code wiki / _hdl_build / nx_rv64_jit.nx

nx_rv64_jit.nx

buildroot/runtime/_hdl_build/nx_rv64_jit.nx

26175 B217 linesdepth 5pulls 6 transitivereach 17 importersview sourcekind librarytopic rv64
docsdependenciesstructsconstsfunctions

about

nx_rv64_jit.nx -- a SOVEREIGN RV64 -> x86-64 JIT (dynamic binary translation): the SOTA emulator technique, above the decode-cache interpreter. It TRANSLATES a basic block of predecoded RV64 ops into NATIVE x86-64 machine code (per-instruction, not a shape template): the 32 RV64 registers live in a caller-provided i64 array (base in rdi per SysV); each RV64 instruction emits x86 that loads operands from reg[], computes in rax, stores back; a backward branch emits a NATIVE conditional jump so a loop runs entirely in host code. The block is mmap'd RWX and CALLED (the RWX + cast-fn-ptr primitive is proven by nx_nxe_loader). Equivalence with the golden behavioral sim is proven by nx_rv64_jit_gate; the speedup is measured there. Subset: LUI/ADDI/ADD/SUB/ANDI/ORI/SLLI + BEQ/BNE/BLT/BGE (backward branches). license_tier: ORIGINAL

dependencies 2 imports · 17 importers

nx_syscalls.nx nx_rv64_fast.nx nx_rv64_jit.nx nx_hwtest_matrix.nx nx_jit_ra_probe.nx nx_rv64_cjit_gate.nx nx_rv64_ctier_gate.nx nx_rv64_difftest.nx nx_rv64_jit_fwd_gate.nx nx_rv64_jit_gate.nx nx_rv64_jit_mem_gate.nx nx_rv64_jit_ra_gate.nx nx_rv64_jit_w_gate.nx

diagram shows first 10 each side; +0 more imports, +7 more importers in the complete lists below.

imports: nx_syscalls.nxnx_rv64_fast.nx

imported by: nx_hwtest_matrix.nxnx_jit_ra_probe.nxnx_rv64_cjit_gate.nxnx_rv64_ctier_gate.nxnx_rv64_difftest.nxnx_rv64_jit_fwd_gate.nxnx_rv64_jit_gate.nxnx_rv64_jit_mem_gate.nxnx_rv64_jit_ra_gate.nxnx_rv64_jit_w_gate.nxnx_rv64_m_gate.nxnx_rv64_mulh_gate.nxnx_rv64_realc_gate.nxnx_rv64_run_bin_tier.nxnx_rv64_tier.nxnx_rv64_tier_gate.nxnx_rv64_toolchain_gate.nx

structs

none

consts

none

functions

12func j_b(out: *u8, o: i64, v: i64) -> i64 { out[o]=(v&0xff) as u8; return o+1 }
called by 41: e_loade_storee_movimme_addimme_andimme_orimm+35
13func j_w32(out: *u8, o: i64, v: i64) -> i64 { out[o]=(v&0xff) as u8; out[o+1]=((v>>8)&0xff) as u8; out[o+2]=((v>>16)&0xff) as u8; out[o+3]=((v>>24)&0xff) as u8; return o+4 }
called by 23: e_loade_storee_movimme_addimme_andimme_orimm+17
15func e_load(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x8B); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // mov rax,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
16func e_store(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x89); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // mov [rdi+ri*8],rax
called by 1: jit_compile calls 2: j_bj_w32
17func e_movimm(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC7); p=j_b(out,p,0xC0); p=j_w32(out,p,imm); return p } // mov rax,imm32(sx)
called by 1: jit_compile calls 2: j_bj_w32
18func e_addimm(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x05); p=j_w32(out,p,imm); return p } // add rax,imm32
called by 1: jit_compile calls 2: j_bj_w32
19func e_andimm(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x25); p=j_w32(out,p,imm); return p } // and rax,imm32
called by 1: jit_compile calls 2: j_bj_w32
20func e_orimm(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x0D); p=j_w32(out,p,imm); return p } // or rax,imm32
called by 1: jit_compile calls 2: j_bj_w32
21func e_opreg(out: *u8, o: i64, opc: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,opc); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // <op> rax,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
22func e_imul_mem(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x0F); p=j_b(out,p,0xAF); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // imul rax,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
23func e_mulhi_u(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xF7); p=j_b(out,p,0xA7); p=j_w32(out,p,ri*8); return p } // mul qword[rdi+ri*8] -> rdx:rax (unsigned)
called by 1: jit_compile calls 2: j_bj_w32
24func e_mulhi_s(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xF7); p=j_b(out,p,0xAF); p=j_w32(out,p,ri*8); return p } // imul qword[rdi+ri*8] -> rdx:rax (signed)
called by 1: jit_compile calls 2: j_bj_w32
25func e_mov_rax_rdx(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x89); p=j_b(out,p,0xD0); return p } // mov rax,rdx (grab the high half)
called by 1: jit_compile calls 1: j_b
26func e_shl(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC1); p=j_b(out,p,0xE0); p=j_b(out,p,sh); return p } // shl rax,imm8
called by 1: jit_compile calls 1: j_b
27func e_jcc(out: *u8, o: i64, cc: i64, rel: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x0F); p=j_b(out,p,cc); p=j_w32(out,p,rel); return p } // jcc rel32
called by 1: jit_compile calls 2: j_bj_w32
30func e_load32(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x8B); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // mov eax,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
31func e_op32_mem(out: *u8, o: i64, opc: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,opc); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // <op> eax,[rdi+ri*8] (03 add / 2B sub)
called by 1: jit_compile calls 2: j_bj_w32
32func e_imul32_mem(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x0F); p=j_b(out,p,0xAF); p=j_b(out,p,0x87); p=j_w32(out,p,ri*8); return p } // imul eax,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
33func e_addimm32(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x05); p=j_w32(out,p,imm); return p } // add eax,imm32
called by 1: jit_compile calls 2: j_bj_w32
34func e_shl32(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0xC1); p=j_b(out,p,0xE0); p=j_b(out,p,sh); return p } // shl eax,imm8
called by 1: jit_compile calls 1: j_b
35func e_shr32(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0xC1); p=j_b(out,p,0xE8); p=j_b(out,p,sh); return p } // shr eax,imm8
called by 1: jit_compile calls 1: j_b
36func e_sar32(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0xC1); p=j_b(out,p,0xF8); p=j_b(out,p,sh); return p } // sar eax,imm8
called by 1: jit_compile calls 1: j_b
37func e_movsxd(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x63); p=j_b(out,p,0xC0); return p } // movsxd rax,eax (sext32->64)
called by 1: jit_compile calls 1: j_b
38func e_ret(out: *u8, o: i64) -> i64 { return j_b(out,o,0xC3) }
called by 2: jit_compilejit_compile_ra calls 1: j_b
39func jit_mmap_rwx(size: i64) -> *u8 { let r: i64 = __syscall(SYS_MMAP, 0, size, 7, 0x22, -1, 0); return r as *u8 }
called by 3: jit_runtier_runtier_run_c
40func j_w64(out: *u8, o: i64, v: i64) -> i64 { var i: i64=0; while i<8 { out[o+i]=((v>>(i*8))&0xff) as u8; i=i+1 } return o+8 }
called by 1: e_movimm64_rsi
44func e_movimm64_rsi(out: *u8, o: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xBE); p=j_w64(out,p,imm); return p } // mov rsi, imm64
called by 1: jit_compile calls 2: j_bj_w64
45func e_ld_rdx(out: *u8, o: i64, ri: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x8B); p=j_b(out,p,0x97); p=j_w32(out,p,ri*8); return p } // mov rdx,[rdi+ri*8]
called by 1: jit_compile calls 2: j_bj_w32
46func e_lw_mem(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x63); p=j_b(out,p,0x04); p=j_b(out,p,0x06); return p } // movsxd rax,dword[rsi+rax]
called by 1: jit_compile calls 1: j_b
47func e_ld_mem(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x8B); p=j_b(out,p,0x04); p=j_b(out,p,0x06); return p } // mov rax,[rsi+rax]
called by 1: jit_compile calls 1: j_b
48func e_sw_mem(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x89); p=j_b(out,p,0x14); p=j_b(out,p,0x06); return p } // mov [rsi+rax],edx (store 32)
called by 1: jit_compile calls 1: j_b
49func e_sd_mem(out: *u8, o: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x89); p=j_b(out,p,0x14); p=j_b(out,p,0x06); return p } // mov [rsi+rax],rdx (store 64)
called by 1: jit_compile calls 1: j_b
50func e_shr(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC1); p=j_b(out,p,0xE8); p=j_b(out,p,sh); return p } // shr rax,imm8 (SRLI)
called by 1: jit_compile calls 1: j_b
51func e_sar(out: *u8, o: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC1); p=j_b(out,p,0xF8); p=j_b(out,p,sh); return p } // sar rax,imm8 (SRAI)
called by 1: jit_compile calls 1: j_b
56func jit_set_membase(v: i64, membytes: i64) -> i64 { jit_membase = v; jit_membytes = membytes; return 0 }
called by 7: mainrun_jit_memmainmainjit_mtier_run+1
61func e_bcheck(out: *u8, o: i64, width: i64) -> i64
called by 1: jit_compile calls 2: j_bj_w32
71func jit_compile(opk: *i64, rd: *i64, rs1: *i64, rs2: *i64, imm: *i64, nidx: i64, out: *u8, x86off: *i64) -> i64
called by 12: mainrun_jitmainmainrun_jit_memmain+6 calls 34: sys_mmapj_be_movimm64_rsie_loade_addimme_store+28
135func jit_run(x86: *u8, len: i64, reg: *i64) -> i64
called by 9: mainrun_jitmainrun_jit_memmainrun_ra+3 calls 1: jit_mmap_rwx
150func ra_pool(s: i64) -> i64 { if s==0 { return 1 } if s==1 { return 2 } return 6 } // rcx, rdx, rsi
called by 2: jit_ra_allocjit_compile_ra
151func era_fill(out: *u8, o: i64, preg: i64, disp: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x8B); p=j_b(out,p,0x87|(preg<<3)); p=j_w32(out,p,disp); return p } // mov preg,[rdi+disp]
called by 1: jit_compile_ra calls 2: j_bj_w32
152func era_spill(out: *u8, o: i64, preg: i64, disp: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x89); p=j_b(out,p,0x87|(preg<<3)); p=j_w32(out,p,disp); return p } // mov [rdi+disp],preg
called by 1: jit_compile_ra calls 2: j_bj_w32
153func era_mov_rr(out: *u8, o: i64, dst: i64, src: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x8B); p=j_b(out,p,0xC0|(dst<<3)|src); return p } // mov dst,src
called by 1: jit_compile_ra calls 1: j_b
154func era_alu_rr(out: *u8, o: i64, opc: i64, dst: i64, src: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,opc); p=j_b(out,p,0xC0|(dst<<3)|src); return p } // <op> dst,src (03/2B/23/0B/3B)
called by 1: jit_compile_ra calls 1: j_b
155func era_alu_ri(out: *u8, o: i64, slash: i64, dst: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0x81); p=j_b(out,p,0xC0|(slash<<3)|dst); p=j_w32(out,p,imm); return p } // add(/0)/or(/1)/and(/4) dst,imm32
called by 1: jit_compile_ra calls 2: j_bj_w32
156func era_shl_ri(out: *u8, o: i64, dst: i64, sh: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC1); p=j_b(out,p,0xE0|dst); p=j_b(out,p,sh); return p } // shl dst,imm8
called by 1: jit_compile_ra calls 1: j_b
157func era_mov_ri(out: *u8, o: i64, dst: i64, imm: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x48); p=j_b(out,p,0xC7); p=j_b(out,p,0xC0|dst); p=j_w32(out,p,imm); return p } // mov dst,imm32(sx)
called by 1: jit_compile_ra calls 2: j_bj_w32
158func era_jcc(out: *u8, o: i64, cc: i64, rel: i64) -> i64 { var p: i64=o; p=j_b(out,p,0x0F); p=j_b(out,p,cc); p=j_w32(out,p,rel); return p }
called by 1: jit_compile_ra calls 2: j_bj_w32
160func jit_ra_alloc(rvmap: *i64, nbox: *i64, r: i64) -> i64 { if r==0 { return 0 } if rvmap[r]>=0 { return 0 } if nbox[0]>=3 { return 0-2 } rvmap[r]=ra_pool(nbox[0]); nbox[0]=nbox[0]+1; return 0 }
called by 1: jit_compile_ra calls 1: ra_pool
164func jit_compile_ra(opk: *i64, rd: *i64, rs1: *i64, rs2: *i64, imm: *i64, nidx: i64, out: *u8, x86off: *i64, rvmap: *i64) -> i64
called by 3: mainrun_ramain calls 13: sys_mmapjit_ra_allocj_bra_poolera_fillera_mov_ri+7
217func jit_main_ignore() -> i64 { return 0 }