ik_llama.cpp/github-data/issues/447 - Compilation Error_ Error C2676.md

📝 #447 - Compilation Error: Error C2676

Author	quasar-of-mikus
State	❌ Closed
Created	2025-05-23
Updated	2025-05-23

---

Description

Got this when trying to compile the latest commit. The last time I ran a build was commit 2ec2229 and that was successful. Windows 10

# usual command
cmake -B ./build -DGGML_CUDA=ON -DGGML_BLAS=OFF -DGGML_SCHED_MAX_COPIES=1
cmake --build ./build --config Release -j 20

Log:

  iqk_gemm_kquants.cpp
  iqk_gemm_ktquants.cpp
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_ktquants.cpp(47,61): error C2676: binary '^': '__m256i' does not define t
his operator or a conversion to a type acceptable to the predefined operator [C:\Textgen\ik_llama.cpp\build\ggml\src\gg
ml.vcxproj]
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_ktquants.cpp(83,46): error C2676: binary '^': '__m256i' does not define t
his operator or a conversion to a type acceptable to the predefined operator [C:\Textgen\ik_llama.cpp\build\ggml\src\gg
ml.vcxproj]
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_ktquants.cpp(120,65): error C2676: binary '^': '__m256i' does not define
this operator or a conversion to a type acceptable to the predefined operator [C:\Textgen\ik_llama.cpp\build\ggml\src\g
gml.vcxproj]
  iqk_gemm_iquants.cpp
  iqk_gemm_iqk_quants.cpp
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_iqk_quants.cpp(810,84): warning C4244: 'argument': conversion from 'const
 uint16_t' to 'char', possible loss of data [C:\Textgen\ik_llama.cpp\build\ggml\src\ggml.vcxproj]
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_iqk_quants.cpp(1279,34): message : see reference to function template ins
tantiation '__m256i `anonymous-namespace'::DequantizerIQ2KS::new_block<Q8<1,block_q8_K>>(int,const Q8 &,__m256 *)' bein
g compiled [C:\Textgen\ik_llama.cpp\build\ggml\src\ggml.vcxproj]
          with
          [
              Q8=Q8<1,block_q8_K>
          ]
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_iqk_quants.cpp(2050,1): message : see reference to function template inst
antiation 'void `anonymous-namespace'::mul_mat_qX_K_q8_K_T<Dequantizer,1>(int,const void *,size_t,const DataInfo &,int)
' being compiled [C:\Textgen\ik_llama.cpp\build\ggml\src\ggml.vcxproj]
          with
          [
              Dequantizer=`anonymous-namespace'::DequantizerIQ2KS
          ]
C:\Textgen\ik_llama.cpp\ggml\src\iqk\iqk_gemm_iqk_quants.cpp(2070,13): message : see reference to function template ins
tantiation 'void `anonymous-namespace'::set_functions<`anonymous-namespace'::DequantizerIQ2KS>(std::array<mul_mat_t,8>
&)' being compiled [C:\Textgen\ik_llama.cpp\build\ggml\src\ggml.vcxproj]
  iqk_gemm_1bit.cpp
  iqk_gemm_legacy_quants.cpp
  iqk_quantize.cpp
  Generating Code...

C:\Textgen\ik_llama.cpp>

---

💬 Conversation

👤 ikawrakow commented the 2025-05-23 at 12:10:01:

Does #448 fix it?

---

👤 quasar-of-mikus commented the 2025-05-23 at 12:30:39:

Yep, it compiles and runs fine with that PR. Don't know if this is related but I saw this message come up even though it built:

C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(554,1): error C3493: 'kBlockSize' cannot be implicit
ly captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\l
lama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(555,1): error C3493: 'kGroupSize' cannot be implicit
ly captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\l
lama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(679,1): error C3493: 'kNg' cannot be implicitly capt
ured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-qu
antize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(694,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(722,1): error C3493: 'kBlockSize' cannot be implicit
ly captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\l
lama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(777,1): error C3493: 'kNumVal' cannot be implicitly
captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llam
a-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(821,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
  llama-gguf.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf.exe
  llama-gguf-hash.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf-hash.exe
  llama-bench-matmult.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-bench-matmult.exe

---

👤 ikawrakow commented the 2025-05-23 at 12:56:37:

These are in the quantize-stats tool that fails to build (but everything else build correctly). Somehow MSVC disagrees with GCC and clang on the scope of constexpr's. Can you check if the commit I just pushed fixes it? Thanks.

---

👤 quasar-of-mikus commented the 2025-05-23 at 13:14:15:

No, on commit f015390 I am still getting an error, same as last time minus an error for kBlockSize:

C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(555,1): error C3493: 'kGroupSize' cannot be implicit
ly captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\l
lama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(678,1): error C3493: 'kNg' cannot be implicitly capt
ured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-qu
antize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(693,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(780,1): error C3493: 'kNumVal' cannot be implicitly
captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llam
a-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(824,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
  llama-gguf.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf.exe
  llama-gguf-hash.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf-hash.exe
  llama-bench-matmult.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-bench-matmult.exe

---

👤 quasar-of-mikus commented the 2025-05-23 at 13:14:15:

No, I am still getting an error, same as last time minus an error for kBlockSize:

C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(555,1): error C3493: 'kGroupSize' cannot be implicit
ly captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\l
lama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(678,1): error C3493: 'kNg' cannot be implicitly capt
ured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-qu
antize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(693,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(780,1): error C3493: 'kNumVal' cannot be implicitly
captured because no default capture mode has been specified [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llam
a-quantize-stats.vcxproj]
C:\Textgen\ik_llama.cpp\examples\quantize-stats\quantize-stats.cpp(824,5): error C2064: term does not evaluate to a fun
ction taking 0 arguments [C:\Textgen\ik_llama.cpp\build\examples\quantize-stats\llama-quantize-stats.vcxproj]
  llama-gguf.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf.exe
  llama-gguf-hash.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-gguf-hash.exe
  llama-bench-matmult.vcxproj -> C:\Textgen\ik_llama.cpp\build\bin\Release\llama-bench-matmult.exe

---

👤 ikawrakow commented the 2025-05-23 at 13:29:23:

And now?

I never work on Windows, but from what I hear from llama.cpp users clang produces faster code than MSVC.

---

👤 quasar-of-mikus commented the 2025-05-23 at 13:44:54:

It works now, no more errors during compilation.

from what I hear from llama.cpp users clang produces faster code than MSVC.

Cool, I'll compare with clang sometime