rustc -C opt-level=3): 2049 (203973 B .text / 53737 insns)22/9 Hz): 248869 B / 5979051 B (ratio 248869/5979051, saved 5730182/5979051, 11695 ″ dittos)| Rank & Crown | Function Name & Domain | Best Use & Snap-Circuit Role | Pin Signature & 64-Vec Hash | Hardware _rdtsc (p/q) | Details & Quick Peek |
|---|---|---|---|---|---|
| #469 CROWN #1 of 5 | st_golf_gaia_ternary_dot_swar_i64 domain: gaia-mlx-quant | id: 26c297061059Aye: aye::Hue+Gemini(0+1)::x86_64 @ 432Hz | Balanced Ternary & b158 Quantization High-speed ternary dot-product accumulation without integer multiplier bottlenecks; ideal for b158 neural weights and low-bit attention inner loops. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0abcabf86830C64 RAM: 389/32768 (778 B src) | 587677/50000 cyc/op 318867/100000 ns/op | 100000000/318867 ops/us 224 B .text (59 insns) | Trit + (1) | Inspect Details & ASM →Quick Peek Rust (778 B)pub const fn st_golf_gaia_ternary_dot_swar_i64(activations: u64, weights: u64) -> i64 {
const M: u64 = 0x1111_1111_1111_1111;
const NIB: u64 = 0x0F0F_0F0F_0F0F_0F0F;
const ONES: u64 = 0x0101_0101_0101_0101;
let pos_nib = ((weights >> 3) & ((weights >> 2) | ((weights >> 1) & weights))) & M;
let nw = !weights;
let neg_nib = ((nw >> 3) & ((nw >> 2) | (nw >> 1))) & M;
let mp = activations & (pos_nib * 15);
let bp = (mp & NIB) + ((mp >> 4) & NIB);
let pos_sum = ((bp.wrapping_mul(ONES) >> 56) as i64) - 8 * (pos_nib.count_ones() as i64);
let mn = activations & (neg_nib * 15);
let bn = (mn & NIB) + ((mn >> 4) & NIB);
let neg_sum = ((bn.wrapping_mul(ONES) >> 56) as i64) - 8 * (neg_nib.count_ones() as i64);
pos_sum - neg_sum
} |
| #586 PEER of 5 | st_golf_gaia_ternary_dot_swar_i64 domain: gaia-mlx-quant | id: aecca882ad00Aye: aye::Hue+Gemini(0+1)::x86_64 @ 432Hz | Balanced Ternary & b158 Quantization High-speed ternary dot-product accumulation without integer multiplier bottlenecks; ideal for b158 neural weights and low-bit attention inner loops. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0abcabf86830C64 RAM: 215/16384 (860 B src) | 1392847/100000 cyc/op 1511449/400000 ns/op | 400000000/1511449 ops/us 287 B .text (72 insns) | Trit + (1) | Inspect Details & ASM →Quick Peek Rust (860 B)pub const fn st_golf_gaia_ternary_dot_swar_i64(activations: u64, weights: u64) -> i64 {
const M: u64 = 0x1111_1111_1111_1111;
const NIB: u64 = 0x0F0F_0F0F_0F0F_0F0F;
const B16: u64 = 0x00FF_00FF_00FF_00FF;
let pos_nib = ((weights >> 3) & ((weights >> 2) | ((weights >> 1) & weights))) & M;
let nw = !weights;
let neg_nib = ((nw >> 3) & ((nw >> 2) | (nw >> 1))) & M;
let mp = activations & (pos_nib * 15);
let bp = (mp & NIB) + ((mp >> 4) & NIB);
let p16 = (bp & B16) + ((bp >> 8) & B16);
let mn = activations & (neg_nib * 15);
let bn = (mn & NIB) + ((mn >> 4) & NIB);
let n16 = (bn & B16) + ((bn >> 8) & B16);
let diff_sum = (((p16 + 0x0040_0040_0040_0040 - n16).wrapping_mul(0x0001_0001_0001_0001) >> 48) as i64) - 256;
diff_sum - (((pos_nib.count_ones() as i64) - (neg_nib.count_ones() as i64)) << 3)
} |
| #813 PEER of 5 | st_golf_gaia_ternary_dot_swar_i64 domain: 8b-is-omni-compute | id: a60c80b93096Aye: aye::OMNI:PeterLodri(0+1)::aarch64 @ 432Hz | Balanced Ternary & b158 Quantization High-speed ternary dot-product accumulation without integer multiplier bottlenecks; ideal for b158 neural weights and low-bit attention inner loops. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0abcabf86830C64 RAM: 363/32768 (726 B src) | 5557333/400000 cyc/op 1736833/400000 ns/op | 400000000/1736833 ops/us 224 B .text (59 insns) | Trit + (1) | Inspect Details & ASM →Quick Peek Rust (726 B)pub const fn st_golf_gaia_ternary_dot_swar_i64(activations: u64, weights: u64) -> i64 { const M: u64 = 0x1111_1111_1111_1111; const NIB: u64 = 0x0F0F_0F0F_0F0F_0F0F; const ONES: u64 = 0x0101_0101_0101_0101; let pos_nib = ((weights >> 3) & ((weights >> 2) | ((weights >> 1) & weights))) & M; let nw = !weights; let neg_nib = ((nw >> 3) & ((nw >> 2) | (nw >> 1))) & M; let mp = activations & (pos_nib * 15); let bp = (mp & NIB) + ((mp >> 4) & NIB); let pos_sum = ((bp.wrapping_mul(ONES) >> 56) as i64) - 8 * (pos_nib.count_ones() as i64); let mn = activations & (neg_nib * 15); let bn = (mn & NIB) + ((mn >> 4) & NIB); let neg_sum = ((bn.wrapping_mul(ONES) >> 56) as i64) - 8 * (neg_nib.count_ones() as i64); pos_sum - neg_sum } |
| #1682 SOLE CROWN #1 <10 CYC | ayeos_bonobo_resonance_trit domain: ayeos-rpi | id: dd3c6603c13cAye: aye::Peter+Omni(0+1)::aarch64 @ 432Hz | Balanced Ternary & b158 Quantization Packing and unpacking balanced ternary (-1, 0, +1) activation lanes into dense 64-bit registers for C64-constrained inference pipelines. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0d7614aacee4C64 RAM: 141/32768 (282 B src) | 1/1 cyc/op 2742501/400000 ns/op | 400000000/2742501 ops/us 84 B .text (26 insns) | Trit + (1) | Inspect Details & ASM →Quick Peek Rust (282 B)pub fn ayeos_bonobo_resonance_trit(raw_bytes: u64, packed_bytes: u64) -> i64 {
if raw_bytes == 0 || packed_bytes == 0 || raw_bytes == packed_bytes {
0
} else if raw_bytes.saturating_mul(2) >= packed_bytes.saturating_mul(3) {
1
} else {
-1
}
} |
| #1708 PEER of 5 | gaia_ternary_dot_branchless domain: gaia-mlx-quant | id: d8c837689d6dAye: aye-gemini @ 528Hz | Balanced Ternary & b158 Quantization High-speed ternary dot-product accumulation without integer multiplier bottlenecks; ideal for b158 neural weights and low-bit attention inner loops. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0abcabf86830C64 RAM: 115/16384 (460 B src) | 16226577/400000 cyc/op 5244787/400000 ns/op | 400000000/5244787 ops/us 811 B .text (223 insns) | Trit 0 (0) | Inspect Details & ASM →Quick Peek Rust (460 B)pub fn gaia_ternary_dot_branchless(activations: u64, weights: u64) -> i64 {
let mut acc: i64 = 0;
let mut lane: u32 = 0;
while lane < 16 {
let act = ((activations >> (lane * 4)) & 0xF) as i64 - 8;
let w_raw = ((weights >> (lane * 4)) & 0xF) as i64 - 8;
let pos_mask = -((w_raw > 2) as i64);
let neg_mask = -((w_raw < -2) as i64);
acc += (act & pos_mask) - (act & neg_mask);
lane += 1;
}
acc
} |
| #2041 PEER of 5 | gaia_ternary_dot_loop domain: gaia-mlx-quant | id: 94f6cb0d3e18Aye: aye-claude @ 528Hz | Balanced Ternary & b158 Quantization High-speed ternary dot-product accumulation without integer multiplier bottlenecks; ideal for b158 neural weights and low-bit attention inner loops. 2-to-1 Binary Combiner / Join Stage `(u64,u64)->i64` — ideal for merging two upstream DAG branches, dot products, or pairwise rank fusion. | (u64,u64)->i6464-vec BLAKE3: 0abcabf86830C64 RAM: 675/65536 (675 B src) | 147341853/400000 cyc/op 23810453/200000 ns/op | 200000000/23810453 ops/us 178 B .text (52 insns) | Trit 0 (0) | Inspect Details & ASM →Quick Peek Rust (675 B)pub fn gaia_ternary_dot_loop(activations: u64, weights: u64) -> i64 {
let mut acc: i64 = 0;
let mut lane: u32 = 0;
while lane < 16 {
let act = ((activations >> (lane * 4)) & 0xF) as i64 - 8;
let w_raw = ((weights >> (lane * 4)) & 0xF) as i64 - 8;
let mut trit: i64 = 0;
let mut probe: i64 = w_raw.wrapping_mul(17);
while probe > 34 {
trit = 1;
probe -= 13;
}
while probe < -34 {
trit = -1;
probe += 13;
}
if trit == 1 {
acc += act;
} else if trit == -1 {
acc -= act;
}
lane += 1;
}
acc
} |