CMSIS-DSP: Added new compilation define for MVE.

Include/arm_math.h

@@ -143,7 +143,13 @@
    *
    * - ARM_MATH_HELIUM:
    *
-   * It implies the flags ARM_MATH_MVEF and ARM_MATH_MVEI and ARM_MATH_FLOAT16.
+   * It implies the flags ARM_MATH_MVEF and ARM_MATH_MVEI and ARM_MATH_MVE_FLOAT16.
+   *
+   * - ARM_MATH_HELIUM_EXPERIMENTAL:
+   *
+   * Only taken into account when ARM_MATH_MVEF, ARM_MATH_MVEI or ARM_MATH_MVE_FLOAT16 are defined.
+   * Enable some vector versions which may have worse performance than scalar
+   * depending on the core / compiler configuration.
    *
    * - ARM_MATH_MVEF:
    *

Source/CMakeLists.txt

@@ -14,6 +14,7 @@ include(configLib)
 
 option(NEON "Neon acceleration" OFF)
 option(NEONEXPERIMENTAL "Neon experimental acceleration" OFF)
+option(HELIUMEXPERIMENTAL "Helium experimental acceleration" OFF)
 option(LOOPUNROLL "Loop unrolling" ON)
 option(ROUNDING "Rounding" OFF)
 option(MATRIXCHECK "Matrix Checks" OFF)

Source/FilteringFunctions/arm_biquad_cascade_df2T_f16.c

@@ -47,7 +47,7 @@
   @return        none
  */
 
-#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if (defined(ARM_MATH_MVE_FLOAT16) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE)
 void arm_biquad_cascade_df2T_f16(
   const arm_biquad_cascade_df2T_instance_f16 * S,
   const float16_t * pSrc,

Source/FilteringFunctions/arm_biquad_cascade_df2T_f32.c

@@ -45,7 +45,7 @@
   @param[in]     blockSize number of samples to process
   @return        none
  */
-#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if (defined(ARM_MATH_MVEF) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE)
 #include "arm_helium_utils.h"
 
 void arm_biquad_cascade_df2T_f32(

Source/FilteringFunctions/arm_biquad_cascade_stereo_df2T_f16.c

@@ -51,7 +51,7 @@
 #pragma GCC warning "Scalar version of arm_biquad_cascade_stereo_df2T_f16 built. Helium version has build issues with gcc."
 #endif 
 
-#if defined(ARM_MATH_MVE_FLOAT16) && !defined(ARM_MATH_AUTOVECTORIZE) && !defined(__CMSIS_GCC_H)
+#if (defined(ARM_MATH_MVE_FLOAT16) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE) && !defined(__CMSIS_GCC_H)
 void arm_biquad_cascade_stereo_df2T_f16(
   const arm_biquad_cascade_stereo_df2T_instance_f16 * S,
   const float16_t * pSrc,

Source/FilteringFunctions/arm_biquad_cascade_stereo_df2T_f32.c

@@ -45,7 +45,7 @@
   @param[in]     blockSize number of samples to process
   @return        none
  */
-#if defined(ARM_MATH_MVEF) && !defined(ARM_MATH_AUTOVECTORIZE)
+#if (defined(ARM_MATH_MVEF) && defined(ARM_MATH_HELIUM_EXPERIMENTAL)) && !defined(ARM_MATH_AUTOVECTORIZE)
 #include "arm_helium_utils.h"
 
 void arm_biquad_cascade_stereo_df2T_f32(

cmsisdspconfig.py

@@ -27,6 +27,7 @@ config["MVEI"]=False
 config["MVEF"]=False
 config["NEON"]=False
 config["HELIUM"]=False
+config["HELIUMEXPERIMENTAL"]=False
 config["Float16"]=True
 config["HOST"]=False
 
@@ -365,6 +366,9 @@ def interpretCmakeOptions(cmake):
     if test(cmake,"MVEF"):
        r.append("-DARM_MATH_MVEF")
 
+    if test(cmake,"HELIUMEXPERIMENTAL"):
+       r.append("-DARM_MATH_HELIUM_EXPERIMENTAL")
+
     if test(cmake,"HELIUM") or test(cmake,"MVEF") or test(cmake,"MVEI"):
        r.append("-IPrivateInclude")
 
@@ -508,6 +512,7 @@ if not forHost:
    check(config,"ROUNDING")
    check(config,"MATRIXCHECK")
    
+   st.sidebar.header('Vector extensions')
    st.sidebar.info("Enable vector code. It is not automatic for Neon. Use of Helium will enable new options to select some interpolation tables.")
    archi=st.sidebar.selectbox("Vector",('None','Helium','Neon'))
    if archi == 'Neon':
@@ -515,6 +520,8 @@ if not forHost:
    if archi == 'Helium':
       multiselect(config,"MVE configuration",["MVEI","MVEF"])
       HELIUM=True
+      st.sidebar.info("When checked some experimental versions will be enabled and may be less performant than scalar version depending on the architecture.")
+      check(config,"HELIUMEXPERIMENTAL")
    if archi != 'None':
        st.sidebar.info("When autovectorization is on, pure C code will be compiled. The version with C intrinsics won't be compiled.")
        check(config,"AUTOVECTORIZE")

configCore.cmake

@@ -280,6 +280,10 @@ function(configcore PROJECTNAME ROOT)
     target_compile_definitions(${PROJECTNAME} PRIVATE ARM_MATH_NEON_EXPERIMENTAL)
   endif()
 
+  if (HELIUMEXPERIMENTAL)
+    target_compile_definitions(${PROJECTNAME} PRIVATE ARM_MATH_HELIUM_EXPERIMENTAL)
+  endif()
+
   if (HELIUM AND CORTEXM)
     target_compile_definitions(${PROJECTNAME} PRIVATE ARM_MATH_HELIUM)
   endif()