Smaller updates.

4 files changed, 42 insertions, 48 deletions

diff --git a/include/stc/cspan.h b/include/stc/cspan.h
index 3f2b300f..e72bb97a 100644
--- a/include/stc/cspan.h
+++ b/include/stc/cspan.h
@@ -63,12 +63,6 @@ int demo2() {
 #include "priv/linkage.h"
 #include "ccommon.h"
 
-#ifdef i_ndebug
-  #define cspan_assert(x) ((void)0)
-#else
-  #define cspan_assert(x) c_assert(x)
-#endif
-
 #define using_cspan(...) c_MACRO_OVERLOAD(using_cspan, __VA_ARGS__)
 #define using_cspan_2(Self, T) \
     using_cspan_3(Self, T, 1); \
@@ -91,7 +85,7 @@ int demo2() {
                                   const int rank, const int32_t a[][2]) { \
         Self s; int outrank; \
         s.data = d + _cspan_slice(s.shape, s.stride.d, &outrank, shape, stri, rank, a); \
-        cspan_assert(outrank == RANK); \
+        c_assert(outrank == RANK); \
         return s; \
     } \
     STC_INLINE Self##_iter Self##_begin(const Self* self) { \
@@ -193,7 +187,7 @@ typedef enum {c_ROWMAJOR, c_COLMAJOR} cspan_layout;
 
 STC_INLINE intptr_t _cspan_size(const int32_t shape[], int rank) {
     intptr_t sz = shape[0];
-    while (--rank > 0) sz *= shape[rank];
+    while (--rank) sz *= shape[rank];
     return sz;
 }
 
@@ -207,14 +201,15 @@ STC_INLINE void _cspan_transpose(int32_t shape[], int32_t stride[], int rank) {
 STC_INLINE intptr_t _cspan_index(int rank, const int32_t shape[], const int32_t stride[], const int32_t a[]) {
     intptr_t off = 0;
     while (rank--) {
-        cspan_assert(c_LTu(a[rank], shape[rank]));
+        c_assert(c_LTu(a[rank], shape[rank]));
         off += stride[rank]*a[rank];
     }
     return off;
 }
 
-STC_API intptr_t _cspan_next2(int32_t pos[], const int32_t shape[], const int32_t stride[], int rank, int* done);
 #define _cspan_next1(pos, shape, stride, rank, done) (*done = ++pos[0]==shape[0], stride[0])
+STC_API intptr_t 
+        _cspan_next2(int32_t pos[], const int32_t shape[], const int32_t stride[], int rank, int* done);
 #define _cspan_next3 _cspan_next2
 #define _cspan_next4 _cspan_next2
 #define _cspan_next5 _cspan_next2
@@ -270,13 +265,13 @@ STC_DEF intptr_t _cspan_slice(int32_t oshape[], int32_t ostride[], int* orank,
     for (; i < rank; ++i) {
         off += stride[i]*a[i][0];
         switch (a[i][1]) {
-            case 0: cspan_assert(c_LTu(a[i][0], shape[i])); continue;
+            case 0: c_assert(c_LTu(a[i][0], shape[i])); continue;
             case -1: end = shape[i]; break;
             default: end = a[i][1];
         }
         oshape[oi] = end - a[i][0];
         ostride[oi] = stride[i];
-        cspan_assert(c_LTu(0, oshape[oi]) & !c_LTu(shape[i], end));
+        c_assert((oshape[oi] > 0) & !c_LTu(shape[i], end));
         ++oi;
     }
     *orank = oi;
@@ -284,7 +279,6 @@ STC_DEF intptr_t _cspan_slice(int32_t oshape[], int32_t ostride[], int* orank,
 }
 
 #endif
-#undef i_ndebug
 #undef i_opt
 #undef i_header
 #undef i_implement
diff --git a/misc/benchmarks/various/cspan_bench.c b/misc/benchmarks/various/cspan_bench.c
index 3b1c3132..bfc0ead3 100644
--- a/misc/benchmarks/various/cspan_bench.c
+++ b/misc/benchmarks/various/cspan_bench.c
@@ -42,7 +42,7 @@ static void Traditional_for_loop(intptr_t n)
     printf("forloop : %.1f ms, %f\n", 1000.0f*t / CLOCKS_PER_SEC, sum);
 }
 
-static void MDRanges_nested_loop(intptr_t n)
+static void MDRanges_loop_over_joined(intptr_t n)
 {
     clock_t t = clock();
     MD3 r_in = cspan_md(Vin, nx, ny, nz);
@@ -52,22 +52,20 @@ static void MDRanges_nested_loop(intptr_t n)
     double sum = 0;
 
     for (intptr_t s = 0; s < n; ++s) {
-        for (int x = 0; x < r_in.shape[0]; ++x) {
-            for (int y = 0; y < r_in.shape[1]; ++y) {
-                for (int z = 0; z < r_in.shape[2]; ++z)
-                {
-                    double d = *cspan_at(&r_in, x,y,z);
-                    *cspan_at(&r_out, x,y,z) += d;
-                    sum += d;
-                }
-            }
+        MD3_iter i = MD3_begin(&r_in);
+        MD3_iter o = MD3_begin(&r_out);
+
+        for (; i.ref; MD3_next(&i), MD3_next(&o))
+        {
+            *o.ref += *i.ref;
+            sum += *i.ref;
         }
     }
     t = clock() - t;
-    printf("nested  : %.1f ms, %f\n", 1000.0f*t / CLOCKS_PER_SEC, sum);
+    printf("joined  : %.1f ms, %f\n", 1000.0f*t / CLOCKS_PER_SEC, sum);
 }
 
-static void MDRanges_loop_over_joined(intptr_t n)
+static void MDRanges_nested_loop(intptr_t n)
 {
     clock_t t = clock();
     MD3 r_in = cspan_md(Vin, nx, ny, nz);
@@ -77,19 +75,22 @@ static void MDRanges_loop_over_joined(intptr_t n)
     double sum = 0;
 
     for (intptr_t s = 0; s < n; ++s) {
-        MD3_iter i = MD3_begin(&r_in);
-        MD3_iter o = MD3_begin(&r_out);
-
-        for (; i.ref; MD3_next(&i), MD3_next(&o))
-        {
-            *o.ref += *i.ref;
-            sum += *i.ref;
+        for (int x = 0; x < r_in.shape[0]; ++x) {
+            for (int y = 0; y < r_in.shape[1]; ++y) {
+                for (int z = 0; z < r_in.shape[2]; ++z)
+                {
+                    double d = *cspan_at(&r_in, x,y,z);
+                    *cspan_at(&r_out, x,y,z) += d;
+                    sum += d;
+                }
+            }
         }
     }
     t = clock() - t;
-    printf("joined  : %.1f ms, %f\n", 1000.0f*t / CLOCKS_PER_SEC, sum);
+    printf("nested  : %.1f ms, %f\n", 1000.0f*t / CLOCKS_PER_SEC, sum);
 }
 
+
 int main(void)
 {
     intptr_t n = 100000;
@@ -97,6 +98,6 @@ int main(void)
         Vin[i] = i + 1.23;
 
     Traditional_for_loop(n);
-    MDRanges_nested_loop(n);
     MDRanges_loop_over_joined(n);
+    MDRanges_nested_loop(n);
 }
diff --git a/misc/examples/spans/matmult.c b/misc/examples/spans/matmult.c
index 266fa121..ec992ff9 100644
--- a/misc/examples/spans/matmult.c
+++ b/misc/examples/spans/matmult.c
@@ -37,7 +37,7 @@ void base_case_matrix_product(Mat2 A, Mat2 B, OutMat C)
 void recursive_matrix_product(Mat2 A, Mat2 B, OutMat C)
 {
   // Some hardware-dependent constant
-  enum {recursion_threshold = 16};
+  enum {recursion_threshold = 32};
   if (C.shape[0] <= recursion_threshold || C.shape[1] <= recursion_threshold) {
     base_case_matrix_product(A, B, C);
   } else {
@@ -63,28 +63,28 @@ void recursive_matrix_product(Mat2 A, Mat2 B, OutMat C)
 
 int main(void)
 {
-  enum {N = 10, D1 = 256, D2 = D1};
+  enum {N = 10, D = 256};
 
   Values values = {0};
-  for (int i=0; i < N*D1*D2; ++i)
+  for (int i=0; i < N*D*D; ++i)
       Values_push(&values, (crandf() - 0.5)*4.0);
 
-  double out[D1*D2];
-  Mat3 data = cspan_md_layout(c_ROWMAJOR, values.data, N, D1, D2);
-  OutMat c = cspan_md_layout(c_ROWMAJOR, out, D1, D2);
+  double out[D*D];
+  Mat3 data = cspan_md_layout(c_ROWMAJOR, values.data, N, D, D);
+  OutMat c = cspan_md_layout(c_COLMAJOR, out, D, D);
   Mat2 a = cspan_submd3(&data, 0);
-  double sum = 0.0;
-  clock_t t = clock();
 
+  clock_t t = clock();
   for (int i=1; i<N; ++i) {
     Mat2 b = cspan_submd3(&data, i);
     memset(out, 0, sizeof out);
     recursive_matrix_product(a, b, c);
     //base_case_matrix_product(a, b, c);
-    sum += *cspan_at(&c, 0, 1);
   }
-
   t = clock() - t;
-  printf("%.16g: %f\n", sum, (double)t*1000.0/CLOCKS_PER_SEC);
+
+  double sum = 0.0;
+  c_foreach (i, Mat2, c) sum += *i.ref;
+  printf("sum=%.16g, %f ms\n", sum, (double)t*1000.0/CLOCKS_PER_SEC);
   Values_drop(&values);
 }
diff --git a/src/libstc.c b/src/libstc.c
index 462c97c4..1305abef 100644
--- a/src/libstc.c
+++ b/src/libstc.c
@@ -1,10 +1,9 @@
+#define STC_IMPLEMENT

 #define i_import

 #include "../include/stc/cregex.h" /* cstr. utf8, and cregex */

-#define i_implement

 #include "../include/stc/csview.h"

-#define i_implement

 #include "../include/stc/crand.h"

+#include "../include/stc/cspan.h"

 #if __STDC_VERSION__ >= 201112L

-#  define i_implement

 #  include "../include/c11/fmt.h"

 #endif