From a632a0442622f228878c4c5706b100dcb3424b58 Mon Sep 17 00:00:00 2001
From: jmorganca <jmorganca@gmail.com>
Date: Sun, 9 Jun 2024 23:53:40 -0400
Subject: [PATCH] fix output

---
 llama/.gitignore |  4 +++-
 llama/Makefile   | 34 +++++++++++++++++++++-------------
 llama/llama.go   |  2 +-
 3 files changed, 25 insertions(+), 15 deletions(-)

diff --git a/llama/.gitignore b/llama/.gitignore
index 9211745f..4290f9e1 100644
--- a/llama/.gitignore
+++ b/llama/.gitignore
@@ -3,5 +3,7 @@
 *.lib
 *.exp
 *.dll
+*.so
 *.o
-ollama_runner_*
\ No newline at end of file
+*.obj
+ollama_runner*
\ No newline at end of file
diff --git a/llama/Makefile b/llama/Makefile
index 76fab709..eb3ac783 100644
--- a/llama/Makefile
+++ b/llama/Makefile
@@ -3,21 +3,17 @@ ARCH := $(or $(ARCH), $(shell uname -m))
 NVCC := nvcc
 HIPCC := "$(HIP_PATH)/bin/hipcc"
 
-ifeq ($(UNAME_ARCH),x86_64)
+ifeq ($(ARCH),x86_64)
     ARCH := amd64
 endif
 
 # Determine object file extension based on OS
-ifeq ($(OS),Windows_NT)
+ifneq (,$(findstring MINGW,$(OS)))
     OBJ_EXT := obj
+	SHARED_EXT := dll
 else
     OBJ_EXT := o
-endif
-
-ifeq ($(OS),Windows_NT)
-    SHARED_EXT := dll
-else
-    SHARED_EXT := so
+	SHARED_EXT := so
 endif
 
 CUDA_SRCS := \
@@ -105,15 +101,27 @@ else
     all: ollama_runner ollama_runner_avx ollama_runner_avx2 ollama_runner_cuda ollama_runner_rocm
 endif
 
-%.cuda.$(OBJ_EXT): %.cu %.c %.cpp
-	nvcc -c $(CUDA_FLAGS) -o $@ $<
+%.cuda.$(OBJ_EXT): %.cu
+	$(NVCC) -c $(CUDA_FLAGS) -o $@ $<
 
-%.hip.$(OBJ_EXT): %.cu %.c %.cpp
-	$(HIPCC) -c $(HIP_FLAGS) -o $@ $<
+%.cuda.$(OBJ_EXT): %.c
+	$(NVCC) -c $(CFLAGS) -o $@ $<
+
+%.cuda.$(OBJ_EXT): %.cpp
+	$(NVCC) -c $(CXXFLAGS) -o $@ $<
 
 ggml_cuda.$(SHARED_EXT): $(CUDA_OBJS)
 	nvcc --shared -lcuda -lcublas -lcudart -lcublasLt $(CUDA_FLAGS) $(CUDA_OBJS) -o $@
 
+%.hip.$(OBJ_EXT): %.cu
+	$(HIPCC) -c $(HIP_FLAGS) -o $@ $<
+
+%.hip.$(OBJ_EXT): %.c
+	$(HIPCC) -c $(CFLAGS) -o $@ $<
+
+%.hip.$(OBJ_EXT): %.cpp
+	$(HIPCC) -c $(CXXFLAGS) -o $@ $<
+
 ggml_hipblas.$(SHARED_EXT): $(HIP_OBJS)
 	$(HIPCC) --shared -lhipblas -lamdhip64 -lrocblas $(HIP_OBJS) -o $@
 
@@ -135,4 +143,4 @@ ollama_runner_rocm: ggml_hipblas.dll
 clean:
 	rm -f $(CUDA_OBJS) $(HIP_OBJS) ggml_cuda.$(SHARED_EXT) ggml_cuda.* ggml_hipblas.* ollama_runner*
 
-.PHONY: all clean ollama_runner_cpu ollama_runner_cpu_avx ollama_runner_cpu_avx2 ollama_runner_cuda ollama_runner_rocm
+.PHONY: all clean ollama_runner ollama_runner_avx ollama_runner_avx2 ollama_runner_cuda ollama_runner_rocm
diff --git a/llama/llama.go b/llama/llama.go
index aa52fb2e..b0cfcc88 100644
--- a/llama/llama.go
+++ b/llama/llama.go
@@ -1,6 +1,6 @@
 package llama
 
-// #cgo CFLAGS: -std=c11 -DNDEBUG -DLOG_DISABLE_LOGS
+// #cgo CFLAGS: -std=c11 -DNDEBUG -DLOG_DISABLE_LOGS -Wno-discarded-qualifiers
 // #cgo CXXFLAGS: -std=c++11 -DNDEBUG -DLOG_DISABLE_LOGS
 // #cgo darwin,arm64 CFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64
 // #cgo darwin,arm64 CXXFLAGS: -DGGML_USE_METAL -DGGML_USE_ACCELERATE -DACCELERATE_NEW_LAPACK -DACCELERATE_LAPACK_ILP64