ref: 6c2f8fd5e4f0e0be44495a5526930fa3760cd87e
dir: /libkern/vlop-sparc.s/
TEXT _mulv(SB), $0 MOVW u1+8(FP), R8 MOVW u2+16(FP), R13 MOVW R13, R16 /* save low parts for later */ MOVW R8, R12 /* * unsigned 32x32 => 64 multiply */ CMP R13, R8 BLE mul1 MOVW R12, R13 MOVW R16, R8 mul1: MOVW R13, Y ANDNCC $0xFFF, R13, R0 BE mul_shortway ANDCC R0, R0, R9 /* zero partial product and clear N and V cond's */ /* long multiply */ MULSCC R8, R9, R9 /* 0 */ MULSCC R8, R9, R9 /* 1 */ MULSCC R8, R9, R9 /* 2 */ MULSCC R8, R9, R9 /* 3 */ MULSCC R8, R9, R9 /* 4 */ MULSCC R8, R9, R9 /* 5 */ MULSCC R8, R9, R9 /* 6 */ MULSCC R8, R9, R9 /* 7 */ MULSCC R8, R9, R9 /* 8 */ MULSCC R8, R9, R9 /* 9 */ MULSCC R8, R9, R9 /* 10 */ MULSCC R8, R9, R9 /* 11 */ MULSCC R8, R9, R9 /* 12 */ MULSCC R8, R9, R9 /* 13 */ MULSCC R8, R9, R9 /* 14 */ MULSCC R8, R9, R9 /* 15 */ MULSCC R8, R9, R9 /* 16 */ MULSCC R8, R9, R9 /* 17 */ MULSCC R8, R9, R9 /* 18 */ MULSCC R8, R9, R9 /* 19 */ MULSCC R8, R9, R9 /* 20 */ MULSCC R8, R9, R9 /* 21 */ MULSCC R8, R9, R9 /* 22 */ MULSCC R8, R9, R9 /* 23 */ MULSCC R8, R9, R9 /* 24 */ MULSCC R8, R9, R9 /* 25 */ MULSCC R8, R9, R9 /* 26 */ MULSCC R8, R9, R9 /* 27 */ MULSCC R8, R9, R9 /* 28 */ MULSCC R8, R9, R9 /* 29 */ MULSCC R8, R9, R9 /* 30 */ MULSCC R8, R9, R9 /* 31 */ MULSCC R0, R9, R9 /* 32; shift only; r9 is high part */ /* * need to correct top word if top bit set */ CMP R8, R0 BGE mul_tstlow ADD R13, R9 /* adjust the high parts */ mul_tstlow: MOVW Y, R13 /* get low part */ BA mul_done mul_shortway: ANDCC R0, R0, R9 /* zero partial product and clear N and V cond's */ MULSCC R8, R9, R9 /* 0 */ MULSCC R8, R9, R9 /* 1 */ MULSCC R8, R9, R9 /* 2 */ MULSCC R8, R9, R9 /* 3 */ MULSCC R8, R9, R9 /* 4 */ MULSCC R8, R9, R9 /* 5 */ MULSCC R8, R9, R9 /* 6 */ MULSCC R8, R9, R9 /* 7 */ MULSCC R8, R9, R9 /* 8 */ MULSCC R8, R9, R9 /* 9 */ MULSCC R8, R9, R9 /* 10 */ MULSCC R8, R9, R9 /* 11 */ MULSCC R0, R9, R9 /* 12; shift only; r9 is high part */ MOVW Y, R8 /* make low part of partial low part & high part */ SLL $12, R9, R13 SRL $20, R8 OR R8, R13 SRA $20, R9 /* high part */ mul_done: /* * mul by high halves if needed */ MOVW R13, 4(R7) MOVW u2+12(FP), R11 CMP R11, R0 BE nomul1 MUL R11, R12 ADD R12, R9 nomul1: MOVW u1+4(FP), R11 CMP R11, R0 BE nomul2 MUL R11, R16 ADD R16, R9 nomul2: MOVW R9, 0(R7) RETURN