Adding strided computation to blas kernels.

I started implementing LQ factorization and immediately realized I needed strided views.
For simplicity, I will just implement them in the most portable, C native way (no vectorization).
Speed can come later.

1 files changed, 10 insertions, 14 deletions

diff --git a/include/libmath.h b/include/libmath.h
index b0ae434..1da414a 100644
--- a/include/libmath.h
+++ b/include/libmath.h
@@ -148,11 +148,13 @@ typedef uint32 blas·Flag;
 void   blas·rot(int len, double *x, double *y, double cos, double sin);
 void   blas·rotg(double *a, double *b, double *cos, double *sin);
 error  blas·rotm(int len, double *x, double *y, double p[5]);
-void   blas·scale(int len, double *x, double a);
+void   blas·scale(int len, double a, double *x);
 void   blas·copy(int len, double *x, double *y);
 void   blas·swap(int len, double *x, double *y);
 void   blas·axpy(int len, double a, double *x, double *y);
-double blas·dot(int len, double *x, double *y);
+double blas·dot(int len, double *x, int incx, double *y, int incy);
+double blas·norm(int len, double *x);
+double blas·sum(int len, double *x);
 int    blas·argmax(int len, double *x);
 int    blas·argmin(int len, double *x);
 
@@ -172,25 +174,19 @@ void   blas·trsm(blas·Flag f, int nrow, int ncol, double a, double *m1, double
 // -----------------------------------------------------------------------
 // higher level linear algebra
 
-struct linalg·Header 
-{
-    void          *h;
-    mem·Allocator heap;
-
-    double *data;
-};
-
 typedef struct math·Vector
 {
-    int       len;
-    struct    linalg·Header;
+    double *data;
+    int     len;
 } math·Vector;
 
+#define math·slicev(vec, lo, hi) (struct math·Vector){.len=((hi)-(lo)), .data=((vec).data + (lo))}
+
 typedef struct math·Matrix
 {
-    int       dim[2];
+    double   *data;
     blas·Flag kind;
-    struct    linalg·Header;
+    int       dim[2];
 } math·Matrix;
 
 // TODO: tensor ala numpy