ogl_beamforming

util.c (17512B)

---

 /* See LICENSE for license details. */
 #define zero_struct(s) mem_clear(s, 0, sizeof(*s))
 function void *
 mem_clear(void *restrict p_, u8 c, iz size)
 {
 	u8 *p = p_;
 	while (size > 0) p[--size] = c;
 	return p;
 }
 
 function void
 mem_copy(void *restrict dest, void *restrict src, uz n)
 {
 	u8 *s = src, *d = dest;
 	for (; n; n--) *d++ = *s++;
 }
 
 function void
 mem_move(u8 *dest, u8 *src, iz n)
 {
 	if (dest < src) mem_copy(dest, src, n);
 	else            while (n) { n--; dest[n] = src[n]; }
 }
 
 
 function u8 *
 arena_commit(Arena *a, iz size)
 {
 	ASSERT(a->end - a->beg >= size);
 	u8 *result = a->beg;
 	a->beg += size;
 	return result;
 }
 
 function void
 arena_pop(Arena *a, iz length)
 {
 	a->beg -= length;
 }
 
 #define push_array(a, t, n) (t *)arena_alloc(a, sizeof(t), _Alignof(t), n)
 #define push_struct(a, t)   (t *)arena_alloc(a, sizeof(t), _Alignof(t), 1)
 function void *
 arena_alloc(Arena *a, iz len, iz align, iz count)
 {
 	/* NOTE: special case 0 arena */
 	if (a->beg == 0)
 		return 0;
 
 	iz padding   = -(uintptr_t)a->beg & (align - 1);
 	iz available = a->end - a->beg - padding;
 	if (available < 0 || count > available / len)
 		ASSERT(0 && "arena OOM\n");
 	void *p = a->beg + padding;
 	a->beg += padding + count * len;
 	/* TODO: Performance? */
 	return mem_clear(p, 0, count * len);
 }
 
 #define arena_capacity(a, t) arena_capacity_(a, sizeof(t), _Alignof(t))
 function iz
 arena_capacity_(Arena *a, iz size, iz alignment)
 {
 	iz padding   = -(uintptr_t)a->beg & (alignment - 1);
 	iz available = a->end - a->beg - padding;
 	iz result    = available / size;
 	return result;
 }
 
 enum { DA_INITIAL_CAP = 4 };
 #define da_reserve(a, s, n) \
   (s)->data = da_reserve_((a), (s)->data, &(s)->capacity, (s)->count + n, \
                           _Alignof(typeof(*(s)->data)), sizeof(*(s)->data))
 
 #define da_push(a, s) \
   ((s)->count == (s)->capacity  \
     ? da_reserve(a, s, 1),      \
       (s)->data + (s)->count++  \
     : (s)->data + (s)->count++)
 
 function void *
 da_reserve_(Arena *a, void *data, iz *capacity, iz needed, iz align, iz size)
 {
 	iz cap = *capacity;
 
 	/* NOTE(rnp): handle both 0 initialized DAs and DAs that need to be moved (they started
 	 * on the stack or someone allocated something in the middle of the arena during usage) */
 	if (!data || a->beg != (u8 *)data + cap * size) {
 		void *copy = arena_alloc(a, size, align, cap);
 		if (data) mem_copy(copy, data, cap * size);
 		data = copy;
 	}
 
 	if (!cap) cap = DA_INITIAL_CAP;
 	while (cap < needed) cap *= 2;
 	arena_alloc(a, size, align, cap - *capacity);
 	*capacity = cap;
 	return data;
 }
 
 function Arena
 sub_arena(Arena *a, iz len, iz align)
 {
 	Arena result = {0};
 
 	iz padding = -(uintptr_t)a->beg & (align - 1);
 	result.beg   = a->beg + padding;
 	result.end   = result.beg + len;
 	arena_commit(a, len + padding);
 
 	return result;
 }
 
 function TempArena
 begin_temp_arena(Arena *a)
 {
 	TempArena result = {.arena = a, .old_beg = a->beg};
 	return result;
 }
 
 function void
 end_temp_arena(TempArena ta)
 {
 	Arena *a = ta.arena;
 	if (a) {
 		assert(a->beg >= ta.old_beg);
 		a->beg = ta.old_beg;
 	}
 }
 
 function u32
 utf8_encode(u8 *out, u32 cp)
 {
 	u32 result = 1;
 	if (cp <= 0x7F) {
 		out[0] = cp & 0x7F;
 	} else if (cp <= 0x7FF) {
 		result = 2;
 		out[0] = ((cp >>  6) & 0x1F) | 0xC0;
 		out[1] = ((cp >>  0) & 0x3F) | 0x80;
 	} else if (cp <= 0xFFFF) {
 		result = 3;
 		out[0] = ((cp >> 12) & 0x0F) | 0xE0;
 		out[1] = ((cp >>  6) & 0x3F) | 0x80;
 		out[2] = ((cp >>  0) & 0x3F) | 0x80;
 	} else if (cp <= 0x10FFFF) {
 		result = 4;
 		out[0] = ((cp >> 18) & 0x07) | 0xF0;
 		out[1] = ((cp >> 12) & 0x3F) | 0x80;
 		out[2] = ((cp >>  6) & 0x3F) | 0x80;
 		out[3] = ((cp >>  0) & 0x3F) | 0x80;
 	} else {
 		out[0] = '?';
 	}
 	return result;
 }
 
 function UnicodeDecode
 utf16_decode(u16 *data, iz length)
 {
 	UnicodeDecode result = {.cp = U32_MAX};
 	if (length) {
 		result.consumed = 1;
 		result.cp = data[0];
 		if (length > 1 && BETWEEN(data[0], 0xD800, 0xDBFF)
 		               && BETWEEN(data[1], 0xDC00, 0xDFFF))
 		{
 			result.consumed = 2;
 			result.cp = ((data[0] - 0xD800) << 10) | ((data[1] - 0xDC00) + 0x10000);
 		}
 	}
 	return result;
 }
 
 function u32
 utf16_encode(u16 *out, u32 cp)
 {
 	u32 result = 1;
 	if (cp == U32_MAX) {
 		out[0] = '?';
 	} else if (cp < 0x10000) {
 		out[0] = cp;
 	} else {
 		u32 value = cp - 0x10000;
 		out[0] = 0xD800 + (value >> 10u);
 		out[1] = 0xDC00 + (value & 0x3FFu);
 		result = 2;
 	}
 	return result;
 }
 
 function Stream
 stream_alloc(Arena *a, iz cap)
 {
 	Stream result = {.cap = cap};
 	result.data = push_array(a, u8, cap);
 	return result;
 }
 
 function s8
 stream_to_s8(Stream *s)
 {
 	s8 result = s8("");
 	if (!s->errors) result = (s8){.len = s->widx, .data = s->data};
 	return result;
 }
 
 function void
 stream_reset(Stream *s, iz index)
 {
 	s->errors = s->cap <= index;
 	if (!s->errors)
 		s->widx = index;
 }
 
 function void
 stream_commit(Stream *s, iz count)
 {
 	s->errors |= !BETWEEN(s->widx + count, 0, s->cap);
 	if (!s->errors)
 		s->widx += count;
 }
 
 function void
 stream_append(Stream *s, void *data, iz count)
 {
 	s->errors |= (s->cap - s->widx) < count;
 	if (!s->errors) {
 		mem_copy(s->data + s->widx, data, count);
 		s->widx += count;
 	}
 }
 
 function void
 stream_append_byte(Stream *s, u8 b)
 {
 	stream_append(s, &b, 1);
 }
 
 function void
 stream_pad(Stream *s, u8 b, i32 n)
 {
 	while (n > 0) stream_append_byte(s, b), n--;
 }
 
 function void
 stream_append_s8(Stream *s, s8 str)
 {
 	stream_append(s, str.data, str.len);
 }
 
 #define stream_append_s8s(s, ...) stream_append_s8s_(s, arg_list(s8, ##__VA_ARGS__))
 function void
 stream_append_s8s_(Stream *s, s8 *strs, iz count)
 {
 	for (iz i = 0; i < count; i++)
 		stream_append(s, strs[i].data, strs[i].len);
 }
 
 function void
 stream_append_u64(Stream *s, u64 n)
 {
 	u8 tmp[64];
 	u8 *end = tmp + sizeof(tmp);
 	u8 *beg = end;
 	do { *--beg = '0' + (n % 10); } while (n /= 10);
 	stream_append(s, beg, end - beg);
 }
 
 function void
 stream_append_hex_u64(Stream *s, u64 n)
 {
 	if (!s->errors) {
 		u8  buf[16];
 		u8 *end = buf + sizeof(buf);
 		u8 *beg = end;
 		while (n) {
 			*--beg = "0123456789abcdef"[n & 0x0F];
 			n >>= 4;
 		}
 		while (end - beg < 2)
 			*--beg = '0';
 		stream_append(s, beg, end - beg);
 	}
 }
 
 function void
 stream_append_i64(Stream *s, i64 n)
 {
 	if (n < 0) {
 		stream_append_byte(s, '-');
 		n *= -1;
 	}
 	stream_append_u64(s, n);
 }
 
 function void
 stream_append_f64(Stream *s, f64 f, i64 prec)
 {
 	if (f < 0) {
 		stream_append_byte(s, '-');
 		f *= -1;
 	}
 
 	/* NOTE: round last digit */
 	f += 0.5f / prec;
 
 	if (f >= (f64)(-1UL >> 1)) {
 		stream_append_s8(s, s8("inf"));
 	} else {
 		u64 integral = f;
 		u64 fraction = (f - integral) * prec;
 		stream_append_u64(s, integral);
 		stream_append_byte(s, '.');
 		for (i64 i = prec / 10; i > 1; i /= 10) {
 			if (i > fraction)
 				stream_append_byte(s, '0');
 		}
 		stream_append_u64(s, fraction);
 	}
 }
 
 function void
 stream_append_f64_e(Stream *s, f64 f)
 {
 	/* TODO: there should be a better way of doing this */
 	#if 0
 	/* NOTE: we ignore subnormal numbers for now */
 	union { f64 f; u64 u; } u = {.f = f};
 	i32 exponent = ((u.u >> 52) & 0x7ff) - 1023;
 	f32 log_10_of_2 = 0.301f;
 	i32 scale       = (exponent * log_10_of_2);
 	/* NOTE: normalize f */
 	for (i32 i = ABS(scale); i > 0; i--)
 		f *= (scale > 0)? 0.1f : 10.0f;
 	#else
 	i32 scale = 0;
 	if (f != 0) {
 		while (f > 1) {
 			f *= 0.1f;
 			scale++;
 		}
 		while (f < 1) {
 			f *= 10.0f;
 			scale--;
 		}
 	}
 	#endif
 
 	i32 prec = 100;
 	stream_append_f64(s, f, prec);
 	stream_append_byte(s, 'e');
 	stream_append_byte(s, scale >= 0? '+' : '-');
 	for (i32 i = prec / 10; i > 1; i /= 10)
 		stream_append_byte(s, '0');
 	stream_append_u64(s, ABS(scale));
 }
 
 function void
 stream_append_v2(Stream *s, v2 v)
 {
 	stream_append_byte(s, '{');
 	stream_append_f64(s, v.x, 100);
 	stream_append_s8(s, s8(", "));
 	stream_append_f64(s, v.y, 100);
 	stream_append_byte(s, '}');
 }
 
 function Stream
 arena_stream(Arena a)
 {
 	Stream result = {0};
 	result.data   = a.beg;
 	result.cap    = a.end - a.beg;
 	return result;
 }
 
 function s8
 arena_stream_commit(Arena *a, Stream *s)
 {
 	ASSERT(s->data == a->beg);
 	s8 result = stream_to_s8(s);
 	arena_commit(a, result.len);
 	return result;
 }
 
 function s8
 arena_stream_commit_zero(Arena *a, Stream *s)
 {
 	b32 error = s->errors || s->widx == s->cap;
 	if (!error)
 		s->data[s->widx] = 0;
 	s8 result = stream_to_s8(s);
 	arena_commit(a, result.len + 1);
 	return result;
 }
 
 /* NOTE(rnp): FNV-1a hash */
 function u64
 s8_hash(s8 v)
 {
 	u64 h = 0x3243f6a8885a308d; /* digits of pi */
 	for (; v.len; v.len--) {
 		h ^= v.data[v.len - 1] & 0xFF;
 		h *= 1111111111111111111; /* random prime */
 	}
 	return h;
 }
 
 function s8
 c_str_to_s8(char *cstr)
 {
 	s8 result = {.data = (u8 *)cstr};
 	if (cstr) { while (*cstr) { result.len++; cstr++; } }
 	return result;
 }
 
 /* NOTE(rnp): returns < 0 if byte is not found */
 function iz
 s8_scan_backwards(s8 s, u8 byte)
 {
 	iz result = s.len;
 	while (result && s.data[result - 1] != byte) result--;
 	result--;
 	return result;
 }
 
 function s8
 s8_cut_head(s8 s, iz cut)
 {
 	s8 result = s;
 	if (cut > 0) {
 		result.data += cut;
 		result.len  -= cut;
 	}
 	return result;
 }
 
 function s8
 s8_alloc(Arena *a, iz len)
 {
 	s8 result = {.data = push_array(a, u8, len), .len = len};
 	return result;
 }
 
 function s8
 s16_to_s8(Arena *a, s16 in)
 {
 	s8 result = s8("");
 	if (in.len) {
 		iz commit = in.len * 4;
 		iz length = 0;
 		u8 *data = arena_commit(a, commit + 1);
 		u16 *beg = in.data;
 		u16 *end = in.data + in.len;
 		while (beg < end) {
 			UnicodeDecode decode = utf16_decode(beg, end - beg);
 			length += utf8_encode(data + length, decode.cp);
 			beg    += decode.consumed;
 		}
 		data[length] = 0;
 		result = (s8){.len = length, .data = data};
 		arena_pop(a, commit - length);
 	}
 	return result;
 }
 
 function s16
 s8_to_s16(Arena *a, s8 in)
 {
 	s16 result = {0};
 	if (in.len) {
 		iz required = 2 * in.len + 1;
 		u16 *data   = push_array(a, u16, required);
 		iz length   = 0;
 		/* TODO(rnp): utf8_decode */
 		for (iz i = 0; i < in.len; i++) {
 			u32 cp  = in.data[i];
 			length += utf16_encode(data + length, cp);
 		}
 		result = (s16){.len = length, .data = data};
 		arena_pop(a, required - length);
 	}
 	return result;
 }
 
 function s8
 push_s8(Arena *a, s8 str)
 {
 	s8 result = s8_alloc(a, str.len);
 	mem_copy(result.data, str.data, result.len);
 	return result;
 }
 
 function s8
 push_s8_zero(Arena *a, s8 str)
 {
 	s8 result   = s8_alloc(a, str.len + 1);
 	result.len -= 1;
 	mem_copy(result.data, str.data, result.len);
 	return result;
 }
 
 function u32
 round_down_power_of_2(u32 a)
 {
 	u32 result = 0x80000000UL >> clz_u32(a);
 	return result;
 }
 
 function b32
 uv2_equal(uv2 a, uv2 b)
 {
 	return a.x == b.x && a.y == b.y;
 }
 
 function b32
 uv3_equal(uv3 a, uv3 b)
 {
 	return a.x == b.x && a.y == b.y && a.z == b.z;
 }
 
 function v3
 cross(v3 a, v3 b)
 {
 	v3 result = {
 		.x = a.y * b.z - a.z * b.y,
 		.y = a.z * b.x - a.x * b.z,
 		.z = a.x * b.y - a.y * b.x,
 	};
 	return result;
 }
 
 function v3
 sub_v3(v3 a, v3 b)
 {
 	v3 result = {
 		.x = a.x - b.x,
 		.y = a.y - b.y,
 		.z = a.z - b.z,
 	};
 	return result;
 }
 
 function f32
 length_v3(v3 a)
 {
 	f32 result = a.x * a.x + a.y * a.y + a.z * a.z;
 	return result;
 }
 
 function v3
 normalize_v3(v3 a)
 {
 	f32 length = length_v3(a);
 	v3 result = {.x = a.x / length, .y = a.y / length, .z = a.z / length};
 	return result;
 }
 
 function v2
 clamp_v2_rect(v2 v, Rect r)
 {
 	v2 result = v;
 	result.x = CLAMP(v.x, r.pos.x, r.pos.x + r.size.x);
 	result.y = CLAMP(v.y, r.pos.y, r.pos.y + r.size.y);
 	return result;
 }
 
 function v2
 add_v2(v2 a, v2 b)
 {
 	v2 result = {
 		.x = a.x + b.x,
 		.y = a.y + b.y,
 	};
 	return result;
 }
 
 function v2
 sub_v2(v2 a, v2 b)
 {
 	v2 result = {
 		.x = a.x - b.x,
 		.y = a.y - b.y,
 	};
 	return result;
 }
 
 function v2
 scale_v2(v2 a, f32 scale)
 {
 	v2 result = {
 		.x = a.x * scale,
 		.y = a.y * scale,
 	};
 	return result;
 }
 
 function v2
 mul_v2(v2 a, v2 b)
 {
 	v2 result = {
 		.x = a.x * b.x,
 		.y = a.y * b.y,
 	};
 	return result;
 }
 
 function v2
 div_v2(v2 a, v2 b)
 {
 	v2 result;
 	result.x = a.x / b.x;
 	result.y = a.y / b.y;
 	return result;
 }
 
 
 function v2
 floor_v2(v2 a)
 {
 	v2 result;
 	result.x = (u32)a.x;
 	result.y = (u32)a.y;
 	return result;
 }
 
 function f32
 magnitude_v2(v2 a)
 {
 	f32 result = sqrt_f32(a.x * a.x + a.y * a.y);
 	return result;
 }
 
 function uv4
 uv4_from_u32_array(u32 v[4])
 {
 	uv4 result;
 	result.E[0] = v[0];
 	result.E[1] = v[1];
 	result.E[2] = v[2];
 	result.E[3] = v[3];
 	return result;
 }
 
 function b32
 uv4_equal(uv4 a, uv4 b)
 {
 	return a.x == b.x && a.y == b.y && a.z == b.z && a.w == b.w;
 }
 
 function v4
 v4_from_f32_array(f32 v[4])
 {
 	v4 result;
 	result.E[0] = v[0];
 	result.E[1] = v[1];
 	result.E[2] = v[2];
 	result.E[3] = v[3];
 	return result;
 }
 
 function v4
 sub_v4(v4 a, v4 b)
 {
 	v4 result;
 	result.x = a.x - b.x;
 	result.y = a.y - b.y;
 	result.z = a.z - b.z;
 	result.w = a.w - b.w;
 	return result;
 }
 
 function void
 split_rect_horizontal(Rect rect, f32 fraction, Rect *left, Rect *right)
 {
 	if (left) {
 		left->pos    = rect.pos;
 		left->size.h = rect.size.h;
 		left->size.w = rect.size.w * fraction;
 	}
 	if (right) {
 		right->pos    = rect.pos;
 		right->pos.x += rect.size.w * fraction;
 		right->size.h = rect.size.h;
 		right->size.w = rect.size.w * (1.0f - fraction);
 	}
 }
 
 function void
 split_rect_vertical(Rect rect, f32 fraction, Rect *top, Rect *bot)
 {
 	if (top) {
 		top->pos    = rect.pos;
 		top->size.w = rect.size.w;
 		top->size.h = rect.size.h * fraction;
 	}
 	if (bot) {
 		bot->pos    = rect.pos;
 		bot->pos.y += rect.size.h * fraction;
 		bot->size.w = rect.size.w;
 		bot->size.h = rect.size.h * (1.0f - fraction);
 	}
 }
 
 function void
 cut_rect_horizontal(Rect rect, f32 at, Rect *left, Rect *right)
 {
 	at = MIN(at, rect.size.w);
 	if (left) {
 		*left = rect;
 		left->size.w = at;
 	}
 	if (right) {
 		*right = rect;
 		right->pos.x  += at;
 		right->size.w -= at;
 	}
 }
 
 function void
 cut_rect_vertical(Rect rect, f32 at, Rect *top, Rect *bot)
 {
 	at = MIN(at, rect.size.h);
 	if (top) {
 		*top = rect;
 		top->size.h = at;
 	}
 	if (bot) {
 		*bot = rect;
 		bot->pos.y  += at;
 		bot->size.h -= at;
 	}
 }
 
 function f64
 parse_f64(s8 s)
 {
 	f64 integral = 0, fractional = 0, sign = 1;
 
 	if (s.len && *s.data == '-') {
 		sign = -1;
 		s.data++;
 		s.len--;
 	}
 
 	while (s.len && *s.data != '.') {
 		integral *= 10;
 		integral += *s.data - '0';
 		s.data++;
 		s.len--;
 	}
 
 	if (*s.data == '.') { s.data++; s.len--; }
 
 	while (s.len) {
 		ASSERT(s.data[s.len - 1] != '.');
 		fractional /= 10;
 		fractional += (f64)(s.data[--s.len] - '0') / 10.0;
 	}
 	f64 result = sign * (integral + fractional);
 	return result;
 }
 
 function FileWatchDirectory *
 lookup_file_watch_directory(FileWatchContext *ctx, u64 hash)
 {
 	FileWatchDirectory *result = 0;
 	for (u32 i = 0; i < ctx->count; i++) {
 		FileWatchDirectory *test = ctx->data + i;
 		if (test->hash == hash) {
 			result = test;
 			break;
 		}
 	}
 	return result;
 }
 
 function void
 fill_kronecker_sub_matrix(i32 *out, i32 out_stride, i32 scale, i32 *b, uv2 b_dim)
 {
 	f32x4 vscale = dup_f32x4(scale);
 	for (u32 i = 0; i < b_dim.y; i++) {
 		for (u32 j = 0; j < b_dim.x; j += 4, b += 4) {
 			f32x4 vb = cvt_i32x4_f32x4(load_i32x4(b));
 			store_i32x4(cvt_f32x4_i32x4(mul_f32x4(vscale, vb)), out + j);
 		}
 		out += out_stride;
 	}
 }
 
 /* NOTE: this won't check for valid space/etc and assumes row major order */
 function void
 kronecker_product(i32 *out, i32 *a, uv2 a_dim, i32 *b, uv2 b_dim)
 {
 	uv2 out_dim = {.x = a_dim.x * b_dim.x, .y = a_dim.y * b_dim.y};
 	ASSERT(out_dim.y % 4 == 0);
 	for (u32 i = 0; i < a_dim.y; i++) {
 		i32 *vout = out;
 		for (u32 j = 0; j < a_dim.x; j++, a++) {
 			fill_kronecker_sub_matrix(vout, out_dim.y, *a, b, b_dim);
 			vout += b_dim.y;
 		}
 		out += out_dim.y * b_dim.x;
 	}
 }
 
 /* NOTE/TODO: to support even more hadamard sizes use the Paley construction */
 function i32 *
 make_hadamard_transpose(Arena *a, u32 dim)
 {
 	read_only local_persist	i32 hadamard_12_12_transpose[] = {
 		1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,  1,
 		1, -1, -1,  1, -1, -1, -1,  1,  1,  1, -1,  1,
 		1,  1, -1, -1,  1, -1, -1, -1,  1,  1,  1, -1,
 		1, -1,  1, -1, -1,  1, -1, -1, -1,  1,  1,  1,
 		1,  1, -1,  1, -1, -1,  1, -1, -1, -1,  1,  1,
 		1,  1,  1, -1,  1, -1, -1,  1, -1, -1, -1,  1,
 		1,  1,  1,  1, -1,  1, -1, -1,  1, -1, -1, -1,
 		1, -1,  1,  1,  1, -1,  1, -1, -1,  1, -1, -1,
 		1, -1, -1,  1,  1,  1, -1,  1, -1, -1,  1, -1,
 		1, -1, -1, -1,  1,  1,  1, -1,  1, -1, -1,  1,
 		1,  1, -1, -1, -1,  1,  1,  1, -1,  1, -1, -1,
 		1, -1,  1, -1, -1, -1,  1,  1,  1, -1,  1, -1,
 	};
 
 	i32 *result = 0;
 
 	b32 power_of_2     = ISPOWEROF2(dim);
 	b32 multiple_of_12 = dim % 12 == 0;
 	b32 multiple_of_20 = dim % 20 == 0;
 	iz elements        = dim * dim;
 
 	u32 base_dim = 0;
 	if (power_of_2) {
 		base_dim  = dim;
 	} else if (multiple_of_20) {
 		/* TODO(rnp): hadamard 20 */
 		//base_dim  = 20;
 		//dim      /= 20;
 	} else if (multiple_of_12 && dim != 36) {
 		base_dim  = 12;
 		dim      /= 12;
 	}
 
 	if (ISPOWEROF2(dim) && base_dim && arena_capacity(a, i32) >= elements * (1 + (dim != base_dim))) {
 		result = push_array(a, i32, elements);
 
 		Arena tmp = *a;
 		i32 *m = dim == base_dim ? result : push_array(&tmp, i32, elements);
 
 		#define IND(i, j) ((i) * dim + (j))
 		m[0] = 1;
 		for (u32 k = 1; k < dim; k *= 2) {
 			for (u32 i = 0; i < k; i++) {
 				for (u32 j = 0; j < k; j++) {
 					i32 val = m[IND(i, j)];
 					m[IND(i + k, j)]     =  val;
 					m[IND(i, j + k)]     =  val;
 					m[IND(i + k, j + k)] = -val;
 				}
 			}
 		}
 		#undef IND
 
 		if (dim != base_dim) {
 			ASSERT(base_dim == 12);
 			i32 *m2 = hadamard_12_12_transpose;
 			kronecker_product(result,
 			                  m,  (uv2){.x = dim,      .y = dim},
 			                  m2, (uv2){.x = base_dim, .y = base_dim});
 		}
 	}
 
 	return result;
 }