Skip to content

Run frontier MoE models on hardware you already own β€” pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦

C

make: the cl.exe 64-bit check matches the arch token, not the English banner

JustVugg97887c414651
diff --git a/c/Makefile b/c/Makefile
index 893b3096..129ad8d3 100644
--- a/c/Makefile
+++ b/c/Makefile
@@ -755,7 +755,8 @@ backend_loader.o: backend_loader.c backend_cuda.h compat.h .build-config
cuda-dll: backend_cuda.cu backend_cuda.h
@command -v "$(NVCC)" >/dev/null 2>&1 || { echo "nvcc not found: set CUDA_HOME or NVCC" >&2; exit 1; }
@command -v cl >/dev/null 2>&1 || { echo "cl.exe (MSVC) not in PATH β€” run vcvars64.bat first" >&2; exit 1; }
- @cl 2>&1 | grep -q "for x64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
+ # The banner is localized ("for x64" / "per x64" / "pour x64"): match the arch token only (#1531).
+ @cl 2>&1 | grep -qE "x64|AMD64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
"$(NVCC)" $(NVCCFLAGS) -shared -DCOLI_CUDA_BUILDING_DLL \
-L"$(CUDA_HOME)/lib/x64" -lcudart \
backend_cuda.cu -o coli_cuda.dll
@@ -831,7 +832,7 @@ deepgemm-fetch $(DEEPGEMM_STAMP):
cuda-dsv4-dll: backend_cuda_dsv4.cu backend_cuda_dsv4.h dsv4.def
@command -v "$(NVCC)" >/dev/null 2>&1 || { echo "nvcc not found: set CUDA_HOME or NVCC" >&2; exit 1; }
@command -v cl >/dev/null 2>&1 || { echo "cl.exe (MSVC) not in PATH β€” run vcvars64.bat first" >&2; exit 1; }
- @cl 2>&1 | grep -q "for x64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
+ @cl 2>&1 | grep -qE "x64|AMD64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
"$(NVCC)" $(NVCCFLAGS) --expt-relaxed-constexpr --expt-extended-lambda -Xcompiler=/Zc:preprocessor -shared \
-Xlinker /DEF:dsv4.def \
-L"$(CUDA_HOME)/lib/x64" -lcublasLt -lcudart -lcuda \
@@ -844,7 +845,7 @@ DEEPGEMM_INC = -I"$(DEEPGEMM_HOME)/deep_gemm/include" -I"$(DEEPGEMM_HOME)/cutlas
cuda-dsv4-dg-dll: backend_cuda_dsv4.cu backend_cuda_dsv4.h dsv4.def $(DEEPGEMM_STAMP)
@command -v "$(NVCC)" >/dev/null 2>&1 || { echo "nvcc not found: set CUDA_HOME or NVCC" >&2; exit 1; }
@command -v cl >/dev/null 2>&1 || { echo "cl.exe (MSVC) not in PATH β€” run vcvars64.bat first" >&2; exit 1; }
- @cl 2>&1 | grep -q "for x64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
+ @cl 2>&1 | grep -qE "x64|AMD64" || { echo "cl.exe is the 32-bit (x86) compiler: nvcc then fails inside cuda_fp16.hpp with 'asm operand type size(8)'. Open 'x64 Native Tools Command Prompt for VS 2022' (or run vcvars64.bat), not the generic Developer Command Prompt (#1405)" >&2; exit 1; }
"$(NVCC)" -O3 -std=c++20 -ftz=false --expt-relaxed-constexpr --expt-extended-lambda -Xcompiler=/Zc:preprocessor \
-gencode arch=compute_120a,code=sm_120a -DCOLI_DSV4_DEEPGEMM \
$(DEEPGEMM_INC) -shared \