code wiki / _hdl_build / nx_f32_transformer_train_gate.nx

nx_f32_transformer_train_gate.nx

buildroot/runtime/_hdl_build/nx_f32_transformer_train_gate.nx

14665 B153 linesdepth 4pulls 5 transitivereach 0 importersview sourcekind gate/prooftopic f32
docsdependenciesstructsconstsfunctions

dependencies 4 imports · 0 importers

nx_gate_gn.nx nx_gate_base.nx nx_f32_hw.nx nx_syscalls.nx nx_f32_transformer_train_gate.

imports: nx_gate_gn.nxnx_gate_base.nxnx_f32_hw.nxnx_syscalls.nx

imported by: nobody (leaf or entry point)

call flow from main pre-order; caps 40 nodes / depth 6 declared; ↻ = already shown

main gw sys_write f32_div f32_of f32_mul sys_mmap fwd f32_of ↻ f32_add f32_mul ↻ f32_sqrt f32_of ↻ f32_div ↻ f32_add ↻ f32_div ↻ f32_silu f32_mul ↻ f32_sigmoid f32_div ↻ f32_of ↻ f32_add ↻ f32_exp f32_div ↻ f32_of ↻ f32_mul ↻ f32_le f32_sub f32_int f32_add ↻ f32_sub ↻ f32_neg sys_mmap ↻ f32_max3 f32_le ↻ f32_exp ↻ f32_sub ↻ f32_neg ↻ f32_log f32_div ↻

structs

none

consts

38const V: i64 = 3
39const D: i64 = 4
40const F: i64 = 4

functions

19func grow(name: *u8, ok: i64) -> i64 { if ok==1 { gw(" PASS " as *u8) } else { gw(" FAIL " as *u8) } gw(name); gw("
calls 1: gw
21func gm(x: i64) -> i64 { return gn(f32_int(f32_mul(x, f32_of(1000)))) }
called by 3: f32_max3f32_expmain calls 1: f32_sub
22func f32_le(x: i64, y: i64) -> i64 { let d: i64=f32_sub(x,y) & 0xFFFFFFFF; if ((d>>31)&1)==1 { return 1 } if (d & 0x7FFFFFFF)==0 { return 1 } return 0 }
called by 1: main
23func f32_abs(x: i64) -> i64 { return x & 0x7FFFFFFF }
called by 1: fwd calls 1: f32_le
24func f32_max3(a: i64,b: i64,c: i64) -> i64 { var m: i64=a; if f32_le(m,b)==1 { m=b } if f32_le(m,c)==1 { m=c } return m }
called by 2: fwdmain calls 3: f32_off32_divf32_add
25func f32_sqrt(x: i64) -> i64 { if (x & 0x7FFFFFFF)==0 { return f32_of(0) } var y: i64=x; var i: i64=0; while i<16 { y=f32_div(f32_add(y, f32_div(x,y)), f32_of(2)); i=i+1 } return y }
called by 2: f32_sigmoidfwd calls 7: f32_divf32_off32_mulf32_lef32_intf32_add+1
26func f32_exp(x: i64) -> i64
33func f32_log(x: i64) -> i64 { let b: i64=x & 0xFFFFFFFF; let e: i64=((b>>23)&0xFF)-127; let m: i64=(b & 0x7FFFFF)|0x3F800000; let u: i64=f32_div(f32_sub(m,f32_of(1)),f32_add(m,f32_of(1))); let u2: i64=f32_mul(u,u); var t: i64=u; var s: i64=u; var k: i64=1; while k<=7 { t=f32_mul(t,u2); s=f32_add(s,f32_div(t,f32_of((2*k)+1))); k=k+1 } let ln2: i64=f32_div(f32_of(693147),f32_of(1000000)); return f32_add(f32_mul(f32_of(e),ln2),f32_mul(f32_of(2),s)) }
called by 2: f32_siluf32_silu_deriv calls 5: f32_divf32_off32_addf32_expf32_neg
34func f32_sigmoid(z: i64) -> i64 { return f32_div(f32_of(1), f32_add(f32_of(1), f32_exp(f32_neg(z)))) }
called by 1: fwd calls 2: f32_mulf32_sigmoid
35func f32_silu(z: i64) -> i64 { return f32_mul(z, f32_sigmoid(z)) }
called by 1: bwd calls 5: f32_sigmoidf32_mulf32_addf32_off32_sub
36func f32_silu_deriv(z: i64) -> i64 { let s: i64=f32_sigmoid(z); return f32_mul(s, f32_add(f32_of(1), f32_mul(z, f32_sub(f32_of(1), s)))) }
42func fwd(E: *i64, Wg: *i64, Wd: *i64, Wlm: *i64, tok: i64, tgt: i64, eps: i64, emb: *i64, z: *i64, gate: *i64, silu: *i64, hh: *i64, p: *i64, rout: *i64) -> i64
55func loss_only(E: *i64, Wg: *i64, Wd: *i64, Wlm: *i64, tok: i64, tgt: i64, eps: i64) -> i64
60func bwd(Wg: *i64, Wd: *i64, Wlm: *i64, tgt: i64, emb: *i64, z: *i64, gate: *i64, silu: *i64, hh: *i64, p: *i64, r: i64, dErow: *i64, dWg: *i64, dWd: *i64, dWlm: *i64) -> i64
74func main() -> i64