From 4d476d894e12fcd854ec6725b1d2095d68b022ab Mon Sep 17 00:00:00 2001
From: jmorganca <jmorganca@gmail.com>
Date: Tue, 11 Jun 2024 23:18:07 -0700
Subject: [PATCH] fix `Makefile`

---
 llama/Makefile  | 23 +++++++++++++----------
 llama/README.md | 20 ++++++++++----------
 2 files changed, 23 insertions(+), 20 deletions(-)

diff --git a/llama/Makefile b/llama/Makefile
index 3b26da8d..029bb961 100644
--- a/llama/Makefile
+++ b/llama/Makefile
@@ -91,19 +91,22 @@ HIP_FLAGS := \
 	-I. \
 	$(foreach arch, $(HIP_ARCHS), --offload-arch=$(arch))
 
-ifeq ($(UNAME_S), Linux)
+ifeq ($(OS), Linux)
 	HIP_FLAGS += $(foreach arch, $(LINUX_HIP_ARCHS), --offload-arch=$(arch))
 endif
 
-ifeq ($(OS),Darwin)
-	ifeq ($(ARCH),arm64)
-		all: ollama_runner
-	else ifeq ($(ARCH),amd64)
-		all: ollama_runner ollama_runner_avx ollama_runner_avx2
-	endif
-else
-	all: ollama_runner ollama_runner_avx ollama_runner_avx2 ollama_runner_cuda ollama_runner_rocm
+RUNNERS := ollama_runner
+ifeq ($(ARCH),amd64)
+	RUNNERS += ollama_runner_avx ollama_runner_avx2
 endif
+ifeq ($(NVCC),)
+	RUNNERS += ollama_runner_cuda
+endif
+ifeq ($(HIPCC),)
+	RUNNERS += ollama_runner_rocm
+endif
+
+all: ollama_runner $(RUNNERS)
 
 %.cuda.$(OBJ_EXT): %.cu
 	$(NVCC) -c $(CUDA_FLAGS) -o $@ $<
@@ -147,4 +150,4 @@ 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 ollama_runner_avx ollama_runner_avx2 ollama_runner_cuda ollama_runner_rocm
+.PHONY: runners clean ollama_runner ollama_runner_avx ollama_runner_avx2 ollama_runner_cuda ollama_runner_rocm
diff --git a/llama/README.md b/llama/README.md
index a448c7db..31c6087d 100644
--- a/llama/README.md
+++ b/llama/README.md
@@ -12,12 +12,11 @@ Supported:
 - [x] Linux CUDA
 - [x] Linux ROCm
 - [x] Llava
-- [ ] Parallel Requests
 
-Extra build steps are required for CUDA and ROCm on Windows since `nvcc` and `hipcc` both require using msvc as the host compiler. For these small dlls are created:
+Extra build steps are required for CUDA and ROCm on Windows since `nvcc` and `hipcc` both require using msvc as the host compiler. For these shared libraries are created:
 
-- `ggml-cuda.dll`
-- `ggml-hipblas.dll`
+- `ggml_cuda.dll` on Windows or `ggml_cuda.so` on Linux
+- `ggml_hipblas.dll` on Windows or `ggml_hipblas.so` on Linux
 
 > Note: it's important that memory is allocated and freed by the same compiler (e.g. entirely by code compiled with msvc or mingw). Issues from this should be rare, but there are some places where pointers are returned by the CUDA or HIP runtimes and freed elsewhere, causing a a crash. In a future change the same runtime should be used in both cases to avoid crashes.
 
@@ -51,7 +50,7 @@ Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-down
 
 ```shell
 make ggml_cuda.so
-go build -tags=avx,cuda .
+go build -tags avx,cuda .
 ```
 
 ### ROCm
@@ -60,7 +59,7 @@ Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-down
 
 ```shell
 make ggml_hipblas.so
-go build -tags=avx,rocm .
+go build -tags avx,rocm .
 ```
 
 ## Windows
@@ -73,7 +72,7 @@ Install the [CUDA toolkit v11.3.1](https://developer.nvidia.com/cuda-11-3-1-down
 
 ```shell
 make ggml_cuda.dll
-go build -tags=avx,cuda .
+go build -tags avx,cuda .
 ```
 
 ### ROCm
@@ -82,13 +81,14 @@ Install [ROCm 5.7.1](https://rocm.docs.amd.com/en/docs-5.7.1/).
 
 ```shell
 make ggml_hipblas.dll
-go build -tags=rocm .
+go build -tags avx,rocm .
 ```
 
-Then build the package with the `rocm` tag:
+## Building runners
 
 ```shell
-go build -tags=avx,rocm .
+# build all runners for this platform
+make -j
 ```
 
 ## Syncing with llama.cpp