Q_IN ## 1);       \
         Q_OUT ## 1 = vaddq_f32(Q_IN ## 2, Q_IN ## 3);       \
@@ -330,8 +324,8 @@ static const float32x4_t DIV_TW81N_NEON = VDUPQ_N_F32(DIV_TW81N);
 
 #define NE10_RADIX4x4_R2C_TW_NEON_KERNEL_LAST(Q_OUT,Q_IN) do {  \
     float32x4_t Q_TMP;  \
-    Q_IN ## 1 = vmulq_f32(Q_IN ## 1, Q_TW_81);  \
-    Q_IN ## 3 = vmulq_f32(Q_IN ## 3, Q_TW_81);  \
+    Q_IN ## 1 = vmulq_n_f32(Q_IN ## 1, CONST_TW_81);  \
+    Q_IN ## 3 = vmulq_n_f32(Q_IN ## 3, CONST_TW_81);  \
     Q_TMP = vsubq_f32(Q_IN ## 1, Q_IN ## 3);    \
     Q_IN ## 3 = vaddq_f32(Q_IN ## 1, Q_IN ## 3);    \
     Q_IN ## 1 = Q_TMP;                      \
@@ -352,8 +346,8 @@ static const float32x4_t DIV_TW81N_NEON = VDUPQ_N_F32(DIV_TW81N);
     Q_TMP = vaddq_f32(Q_OUT ## 1, Q_OUT ## 3);  \
     Q_OUT ## 3 = vsubq_f32(Q_OUT ## 3, Q_OUT ## 1);  \
     Q_OUT ## 1 = Q_TMP; \
-    Q_OUT ## 1 = vmulq_f32( Q_OUT ## 1, DIV_TW81_NEON); \
-    Q_OUT ## 3 = vmulq_f32( Q_OUT ## 3, DIV_TW81_NEON); \
+    Q_OUT ## 1 = vmulq_n_f32( Q_OUT ## 1, DIV_TW81); \
+    Q_OUT ## 3 = vmulq_n_f32( Q_OUT ## 3, DIV_TW81); \
     Q_OUT ## 0 = vaddq_f32( Q_OUT ## 0, Q_OUT ## 0 );   \
     Q_OUT ## 2 = vaddq_f32( Q_OUT ## 2, Q_OUT ## 2 );   \
 } while(0);

Ne10/CMSIS_NE10_fft_generic_float16.neonintrisic.c

@@ -57,6 +57,7 @@ POSSIBILITY OF SUCH DAMAGE.
  * 
  */
 
+#include "arm_compiler_specific.h"
 #include <arm_neon.h>
 
 #include "CMSIS_NE10_types.h"