Skip to content

lagoon: %int2 comparison + reduction array jets #67

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
sigilante wants to merge 1 commit into from
Closed

lagoon: %int2 comparison + reduction array jets #67

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
232 changes: 232 additions & 0 deletions lagoon/vere/noun/jets/i/lagoon.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@

#include <math.h> // for pow()
#include <stdio.h>
#include <string.h> // for memcpy (%int2 reduction output marshalling)

#define f16_ceil(a) f16_roundToInt( a, softfloat_round_max, false )
#define f32_ceil(a) f32_roundToInt( a, softfloat_round_max, false )
Expand Down Expand Up @@ -2261,6 +2262,187 @@
return r_data;
}

/* %int2 element-wise comparison (gth/gte/lth/lte): per lane a signed
** two's-complement compare (native signed-int compare IS that order), writing
** 1/0 into the result ray (same shape/bloq). Mirrors the i754 gth jet shape.
*/
typedef enum { _LA_GTH, _LA_LTH, _LA_LTE, _LA_GTE } _la_cop;

static u3_noun
_la_int2_cmp(u3_noun x_data, u3_noun y_data, u3_noun shape, u3_noun bloq, _la_cop op)
{
c3_d bl = u3x_atom(bloq);
if ( bl < 3 || bl > 7 ) return u3_none;
c3_d len = _get_length(shape);
c3_d lb = (c3_d)1 << (bl - 3);
c3_d syz = len * lb;
c3_y* xb = (c3_y*)u3a_malloc(syz * sizeof(c3_y));
c3_y* yb = (c3_y*)u3a_malloc((syz + 1) * sizeof(c3_y));
u3r_bytes(0, syz, xb, x_data);
u3r_bytes(0, syz + 1, yb, y_data);

#define _LA_CMP(UT, ST) \
{ UT* X = (UT*)xb; UT* Y = (UT*)yb; \
for ( c3_d i = 0; i < len; i++ ) { \
ST a = (ST)X[i], b = (ST)Y[i]; c3_t t; \
switch ( op ) { \
case _LA_GTH: t = (c3_t)(a > b); break; \
case _LA_LTH: t = (c3_t)(a < b); break; \
case _LA_LTE: t = (c3_t)(a <= b); break; \
default: t = (c3_t)(a >= b); break; \
} \
Y[i] = (UT)(t ? 1 : 0); \
} }
switch ( bl ) {
case 3: _LA_CMP(c3_b, int8_t); break;
case 4: _LA_CMP(c3_s, c3_ss); break;
case 5: _LA_CMP(c3_w, c3_ws); break;
case 6: _LA_CMP(c3_d, c3_ds); break;
case 7: _LA_CMP(_twoc_u128, _twoc_s128); break;
}
#undef _LA_CMP

u3_noun r_data = u3i_bytes((syz + 1) * sizeof(c3_y), yb);
u3a_free(xb); u3a_free(yb);
return r_data;
}

/* %int2 reductions to a 1-lane scalar ray (data = [value, 0x1 sentinel]),
** mirroring the i754 cumsum/min/max/dot output shape. sum/dot wrap mod
** 2^width (unsigned native add/mul); min/max use signed (two's-complement)
** order. `ob` holds one lane + the sentinel byte.
*/
static u3_noun
_la_int2_sum(u3_noun x_data, u3_noun shape, u3_noun bloq)
{
c3_d bl = u3x_atom(bloq);
if ( bl < 3 || bl > 7 ) return u3_none;
c3_d len = _get_length(shape);
c3_d lb = (c3_d)1 << (bl - 3);
c3_d syz = len * lb;
c3_y* xb = (c3_y*)u3a_malloc((syz + 1) * sizeof(c3_y));
u3r_bytes(0, syz + 1, xb, x_data);
c3_y ob[17] = {0};
#define _LA_SUM(UT) { UT* X = (UT*)xb; UT acc = 0; \
for ( c3_d i = 0; i < len; i++ ) acc = (UT)(acc + X[i]); \
memcpy(ob, &acc, sizeof(UT)); }
switch ( bl ) {
case 3: _LA_SUM(c3_b); break;
case 4: _LA_SUM(c3_s); break;
case 5: _LA_SUM(c3_w); break;
case 6: _LA_SUM(c3_d); break;
case 7: _LA_SUM(_twoc_u128); break;
}
#undef _LA_SUM
ob[lb] = 0x1;
u3_noun r_data = u3i_bytes((lb + 1) * sizeof(c3_y), ob);
u3a_free(xb);
return r_data;
}

static u3_noun
_la_int2_dot(u3_noun x_data, u3_noun y_data, u3_noun shape, u3_noun bloq)
{
c3_d bl = u3x_atom(bloq);
if ( bl < 3 || bl > 7 ) return u3_none;
c3_d len = _get_length(shape);
c3_d lb = (c3_d)1 << (bl - 3);
c3_d syz = len * lb;
c3_y* xb = (c3_y*)u3a_malloc(syz * sizeof(c3_y));
c3_y* yb = (c3_y*)u3a_malloc((syz + 1) * sizeof(c3_y));
u3r_bytes(0, syz, xb, x_data);
u3r_bytes(0, syz + 1, yb, y_data);
c3_y ob[17] = {0};
#define _LA_DOT(UT) { UT* X = (UT*)xb; UT* Y = (UT*)yb; UT acc = 0; \
for ( c3_d i = 0; i < len; i++ ) acc = (UT)(acc + (UT)(X[i] * Y[i])); \
memcpy(ob, &acc, sizeof(UT)); }
switch ( bl ) {
case 3: _LA_DOT(c3_b); break;
case 4: _LA_DOT(c3_s); break;
case 5: _LA_DOT(c3_w); break;
case 6: _LA_DOT(c3_d); break;
case 7: _LA_DOT(_twoc_u128); break;
}
#undef _LA_DOT
ob[lb] = 0x1;
u3_noun r_data = u3i_bytes((lb + 1) * sizeof(c3_y), ob);
u3a_free(xb); u3a_free(yb);
return r_data;
}

static u3_noun
_la_int2_minmax(u3_noun x_data, u3_noun shape, u3_noun bloq, c3_t is_max)
{
c3_d bl = u3x_atom(bloq);
if ( bl < 3 || bl > 7 ) return u3_none;
c3_d len = _get_length(shape);
c3_d lb = (c3_d)1 << (bl - 3);
c3_d syz = len * lb;
c3_y* xb = (c3_y*)u3a_malloc((syz + 1) * sizeof(c3_y));
u3r_bytes(0, syz + 1, xb, x_data);
c3_y ob[17] = {0};
#define _LA_MM(UT, ST) { UT* X = (UT*)xb; ST best = (ST)X[0]; \
for ( c3_d i = 0; i < len; i++ ) { ST v = (ST)X[i]; \
if ( is_max ? (v > best) : (v < best) ) best = v; } \
{ UT o = (UT)best; memcpy(ob, &o, sizeof(UT)); } }
switch ( bl ) {
case 3: _LA_MM(c3_b, int8_t); break;
case 4: _LA_MM(c3_s, c3_ss); break;
case 5: _LA_MM(c3_w, c3_ws); break;
case 6: _LA_MM(c3_d, c3_ds); break;
case 7: _LA_MM(_twoc_u128, _twoc_s128); break;
}
#undef _LA_MM
ob[lb] = 0x1;
u3_noun r_data = u3i_bytes((lb + 1) * sizeof(c3_y), ob);
u3a_free(xb);
return r_data;
}

/* %int2 argmin/argmax -> bare index atom. Mirrors the i754 jet exactly:
** strict compare, reversed lane index (len - i - 1).
*/
static u3_noun
_la_int2_argminmax(u3_noun x_data, u3_noun shape, u3_noun bloq, c3_t is_max)
{
c3_d bl = u3x_atom(bloq);
if ( bl < 3 || bl > 7 ) return u3_none;
c3_d len = _get_length(shape);
c3_d lb = (c3_d)1 << (bl - 3);
c3_d syz = len * lb;
c3_y* xb = (c3_y*)u3a_malloc(syz * sizeof(c3_y));
u3r_bytes(0, syz, xb, x_data);
c3_w idx = 0;
#define _LA_ARG(UT, ST) { UT* X = (UT*)xb; ST best = (ST)X[0]; \
for ( c3_d i = 0; i < len; i++ ) { ST v = (ST)X[i]; \
if ( is_max ? (v > best) : (v < best) ) { best = v; idx = (c3_w)i; } } }
switch ( bl ) {
case 3: _LA_ARG(c3_b, int8_t); break;
case 4: _LA_ARG(c3_s, c3_ss); break;
case 5: _LA_ARG(c3_w, c3_ws); break;
case 6: _LA_ARG(c3_d, c3_ds); break;
case 7: _LA_ARG(_twoc_u128, _twoc_s128); break;
}
#undef _LA_ARG
u3a_free(xb);
return u3i_chub(idx);
}

/* box a scalar reduction result as a ray, matching the Hoon +scalar-to-ray:
** shape = all-1s of the INPUT's rank (reap (lent shape) 1); the data already
** carries its leading-0x1 sentinel (== +spac on a 1-element ray).
*/
static u3_noun
_la_int2_box(u3_noun x_shape, u3_noun x_bloq, u3_noun x_kind,
u3_noun x_tail, u3_noun r_data)
{
c3_d rank = 0; u3_noun s = x_shape;
while ( c3y == u3du(s) ) { rank++; s = u3t(s); }
u3_noun sh = u3_nul;
for ( c3_d i = 0; i < rank; i++ ) sh = u3nc(0x1, sh);
return u3nc(u3nq(sh, u3k(x_bloq), u3k(x_kind), u3k(x_tail)), r_data);
}

u3_noun
u3wi_la_add(u3_noun cor)
{
Expand Down Expand Up @@ -2561,6 +2743,12 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3nq(u3nc(0x1, u3_nul), u3k(x_bloq), u3k(x_kind), u3k(x_tail)), r_data);

case c3__int2: {
u3_noun r_data = _la_int2_sum(x_data, x_shape, x_bloq);
if (r_data == u3_none) { return u3_none; }
return _la_int2_box(x_shape, x_bloq, x_kind, x_tail, r_data);
}

default:
return u3_none;
}
Expand Down Expand Up @@ -2599,6 +2787,10 @@
// bare atom (@ index)
return r_data;}

case c3__int2: {
u3_noun r_data = _la_int2_argminmax(x_data, x_shape, x_bloq, 0);
return r_data;}

default:
return u3_none;
}
Expand Down Expand Up @@ -2675,6 +2867,10 @@
// bare atom (@ index)
return r_data;}

case c3__int2: {
u3_noun r_data = _la_int2_argminmax(x_data, x_shape, x_bloq, 1);
return r_data;}

default:
return u3_none;
}
Expand Down Expand Up @@ -2714,6 +2910,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3nq(u3nt(0x1, 0x1, u3_nul), u3k(x_bloq), u3k(x_kind), u3k(x_tail)), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_minmax(x_data, x_shape, x_bloq, 0);
if (r_data == u3_none) { return u3_none; }
return _la_int2_box(x_shape, x_bloq, x_kind, x_tail, r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -2753,6 +2954,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3nq(u3nt(0x1, 0x1, u3_nul), u3k(x_bloq), u3k(x_kind), u3k(x_tail)), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_minmax(x_data, x_shape, x_bloq, 1);
if (r_data == u3_none) { return u3_none; }
return _la_int2_box(x_shape, x_bloq, x_kind, x_tail, r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -2833,6 +3039,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_cmp(x_data, y_data, x_shape, x_bloq, _LA_GTH);
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -2875,6 +3086,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_cmp(x_data, y_data, x_shape, x_bloq, _LA_GTE);
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -2917,6 +3133,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_cmp(x_data, y_data, x_shape, x_bloq, _LA_LTH);
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -2959,6 +3180,11 @@
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

case c3__int2: {
u3_noun r_data = _la_int2_cmp(x_data, y_data, x_shape, x_bloq, _LA_LTE);
if (r_data == u3_none) { return u3_none; }
return u3nc(u3k(x_meta), r_data);}

default:
return u3_none;
}
Expand Down Expand Up @@ -3188,6 +3414,12 @@
u3a_free(dim_x);
return u3nc(u3nq(u3nt(len_x0, 0x1, u3_nul), u3k(x_bloq), u3k(x_kind), u3k(x_tail)), r_data);

case c3__int2: {
u3_noun r_data = _la_int2_dot(x_data, y_data, x_shape, x_bloq);
if (r_data == u3_none) { return u3_none; }
return _la_int2_box(x_shape, x_bloq, x_kind, x_tail, r_data);
}

default:
return u3_none;
}
Expand Down