vulkan : fix FA mask load with bounds check (coopmat2) (llama/17606)

diff --git a/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp b/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp
index 617d85108698aa317d8e2363911c3885066aa84f..9a71996383d256eb98edcfa3d809f34911c0ee77 100644 (file)
--- a/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp
+++ b/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp
@@ -156,7 +156,7 @@ void main() {
                 tensorLayoutM = setTensorLayoutStrideNV(tensorLayoutM, m_stride, 1);
                 tensorLayoutM = setTensorLayoutClampValueNV(tensorLayoutM, 0xfc00); // -inf in float16_t
 
-                coopmat<float16_t, gl_ScopeWorkgroup, Br, Bc, gl_MatrixUseAccumulator> mv, mvmax;
+                coopmat<float16_t, gl_ScopeWorkgroup, Br, Bc, gl_MatrixUseAccumulator> mvmax;
 
                 coopMatLoadTensorNV(mv, data_m, m_offset, sliceTensorLayoutNV(tensorLayoutM, i * Br, Br, j * Bc, Bc));