h48

commit 0ebba869512abc8e37e461e7587f686fe7c91c6c
parent 8954b4b7eaafa8e306f2aa6cbb7e32334107e6e1
Author: Sebastiano Tronto <sebastiano@tronto.net>
Date:   Sat, 25 May 2024 10:25:16 +0200

Updated TODO and small changes to table gen

Diffstat:
M
TODO.txt
 | 
37
+++++++++++++++++++++++++++++--------
M
src/solve_h48.h
 | 
27
++++++++++++++++++++-------

2 files changed, 49 insertions(+), 15 deletions(-)
diff --git a/TODO.txt b/TODO.txt
@@ -1,11 +1,32 @@
-TODO
-  - add back benchmark, test performance of table computation
-  - table generation is still slow, how to improve? profile!
-    selfsim could be a list of cubes (representatives)
-  - add h48 table for different values of k
-
-TODO checkdata (available from cube.h) and hash check for cocsep
-TODO benchmarks for solve and table generation
+Table generation: transform only edges
+  (cube_avx2 and cube_portable)
+  - implement compose_fast_edges and compose_fast_corners
+  - tests not needed: just base the full transform on the other two
+  - cube_portable: easy for compose_fast to call the other two
+  - cube_avx2: split in (EP+EO+CP) and (CO); compose_edges call
+    (EP+EO+CP), compose_corners calls the full compose
+  (cube_routines)
+  - refactor transform(): use big array of cubes instead of switch
+  - add transform_edges and transform_corners
+  (solve_h48)
+  - uncomment transform_edges() and transform_corners()
+
+Solver
+  - write a solver (how many tricks? some, but not all are needed)
+
+More utilities for tables (in cube.h)
+  - a "dryrun" function that only tells you the size needed
+  - check hash of generated data
+
+Goal: find out which k value is best
+  - temporarily call current table and solver "k4" instead of h48
+  - write table generation and solver for k2 and k1
+  - benchmark for different sizes!
+
+Refactoring
+  - remove cube type and some low-level utilities from interface,
+    rename cube_fast_t to cube_t
+  - add b64 i/o format, base64 encoded cube, one 6-bit word per piece
 
 ## H48 optimal solver (some has already been implemented)
 
diff --git a/src/solve_h48.h b/src/solve_h48.h
@@ -35,6 +35,7 @@ typedef struct {
 } bfsarg_esep_t;
 
 _static_inline int64_t coord_h48(cube_fast_t, const uint32_t *, uint8_t);
+_static_inline int64_t coord_h48_edges(cube_fast_t, int64_t, uint8_t, uint8_t);
 _static_inline cube_fast_t invcoord_h48(int64_t, const cube_fast_t *, uint8_t);
 
 _static size_t gendata_cocsep(void *, uint64_t *, cube_fast_t *);
@@ -50,8 +51,7 @@ _static_inline void set_esep_pval(uint32_t *, int64_t, uint8_t);
 _static_inline int64_t
 coord_h48(cube_fast_t c, const uint32_t *cocsepdata, uint8_t h)
 {
-	cube_fast_t d;
-	int64_t cocsep, coclass, esep, eo, ret;
+	int64_t cocsep, coclass;
 	uint32_t data;
 	uint8_t ttrep;
 
@@ -62,17 +62,24 @@ coord_h48(cube_fast_t c, const uint32_t *cocsepdata, uint8_t h)
 	coclass = (data & (0xFFFFU << 16U)) >> 16U;
 	ttrep = (data & (0xFFU << 8U)) >> 8U;
 
-	d = transform(c, ttrep); /* TODO: transform only edges */
+	return coord_h48_edges(c, coclass, ttrep, h);
+}
+
+_static_inline int64_t
+coord_h48_edges(cube_fast_t c, int64_t coclass, uint8_t t, uint8_t h)
+{
+	cube_fast_t d;
+	int64_t esep, eo;
+
+	//d = transform_edges(c, t);
+	d = transform(c, t);
 	esep = coord_fast_esep(d);
 	eo = coord_fast_eo(d);
 
-	ret = (coclass * H48_ESIZE(h)) + (esep << h) + (eo >> (11-h));
-
-	return ret;
+	return (coclass * H48_ESIZE(h)) + (esep << h) + (eo >> (11-h));
 }
 
 /*
-
 This function does not necessarily return a cube whose coordinate is
 the given value, because it works up to symmetry. This means that the
 returned cube is a transformed cube of one that gives the correct value.
@@ -174,6 +181,7 @@ gendata_cocsep_dfs(dfsarg_cocsep_t *arg)
 			return 0;
 
 		for (t = 0, cc = 0; t < 48; t++) {
+			//d = transform_corners(arg->cube, t);
 			d = transform(arg->cube, t);
 			ii = coord_fast_cocsep(d);
 			arg->selfsim[*arg->n] |= (i == ii) << t;
@@ -267,6 +275,11 @@ gendata_esep_bfs(bfsarg_esep_t *arg)
 			continue;
 		cube = invcoord_h48(i, arg->crep, arg->h);
 		for (m = 0; m < 18; m++) {
+			/*
+			 * TODO: here we can optimize by computing at first
+			 * only the corner part of the coordinate, and then
+			 * the edge parts for each transformation.
+			 */
 			moved = move(cube, m);
 			j = coord_h48(moved, arg->cocsepdata, arg->h);
 			x = get_esep_pval(arg->buf32, j);