Use AVX2-style SIMDE path for RISC-V

Author: carlosqwqqwq

CMakeLists.txt

@@ -172,10 +172,13 @@ elseif(_ABPOA_WASM)
     target_compile_options(abpoa PRIVATE -msimd128)
 
 elseif(_ABPOA_RISCV)
-    # RISC-V: use the existing SIMDE-backed portable SIMD path without
-    # injecting host-specific march flags or x86 dispatch machinery.
+    # RISC-V: like ARM, use the existing SIMDE-backed AVX2-flavored path
+    # without injecting host-specific march flags or x86 dispatch machinery.
+    # Leaving __AVX2__ undefined falls back to the 128-bit SSE2-style path,
+    # which is not the project's validated portable configuration.
     message(STATUS "RISC-V target detected; building abPOA with the portable SIMDE path")
     target_sources(abpoa PRIVATE src/abpoa_align_simd.c)
+    target_compile_definitions(abpoa PRIVATE __AVX2__)
 
 else()
     # Fallback: single compile with -march=native

Makefile

@@ -73,7 +73,7 @@ endif
 endif
 
 ifneq ($(filter riscv%,$(ARCH)),)
-	SIMD_FLAG =
+	SIMD_FLAG = -D__AVX2__
 	OBJS = ${BASIC_OBJS} $(addprefix $(SRC_DIR)/, abpoa_align_simd.o)
 endif
 

README.md

@@ -80,7 +80,8 @@ significant speed improvement over existing tools.
 abPOA supports three alignment modes (global, local, extension) and flexible scoring schemes that allow linear, affine and convex gap penalties. 
 It right now supports SSE2/SSE4.1/AVX2 vectorization.
 For `riscv64`, the current portability path uses the existing SIMDE-backed
-generic build rather than x86 runtime dispatch or a dedicated RVV backend.
+AVX2-style generic build rather than x86 runtime dispatch or a dedicated RVV
+backend.
 
 For more information, please refer to our [paper](https://dx.doi.org/10.1093/bioinformatics/btaa963) published in Bioinformatics.
 

setup.py

@@ -27,7 +27,7 @@
     elif machine_arch in ["aarch32"]:
         simd_flag = ['-march=armv8-a+simd', '-mfpu=auto -D__AVX2__']
     elif machine_arch.startswith("riscv"):
-        simd_flag = []
+        simd_flag = ['-D__AVX2__']
     else: # x86_64
         simd_flag=['-march=native']
         if os.getenv('SSE4', False):