ggml: fix MSVC linkage conflict on the TurboQuant vec_dot

.github/workflows/turboquant-ci.yml

@@ -62,6 +62,32 @@ jobs:
       - name: Backend ops (setup + quantization paths)
         run: ./build/bin/test-backend-ops -o FLASH_ATTN_EXT
 
+  # MSVC rejects things gcc and clang accept silently. A duplicate declaration of
+  # the TurboQuant vec_dot — once with GGML_API, once without — was a hard error
+  # here (C2375, different linkage) and invisible everywhere else. Cheap on a
+  # public repo, and it covers a compiler neither other job does.
+  windows:
+    name: Windows MSVC compile check
+    runs-on: windows-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Configure
+        run: |
+          cmake -B build `
+            -DGGML_NATIVE=OFF `
+            -DLLAMA_BUILD_TESTS=ON `
+            -DLLAMA_BUILD_EXAMPLES=OFF `
+            -DLLAMA_BUILD_SERVER=OFF
+
+      # ggml-cpu is the target the linkage conflict broke; test-turboquant pulls
+      # in the quantization code itself and can actually run here.
+      - name: Build
+        run: cmake --build build --config Release --target ggml-cpu test-turboquant
+
+      - name: TurboQuant CPU reference tests
+        run: ./build/bin/Release/test-turboquant.exe
+
   cuda:
     name: CUDA compile check
     runs-on: ubuntu-latest

ggml/src/ggml-cpu/ggml-cpu.c

@@ -6,6 +6,7 @@
 #include "traits.h"
 #include "ggml-cpu-impl.h"
 #include "ggml-impl.h"
+#include "ggml-quants.h" // TurboQuant vec_dot lives in ggml-base
 #include "quants.h"
 #include "ggml-threading.h"
 #include "unary-ops.h"

ggml/src/ggml-cpu/quants.h

@@ -42,10 +42,10 @@ void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const voi
 void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
 void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
 
-// TurboQuant. Defined in ggml-quants.c (not arch-specific): dequantizes the
-// left-hand side internally, so the right-hand side is plain f32.
-void ggml_vec_dot_turbo3_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
-void ggml_vec_dot_turbo4_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
+// TurboQuant's vec_dot is declared in ggml-quants.h instead of here. It is the
+// only one defined in ggml-quants.c (ggml-base) rather than in this library, so
+// it needs GGML_API to cross the library boundary. Declaring it here as well —
+// without that attribute — is a linkage conflict MSVC rejects outright.
 
 void ggml_vec_dot_mxfp4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
 void ggml_vec_dot_nvfp4_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);