Add a new target which generates a C-only API shared library, compatible with MSVC

.github/workflows/c-api-linux.yaml

@@ -0,0 +1,51 @@
+name: Build C-API shared library with g++ for Linux
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  build-linux:
+    runs-on: ubuntu-latest
+    env:
+       NAME: x86simdsort
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Define variables
+        run: |
+          echo "LIBNAME=${NAME}-c-api" >> $GITHUB_ENV
+          echo "FOLDER=${NAME}-${GITHUB_SHA::8}" >> $GITHUB_ENV
+
+      - name: Install g++
+        run: |
+             sudo apt-get update
+             sudo apt-get install -y g++
+
+      - name: Build (Linux)
+        run: |
+          cd make-c-api
+          make all -j2
+
+      - name: Run executable (Linux sanity check)
+        run: |
+          cd make-c-api
+          LD_LIBRARY_PATH=$LD_LIBRARY_PATH:./ ./smoke.exe
+
+      - name: Package (Linux)
+        run: |
+          mkdir -p artifacts/${NAME}
+          cp make-c-api/*.{so,h} artifacts/${NAME}/
+          cd artifacts
+          tar -czvf ${NAME}-${GITHUB_SHA::8}.tgz ${NAME}
+
+      - name: Upload artifacts (Linux)
+        uses: actions/upload-artifact@v4
+        with:
+          name: c-api-linux-build
+          path: artifacts/*.tgz
+
+

.github/workflows/c-api-win-msvc.yaml

@@ -0,0 +1,71 @@
+name: MSYS2 DLL + MSVC EXE
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  build-windows-mingw:
+    runs-on: windows-latest
+    env:
+       NAME: x86simdsort
+
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@v4
+
+      - name: Define variables
+        run: |
+          echo "LIBNAME=${NAME}-c-api" >> $GITHUB_ENV
+          echo "FOLDER=${NAME}-${GITHUB_SHA::8}" >> $GITHUB_ENV
+
+      # Build DLL with MSYS2
+      - name: Install MSYS2
+        uses: msys2/setup-msys2@v2
+        with:
+           msystem: MINGW64
+           install: >-
+               mingw-w64-x86_64-toolchain
+               mingw-w64-x86_64-clang
+               make
+               zip
+
+      - name: Build DLL (MinGW/Clang)
+        shell: msys2 {0}
+        run: |
+          cd make-c-api
+          make all -j2 TARGET=DLL
+
+      - name: Package (Windows)
+        shell: msys2 {0}
+        run: |
+          mkdir -p artifacts/${NAME}
+          cp make-c-api/*.{lib,dll,h} artifacts/${NAME}/
+          cd artifacts
+          zip -r ${NAME}-${GITHUB_SHA::8}.zip ${NAME}
+
+       # --- Build EXE test with MSVC ---
+      - name: Configure MSVC environment
+        uses: ilammy/msvc-dev-cmd@v1
+        with:
+           arch: x64 # Or other architectures like x86, amd64_x86, etc.
+
+      - name: Build EXE with MSVC
+        run: |
+           cd make-c-api
+           cl /EHsc /O2 /arch:AVX2 /std:c++20 /GL /Gy /MD /nologo /DNDEBUG smoke.cpp x86simdsort-c-api.dll.lib /Fe:smoke.exe
+
+      # --- Run test ---
+      - name: Run EXE
+        shell: cmd
+        run: |
+          cd make-c-api
+          .\smoke.exe
+
+      - name: Upload artifacts (Windows)
+        uses: actions/upload-artifact@v4
+        with:
+          name: c-api-windows-build
+          path: artifacts/*.zip

.github/workflows/c-api-win.yaml

@@ -0,0 +1,45 @@
+name: Build C-API shared library with mingw-64-clang++ for Windows
+
+on:
+  push:
+    branches: [ main ]
+  pull_request:
+    branches: [ main ]
+
+jobs:
+  build-windows-mingw:
+    runs-on: ubuntu-latest
+    env:
+       NAME: x86simdsort
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Define variables
+        run: |
+          echo "LIBNAME=${NAME}-c-api" >> $GITHUB_ENV
+          echo "FOLDER=${NAME}-${GITHUB_SHA::8}" >> $GITHUB_ENV
+
+      - name: Install MinGW-w64
+        run: |
+          sudo apt-get update
+          sudo apt install -y mingw-w64
+          sudo apt install -y clang
+
+      - name: Build (Windows cross-compile DLL)
+        run: |
+          cd make-c-api
+          make all -j2 TARGET=DLL
+
+      - name: Package (Windows)
+        run: |
+          mkdir -p artifacts/${NAME}
+          cp make-c-api/*.{lib,dll,h} artifacts/${NAME}/
+          cd artifacts
+          zip -r ${NAME}-${GITHUB_SHA::8}.zip ${NAME}
+
+      - name: Upload artifacts (Windows)
+        uses: actions/upload-artifact@v4
+        with:
+          name: c-api-windows-build
+          path: artifacts/*.zip

README.md

@@ -109,6 +109,37 @@ xss_dep = xss.get_variable('x86simdsortcpp_dep')
 For more detailed instructions please refer to Meson
 [documentation](https://mesonbuild.com/Subprojects.html#using-a-subproject).
 
+## Build with C-only API, targeting Windows and MSVC
+
+The folder `make-c-api` contains a `Makefile` to build a shared library exporting only C-symbols.
+This avoid potential conflicts of inconsistent mangling and calling convention across compilers.
+
+Only a reduced set of functions are exported (`qselect`, `qsort`, `partial_qsort`, `keyvalue_qsort`, `keyvalue_partial_qsort`, `keyvalue_qselect`)
+and only for a reduces set of types (`int32`, `uint32`, `int64`, `uint64`, `float`, `double`).
+
+In the subfolder `make-c-api` a `Makefile` is provided which supports the following combinations of operating systems, compilers and targets:
+
+```
++--------------+------------------+------------+----------------------+---------------------+
+| Compilation  | Compiler         | Target     | Files Generated      | Build Command line  |
+| Platform     |                  | Platform   |                      |                     |
++--------------+------------------+------------+----------------------+---------------------+
+| LINUX        | g++              | LINUX      | libx86simdsort.so    | make                |
++--------------+------------------+------------+----------------------+---------------------+
+| CYGWIN       | clang++          | CYGWIN     | libx86simdsort.so    | make                |
++--------------+------------------+------------+----------------------+---------------------+
+| ANY          | mingw32 clang++  | Windows    | x86simdsort.dll      | make TARGET=DLL     |
+|              |                  |            | x86simdsort.dll.lib  |                     |
++--------------+------------------+------------+----------------------+---------------------+
+```
+
+In all cases, exceptions and memory allocations never cross the shared library boundary.
+In particular, if the target is a DLL, the library is self-contained, i.e. it does not depend on any other DLL.
+
+Compilation is tested with clang 20 and gcc 15.
+
+An include header `x86simdsort-c-api.h` and a `smoke.exe` executable test are also generated.
+
 ## Example usage
 
 #### Sort an array of floats

lib/c-api-def.h

@@ -0,0 +1,29 @@
+#include "c-api-headers.h"
+
+LIBRARY x86simdsortcpp
+EXPORTS
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(qsort,n)
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(qsort,n)
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(qselect,n)
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(qselect,n)
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(partial_qsort,n)
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(partial_qsort,n)
+#include "c-api-x-macro1.h"
+
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2)
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2)
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2)
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2)
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2)
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2)
+#include "c-api-x-macro2.h"

lib/c-api-export.h

@@ -0,0 +1,37 @@
+#include "c-api-headers.h"
+
+#ifdef XSS_EXPORTING
+#	if defined(__MINGW64__)
+#		define XSS_C_EXPORT __declspec(dllexport)
+#	else
+#		define XSS_C_EXPORT __attribute__((visibility("default")))
+#	endif
+#	define XSS_C_BODY(body) { try { body; return true; } catch(...) { return false; } }
+#else
+#	define XSS_C_EXPORT XSS_DLL_IMPORT
+#	define XSS_C_BODY(body) ;
+#endif
+
+#define XSS_XI1(n,t) XSS_C_EXPORT XSS_QSORT_HEADER_INT(n,t) XSS_C_BODY(x86simdsort::qsort(ar, size, false, descending))
+#define XSS_XF1(n,t) XSS_C_EXPORT XSS_QSORT_HEADER_FLT(n,t) XSS_C_BODY(x86simdsort::qsort(ar, size, hasnan, descending))
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXPORT XSS_QSELECT_HEADER_INT(n,t) XSS_C_BODY(x86simdsort::qselect(ar, k, size, false, descending))
+#define XSS_XF1(n,t) XSS_C_EXPORT XSS_QSELECT_HEADER_FLT(n,t) XSS_C_BODY(x86simdsort::qselect(ar, k, size, hasnan, descending))
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXPORT XSS_QPSORT_HEADER_INT(n,t) XSS_C_BODY(x86simdsort::partial_qsort(ar, k, size, false, descending))
+#define XSS_XF1(n,t) XSS_C_EXPORT XSS_QPSORT_HEADER_FLT(n,t) XSS_C_BODY(x86simdsort::partial_qsort(ar, k, size, hasnan, descending))
+#include "c-api-x-macro1.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVSORT_HEADER_INT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_qsort(keys, vals, size, false, descending))
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVSORT_HEADER_FLT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_qsort(keys, vals, size, hasnan, descending))
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVSEL_HEADER_INT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_select(keys, vals, k, size, false, descending))
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVSEL_HEADER_FLT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_select(keys, vals, k, size, hasnan, descending))
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVPSORT_HEADER_INT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_partial_sort(keys, vals, k, size, false, descending))
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXPORT XSS_QKVPSORT_HEADER_FLT(n1,t1,n2,t2) XSS_C_BODY(x86simdsort::keyvalue_partial_sort(keys, vals, k, size, hasnan, descending))
+#include "c-api-x-macro2.h"

lib/c-api-header-gen.h

@@ -0,0 +1,60 @@
+COMMENT This is an auto-generated file
+COMMENT This header is intended to be used for shared libraries (so or dll) compiled to the shared_c_api target
+
+PRAGMA ifndef __X86_SIMD_SORT_C_API_H__
+PRAGMA define __X86_SIMD_SORT_C_API_H__
+
+EMPTYLINE
+PRAGMA include <cstdint>
+EMPTYLINE
+
+PRAGMA define XSS_DLL_IMPORT
+
+#include "c-api-headers.h"
+
+EMPTYLINE
+COMMENT DLL import declarations
+
+PRAGMA ifdef __cplusplus
+extern "C" {
+PRAGMA endif
+#include "c-api-export.h"
+PRAGMA ifdef __cplusplus
+} COMMENT extern "C"
+PRAGMA endif
+
+EMPTYLINE
+PRAGMA ifdef __cplusplus
+COMMENT C++ overloaded dispatchers
+namespace xss
+{
+
+#define XSS_XI1(n,t) inline bool qsort(t* ar, uint64_t sz, bool desc) { return XSS_C_EXP_NAME1(qsort,n)( ar, sz, desc ); }
+#define XSS_XF1(n,t) inline bool qsort(t* ar, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME1(qsort,n)( ar, sz, hasnan, desc ); }
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) inline bool qselect(t* ar, uint64_t k, uint64_t sz, bool desc) { return XSS_C_EXP_NAME1(qselect,n)( ar, k, sz, desc ); }
+#define XSS_XF1(n,t) inline bool qselect(t* ar, uint64_t k, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME1(qselect,n)( ar, k, sz, hasnan, desc ); }
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) inline bool partial_qsort(t* ar, uint64_t k, uint64_t sz, bool desc) { return XSS_C_EXP_NAME1(partial_qsort,n)( ar, k, sz, desc ); }
+#define XSS_XF1(n,t) inline bool partial_qsort(t* ar, uint64_t k, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME1(partial_qsort,n)( ar, k, sz, hasnan, desc ); }
+#include "c-api-x-macro1.h"
+
+#define XSS_XI2(n1,t1, n2,t2) inline bool keyvalue_qsort(t1* keys, t2* vals, uint64_t sz, bool desc) { return XSS_C_EXP_NAME2(keyvalue_qsort,n1, n2)( keys, vals, sz, desc ); }
+#define XSS_XF2(n1,t1, n2,t2) inline bool keyvalue_qsort(t1* keys, t2* vals, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME2(keyvalue_qsort,n1, n2)( keys, vals, sz, hasnan, desc ); }
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) inline bool keyvalue_qselect(t1* keys, t2* vals, uint64_t k, uint64_t sz, bool desc) { return XSS_C_EXP_NAME2(keyvalue_qselect,n1, n2)( keys, vals, k, sz, desc ); }
+#define XSS_XF2(n1,t1, n2,t2) inline bool keyvalue_qselect(t1* keys, t2* vals, uint64_t k, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME2(keyvalue_qselect,n1, n2)( keys, vals, k, sz, hasnan, desc ); }
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) inline bool keyvalue_partial_qsort(t1* keys, t2* vals, uint64_t k, uint64_t sz, bool desc) { return XSS_C_EXP_NAME2(keyvalue_partial_qsort,n1, n2)( keys, vals, k, sz, desc ); }
+#define XSS_XF2(n1,t1, n2,t2) inline bool keyvalue_partial_qsort(t1* keys, t2* vals, uint64_t k, uint64_t sz, bool hasnan, bool desc)  { return XSS_C_EXP_NAME2(keyvalue_partial_qsort,n1, n2)( keys, vals, k, sz, hasnan, desc ); }
+#include "c-api-x-macro2.h"
+
+} COMMENT namespace xss
+
+PRAGMA endif COMMENT ifdef __cplusplus
+EMPTYLINE
+PRAGMA endif COMMENT ifndef __X86_SIMD_SORT_C_API_H__

lib/c-api-headers.h

@@ -0,0 +1,25 @@
+#pragma once
+
+#define XSS_C_EXP_NAME1(name, ty) c_xss_##name##_##ty
+#define XSS_C_EXP_NAME2(name, ty1, ty2) c_xss_##name##_##ty1##_##ty2
+
+#define XSS_QSORT_HEADER_INT(n,t)   bool XSS_C_EXP_NAME1(qsort, n)(t *ar, uint64_t size, bool descending)
+#define XSS_QSORT_HEADER_FLT(n,t)   bool XSS_C_EXP_NAME1(qsort, n)(t *ar, uint64_t size, bool hasnan, bool descending)
+
+#define XSS_QSELECT_HEADER_INT(n,t)   bool XSS_C_EXP_NAME1(qselect, n)(t *ar, uint64_t k, uint64_t size, bool descending)
+#define XSS_QSELECT_HEADER_FLT(n,t)   bool XSS_C_EXP_NAME1(qselect, n)(t *ar, uint64_t k, uint64_t size, bool hasnan, bool descending)
+
+#define XSS_QPSORT_HEADER_INT(n,t)   bool XSS_C_EXP_NAME1(partial_qsort, n)(t *ar, uint64_t k, uint64_t size, bool descending)
+#define XSS_QPSORT_HEADER_FLT(n,t)   bool XSS_C_EXP_NAME1(partial_qsort, n)(t *ar, uint64_t k, uint64_t size, bool hasnan, bool descending)
+
+#define XSS_QKVSORT_HEADER_INT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2)(t1 *keys, t2* vals, uint64_t size, bool descending)
+#define XSS_QKVSORT_HEADER_FLT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2)(t1 *keys, t2* vals, uint64_t size, bool hasnan, bool descending)
+
+#define XSS_QKVSEL_HEADER_INT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2)(t1 *keys, t2* vals, uint64_t k, uint64_t size, bool descending)
+#define XSS_QKVSEL_HEADER_FLT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2)(t1 *keys, t2* vals, uint64_t k, uint64_t size, bool hasnan, bool descending)
+
+#define XSS_QKVPSORT_HEADER_INT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2)(t1 *keys, t2* vals, uint64_t k, uint64_t size, bool descending)
+#define XSS_QKVPSORT_HEADER_FLT(n1,t1,n2,t2)   bool XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2)(t1 *keys, t2* vals, uint64_t k, uint64_t size, bool hasnan, bool descending)
+
+
+

lib/c-api-ver.h

@@ -0,0 +1,33 @@
+#include "c-api-headers.h"
+
+{
+  global:
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(qsort,n);
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(qsort,n);
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(qselect,n);
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(qselect,n);
+#include "c-api-x-macro1.h"
+
+#define XSS_XI1(n,t) XSS_C_EXP_NAME1(partial_qsort,n);
+#define XSS_XF1(n,t) XSS_C_EXP_NAME1(partial_qsort,n);
+#include "c-api-x-macro1.h"
+
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2);
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qsort, n1, n2);
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2);
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_qselect, n1, n2);
+#include "c-api-x-macro2.h"
+
+#define XSS_XI2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2);
+#define XSS_XF2(n1,t1, n2,t2) XSS_C_EXP_NAME2(keyvalue_partial_qsort, n1, n2);
+#include "c-api-x-macro2.h"
+
+  local:
+    *;
+};