Port to Visual C++.

- Combined nmake/Unix Makefile.
- _alloca instead of variable size array.
- Cast void* to char* for math.
- C++20 for designated initializers.

It builds. I haven't run it yet.

Makefile

@@ -1,3 +1,47 @@
+# This one Makefile works with Microsoft nmake and Unix make.
+# They use different conditional syntax, but each can be nested and inverted within the other.
+
+all: default
+
+ifdef MAKEDIR:
+!ifdef MAKEDIR
+
+# Windows code.
+
+CXX = cl
+# C++20 for designated initializers
+# TODO: Detect AVX.
+CXXFLAGS = /MD /Gy /Z7 /EHsc /O2 /arch:AVX2 /std:c++20
+CC = $(CXX)
+CFLAGS = $(CXXFLAGS)
+LDFLAGS=/incremental:no
+LINK_OUT = /out:
+CC_OUT = /Fo
+O = obj
+EXE = .exe
+RM_F = del 2>nul /f
+UNIX_SPACE =
+START_LINK_FLAGS = /link
+
+# The need for this is surprising but otherwise there is an access violation
+# running main -h.
+SLASH=^\
+
+!else
+else
+
+# Unix code.
+
+O = o
+EXE=
+UNIX_SPACE = " "
+UNIX_SPACE := $(UNIX_SPACE:"=)
+LINK_OUT = -o$(UNIX_SPACE)
+CC_OUT=-o$(UNIX_SPACE)
+RM_F = rm -f
+START_LINK_FLAGS=
+SLASH=/
+
 ifndef UNAME_S
 UNAME_S := $(shell uname -s)
 endif
@@ -172,27 +216,30 @@ $(info I CC:       $(CCV))
 $(info I CXX:      $(CXXV))
 $(info )
 
-default: main quantize
+endif
+!endif :
+
+default: main$(EXE) quantize$(EXE)
 
 #
 # Build library
 #
 
-ggml.o: ggml.c ggml.h
+ggml.$O: ggml.c ggml.h
-	$(CC)  $(CFLAGS)   -c ggml.c -o ggml.o
+	$(CC)  $(CFLAGS) -c ggml.c $(CC_OUT)ggml.$O
 
-utils.o: utils.cpp utils.h
+utils.$O: utils.cpp utils.h
-	$(CXX) $(CXXFLAGS) -c utils.cpp -o utils.o
+	$(CXX) $(CXXFLAGS) -c utils.cpp $(CC_OUT)utils.$O
 
 clean:
-	rm -f *.o main quantize
+	$(RM_F) *.$O main$(EXE) quantize$(EXE)
 
-main: main.cpp ggml.o utils.o
+main$(EXE): main.cpp ggml.$O utils.$O
-	$(CXX) $(CXXFLAGS) main.cpp ggml.o utils.o -o main $(LDFLAGS)
+	$(CXX) $(CXXFLAGS) main.cpp $(START_LINK_FLAGS) ggml.$O utils.$O $(LINK_OUT)main$(EXE) $(LDFLAGS)
-	./main -h
+	.$(SLASH)main.exe -h
 
-quantize: quantize.cpp ggml.o utils.o
+quantize$(EXE): quantize.cpp ggml.$O utils.$O
-	$(CXX) $(CXXFLAGS) quantize.cpp ggml.o utils.o -o quantize $(LDFLAGS)
+	$(CXX) $(CXXFLAGS) quantize.cpp $(START_LINK_FLAGS) ggml.$O utils.$O $(LINK_OUT)quantize$(EXE) $(LDFLAGS)
 
 #
 # Tests

ggml.c

@@ -407,8 +407,8 @@ void quantize_row_q4_0(const float * restrict x, void * restrict y, int k) {
     const int nb = k / QK;
     const size_t bs = sizeof(float) + QK/2;
 
-    uint8_t * restrict pd = (uint8_t *) (y + 0*bs);
+    uint8_t * restrict pd = (uint8_t *) ((char*)y + 0*bs);
-    uint8_t * restrict pb = (uint8_t *) (y + 0*bs + sizeof(float));
+    uint8_t * restrict pb = (uint8_t *) ((char*)y + 0*bs + sizeof(float));
 
     uint8_t pp[QK/2];
 
@@ -654,8 +654,8 @@ void dequantize_row_q4_0(const void * restrict x, float * restrict y, int k) {
     const int nb = k / QK;
     const size_t bs = sizeof(float) + QK/2;
 
-    const uint8_t * restrict pd = (const uint8_t *) (x + 0*bs);
+    const uint8_t * restrict pd = (const uint8_t *) ((char*)x + 0*bs);
-    const uint8_t * restrict pb = (const uint8_t *) (x + 0*bs + sizeof(float));
+    const uint8_t * restrict pb = (const uint8_t *) ((char*)x + 0*bs + sizeof(float));
 
     // scalar
     for (int i = 0; i < nb; i++) {
@@ -1301,11 +1301,11 @@ inline static void ggml_vec_dot_q4_0(const int n, float * restrict s, const void
 
     const size_t bs = sizeof(float) + QK/2;
 
-    const uint8_t * restrict pd0 = (const uint8_t *) (x + 0*bs);
+    const uint8_t * restrict pd0 = (const uint8_t *) ((char*)x + 0*bs);
-    const uint8_t * restrict pd1 = (const uint8_t *) (y + 0*bs);
+    const uint8_t * restrict pd1 = (const uint8_t *) ((char*)y + 0*bs);
 
-    const uint8_t * restrict pb0 = (const uint8_t *) (x + 0*bs + sizeof(float));
+    const uint8_t * restrict pb0 = (const uint8_t *) ((char*)x + 0*bs + sizeof(float));
-    const uint8_t * restrict pb1 = (const uint8_t *) (y + 0*bs + sizeof(float));
+    const uint8_t * restrict pb1 = (const uint8_t *) ((char*)y + 0*bs + sizeof(float));
 
     float sumf = 0.0;
 
@@ -1731,8 +1731,8 @@ inline static void ggml_vec_mad_q4_0(const int n, float * restrict y, void * res
     const int nb = n / QK;
     const size_t bs = sizeof(float) + QK/2;
 
-    const uint8_t * restrict pd = (const uint8_t *) (x + 0*bs);
+    const uint8_t * restrict pd = (const uint8_t *) ((char*)x + 0*bs);
-    const uint8_t * restrict pb = (const uint8_t *) (x + 0*bs + sizeof(float));
+    const uint8_t * restrict pb = (const uint8_t *) ((char*)x + 0*bs + sizeof(float));
 
 #if __ARM_NEON
 #if QK == 32

main.cpp

@@ -733,6 +733,7 @@ bool llama_eval(
 }
 
 int main(int argc, char ** argv) {
+    ggml_time_init();
     const int64_t t_main_start_us = ggml_time_us();
 
     gpt_params params;

utils.cpp

@@ -453,7 +453,11 @@ size_t ggml_quantize_q4_0(float * src, void * dst, int n, int k, int qk, int64_t
 
     assert(k % qk == 0);
 
+#if _MSC_VER
+    uint8_t* pp = (uint8_t*)_alloca(qk / 2);
+#else
     uint8_t pp[qk/2];
+#endif
 
     char * pdst = (char *) dst;
 
@@ -507,7 +511,11 @@ size_t ggml_quantize_q4_1(float * src, void * dst, int n, int k, int qk, int64_t
 
     assert(k % qk == 0);
 
+#if _MSC_VER
+    uint8_t* pp = (uint8_t*)_alloca(qk / 2);
+#else
     uint8_t pp[qk/2];
+#endif
 
     char * pdst = (char *) dst;