The current code generates the following for convertFloat32ToInt32 with the error noted: 0x405137b8: vcvt.s32.f32 s2, s0 0x405137bc: vmov r1, s2 0x405137c0: vcvt.f32.s32 s2, s2 0x405137c4: vcmp.f32 s0, s2 0x405137c8: vmrs APSR_nzcv, fpscr 0x405137cc: bne 0x405138e4 0x405137d0: cmp r1, #0 0x405137d4: vmoveq.32 r1, d0[0] <<<< error 0x405137d8: cmpeq r1, #-2147483648 ; 0x80000000 0x405137ec: beq 0x405138e4 0x405137e0: cmp r1, #7 The follow appears to be what is needed: 0x40513a44: vldr s0, [pc, #204] ; 0x40513b18 0x40513a48: vcvt.s32.f32 s2, s0 0x40513a4c: vmov r1, s2 0x40513a50: vcvt.f32.s32 s2, s2 0x40513a54: vcmp.f32 s0, s2 0x40513a58: vmrs APSR_nzcv, fpscr 0x40513a5c: bne 0x40513bc8 0x40513a60: cmp r1, #0 0x40513a64: vmoveq r1, s0 <<<< Correction 0x40513a68: cmpeq r1, #-2147483648 ; 0x80000000 0x40513a6c: beq 0x40513bc8

Comment on attachment 811143 [details] [diff] [review] Fix the negative zero check path in convertFloat32ToInt32 Review of attachment 811143 [details] [diff] [review]: ----------------------------------------------------------------- I clearly think Marty is more qualified for this review than I am.

Comment on attachment 811143 [details] [diff] [review] Fix the negative zero check path in convertFloat32ToInt32 Review of attachment 811143 [details] [diff] [review]: ----------------------------------------------------------------- I swear, I'm actually on vacation.