From 794a38a2e895a34b957a77707e5510ef9c4468c7 Mon Sep 17 00:00:00 2001
From: Concedo <39025047+LostRuins@users.noreply.github.com>
Date: Fri, 21 Apr 2023 21:02:40 +0800
Subject: [PATCH] Revert "cublas is not feasible at this time. removed for now"

This reverts commit 3687db7cf7b45c4e9cca59aa08732f3fd472d9f2.
---
 Makefile | 19 ++++++++++++++++---
 1 file changed, 16 insertions(+), 3 deletions(-)

diff --git a/Makefile b/Makefile
index 9a8e472b1..3d12d6685 100644
--- a/Makefile
+++ b/Makefile
@@ -1,4 +1,4 @@
-default: koboldcpp koboldcpp_noavx2 koboldcpp_openblas koboldcpp_openblas_noavx2 koboldcpp_clblast
+default: koboldcpp koboldcpp_noavx2 koboldcpp_openblas koboldcpp_openblas_noavx2 koboldcpp_clblast koboldcpp_cublas
 simple: koboldcpp koboldcpp_noavx2
 dev: koboldcpp_openblas
 
@@ -55,6 +55,7 @@ BONUSCFLAGS2 =
 
 OPENBLAS_FLAGS = -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
 CLBLAST_FLAGS = -DGGML_USE_CLBLAST -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
+CUBLAS_FLAGS = -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I./include/cuda -I./include/cuda/crt
 
 #lets try enabling everything
 CFLAGS   += -pthread -s
@@ -123,7 +124,6 @@ ifndef LLAMA_NO_ACCELERATE
 	endif
 endif
 
-#note: koboldcpp does not officially support cublas. You can manually link it if you want, but it will not be a regular feature
 ifdef LLAMA_CUBLAS
 	CFLAGS  += -DGGML_USE_CUBLAS -I/usr/local/cuda/include
 	LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64
@@ -157,6 +157,7 @@ NOAVX2_BUILD =
 OPENBLAS_BUILD =
 OPENBLAS_NOAVX2_BUILD =
 CLBLAST_BUILD =
+CUBLAS_BUILD =
 
 ifeq ($(OS),Windows_NT)
 	DEFAULT_BUILD = $(CXX) $(CXXFLAGS)  $^ -shared -o $@.dll $(LDFLAGS)
@@ -164,6 +165,7 @@ ifeq ($(OS),Windows_NT)
 	OPENBLAS_BUILD = $(CXX) $(CXXFLAGS) $^ lib/libopenblas.lib -shared -o $@.dll $(LDFLAGS)
 	OPENBLAS_NOAVX2_BUILD = $(CXX) $(CXXFLAGS) $^ lib/libopenblas.lib -shared -o $@.dll $(LDFLAGS)
 	CLBLAST_BUILD = $(CXX) $(CXXFLAGS) $^ lib/OpenCL.lib lib/clblast.lib -shared -o $@.dll $(LDFLAGS)
+	CUBLAS_BUILD = $(CXX) $(CXXFLAGS) $^  -shared -o $@.dll $(LDFLAGS)
 else
 	DEFAULT_BUILD = $(CXX) $(CXXFLAGS)  $^ -shared -o $@.so $(LDFLAGS)
 	NOAVX2_BUILD = $(CXX) $(CXXFLAGS) $^ -shared -o $@.so $(LDFLAGS)
@@ -174,12 +176,17 @@ else
 	ifdef LLAMA_CLBLAST
 	CLBLAST_BUILD = $(CXX) $(CXXFLAGS) $^ -lclblast -lOpenCL $(ARCH_ADD) -lopenblas -shared -o $@.so $(LDFLAGS)
 	endif
+	ifdef LLAMA_CUBLAS
+	CUBLAS_BUILD = $(CXX) $(CXXFLAGS) $^ -lcublas_static -lculibos -lcudart_static -lcublasLt_static -lpthread -ldl -L/usr/local/cuda/lib64 -shared -o $@.so $(LDFLAGS)
+	endif
 
 	ifndef LLAMA_OPENBLAS
 	ifndef LLAMA_CLBLAST
+	ifndef LLAMA_CUBLAS
 	OPENBLAS_BUILD = @echo 'Your OS $(OS) does not appear to be Windows. For faster speeds, install and link a BLAS library. Set LLAMA_OPENBLAS=1 to compile with OpenBLAS support or LLAMA_CLBLAST=1 to compile with ClBlast support. This is just a reminder, not an error.'
 	endif
 	endif
+	endif
 endif
 
 #
@@ -216,6 +223,9 @@ ggml_openblas_noavx2.o: ggml.c ggml.h
 ggml_clblast.o: ggml.c ggml.h
 	$(CC)  $(CFLAGS) $(BONUSCFLAGS1) $(BONUSCFLAGS2) $(CLBLAST_FLAGS) -c $< -o $@
 
+ggml_cublas.o: ggml.c ggml.h
+	$(CC)  $(CFLAGS) $(BONUSCFLAGS1) $(BONUSCFLAGS2) $(CUBLAS_FLAGS) -c $< -o $@
+
 ggml_v1.o: otherarch/ggml_v1.c otherarch/ggml_v1.h
 	$(CC)  $(CFLAGS) $(BONUSCFLAGS1) $(BONUSCFLAGS2) -c $< -o $@
 
@@ -238,7 +248,7 @@ gpttype_adapter.o: gpttype_adapter.cpp
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
 clean:
-	rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize-stats perplexity embedding benchmark-q4_0-matmult main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe koboldcpp.dll koboldcpp_openblas.dll koboldcpp_noavx2.dll koboldcpp_openblas_noavx2.dll koboldcpp_clblast.dll koboldcpp.so koboldcpp_openblas.so koboldcpp_noavx2.so koboldcpp_openblas_noavx2.so koboldcpp_clblast.so gptj.exe gpt2.exe
+	rm -vf *.o main quantize_llama quantize_gpt2 quantize_gptj quantize-stats perplexity embedding benchmark-q4_0-matmult main.exe quantize_llama.exe quantize_gptj.exe quantize_gpt2.exe koboldcpp.dll koboldcpp_openblas.dll koboldcpp_noavx2.dll koboldcpp_openblas_noavx2.dll koboldcpp_clblast.dll koboldcpp_cublas.dll koboldcpp.so koboldcpp_openblas.so koboldcpp_noavx2.so koboldcpp_openblas_noavx2.so koboldcpp_clblast.so koboldcpp_cublas.so gptj.exe gpt2.exe
 
 main: examples/main/main.cpp ggml.o llama.o common.o $(OBJS)
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)
@@ -260,6 +270,9 @@ koboldcpp_openblas_noavx2: ggml_openblas_noavx2.o ggml_rwkv.o ggml_v1_noavx2.o e
 
 koboldcpp_clblast: ggml_clblast.o ggml_rwkv.o ggml_v1.o expose.o common.o gpttype_adapter.o 
 	$(CLBLAST_BUILD)
+
+koboldcpp_cublas: ggml_cublas.o ggml_rwkv.o ggml_v1.o expose.o common.o llama_adapter.o gpttype_adapter.o 
+	$(CUBLAS_BUILD)
 		
 quantize_llama: examples/quantize/quantize.cpp ggml.o llama.o
 	$(CXX) $(CXXFLAGS) $^ -o $@ $(LDFLAGS)