On Tue, Sep 29, 2026 at 5:53 PM Matt Turner <[email protected]> wrote: > > On Sun, Sep 27, 2026 at 5:00 AM Richard Henderson > <[email protected]> wrote: > > > > Signed-off-by: Richard Henderson <[email protected]> > > --- > > include/tcg/tcg-op-common.h | 2 -- > > include/tcg/tcg-op-def.h.inc | 1 + > > tcg/tcg-op.c | 15 +++++---------- > > 3 files changed, 6 insertions(+), 12 deletions(-) > > > > diff --git a/include/tcg/tcg-op-common.h b/include/tcg/tcg-op-common.h > > index 8444044ed57..ff15e3b88ae 100644 > > --- a/include/tcg/tcg-op-common.h > > +++ b/include/tcg/tcg-op-common.h > > @@ -124,7 +124,6 @@ void tcg_gen_plugin_mem_cb(TCGv_i64 addr, unsigned > > meminfo); > > /* 32 bit ops */ > > > > void tcg_gen_subfi_i32(TCGv_i32 ret, int32_t arg1, TCGv_i32 arg2); > > -void tcg_gen_subi_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2); > > void tcg_gen_andi_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2); > > void tcg_gen_ori_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2); > > void tcg_gen_xori_i32(TCGv_i32 ret, TCGv_i32 arg1, int32_t arg2); > > @@ -225,7 +224,6 @@ void tcg_gen_not_i32(TCGv_i32 ret, TCGv_i32 arg); > > /* 64 bit ops */ > > > > void tcg_gen_subfi_i64(TCGv_i64 ret, int64_t arg1, TCGv_i64 arg2); > > -void tcg_gen_subi_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2); > > void tcg_gen_andi_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2); > > void tcg_gen_ori_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2); > > void tcg_gen_xori_i64(TCGv_i64 ret, TCGv_i64 arg1, int64_t arg2); > > diff --git a/include/tcg/tcg-op-def.h.inc b/include/tcg/tcg-op-def.h.inc > > index 51d5ec6e06a..55249c4dd6f 100644 > > --- a/include/tcg/tcg-op-def.h.inc > > +++ b/include/tcg/tcg-op-def.h.inc > > @@ -7,3 +7,4 @@ DEF1(discard, TCGV) > > DEF2(mov, TCGV, TCGV) > > DEF2(movi, TCGV, TINT) > > DEF3(sub, TCGV, TCGV, TCGV) > > +DEF3(subi, TCGV, TCGV, TINT) > > diff --git a/tcg/tcg-op.c b/tcg/tcg-op.c > > index 2865de65d39..9bcf79ccbfc 100644 > > --- a/tcg/tcg-op.c > > +++ b/tcg/tcg-op.c > > @@ -399,6 +399,11 @@ void tcg_gen_sub(TCGType type, TCGTemp *dst, TCGTemp > > *src1, TCGTemp *src2) > > tcg_gen_op_ttt(INDEX_op_sub, type, dst, src1, src2); > > } > > > > +void tcg_gen_subi(TCGType type, TCGTemp *dst, TCGTemp *src1, int64_t src2) > > +{ > > + tcg_gen_addi(type, dst, src1, -src2); > > Double check me on this, but since src2's type changed from int32_t to > int64_t, I think this will produce a value that isn't sign-extended > for I32. > > E.g. for I32 with src2 == INT32_MIN this now produces 0x80000000 > rather than the 0xffffffff80000000 the old code produced. > tcg_constant_internal() doesn't canonicalize, so we end up with an I32 > constant that isn't sign-extended, which optimize.c assumes can't > happen. > > Going to make a quick sweep over the rest of the series to look for > this issue specifically.
Found similar cases in 087 and 096. If this is a real problem, it might be simplest to sign-extend I32 values in tcg_constant_internal() rather than fix each caller?
