Re: [Qemu-devel] [PATCH v2 3/5] target-m68k: Inline shifts

[Laurent Vivier]

Le 09/11/2016 à 14:46, Richard Henderson a écrit :
> diff --git a/target-m68k/translate.c b/target-m68k/translate.c
> index 4f224d7..1b3765f 100644
> --- a/target-m68k/translate.c
> +++ b/target-m68k/translate.c

> +static inline void shift_reg(DisasContext *s, uint16_t insn, int opsize)
> +{
> +    int logical = insn & 8;
> +    int left = insn & 0x100;
> +    int bits = opsize_bytes(opsize) * 8;
> +    TCGv reg = gen_extend(DREG(insn, 0), opsize, !logical);
> +    TCGv s32;
> +    TCGv_i64 t64, s64;
> +
> +    t64 = tcg_temp_new_i64();
> +    s64 = tcg_temp_new_i64();
> +    s32 = tcg_temp_new();
> +
> +    /* Note that m68k truncates the shift count modulo 64, not 32.
> +       In addition, a 64-bit shift makes it easy to find "the last
> +       bit shifted out", for the carry flag.  */
> +    tcg_gen_andi_i32(s32, DREG(insn, 9), 63);
> +    tcg_gen_extu_i32_i64(s64, s32);
> +    tcg_gen_extu_i32_i64(t64, reg);
> +
> +    /* Optimistically set V=0.  Also used as a zero source below.  */
> +    tcg_gen_movi_i32(QREG_CC_V, 0);
> +    if (left) {
> +        tcg_gen_shl_i64(t64, t64, s64);
> +
> +        tcg_gen_extr_i64_i32(QREG_CC_N, QREG_CC_C, t64);

This does not extract correctly the C flag when the opsize is word or byte.
I think we should use a shift instead:

-        tcg_gen_extr_i64_i32(QREG_CC_N, QREG_CC_C, t64);
-
-        /* Note that C=0 if shift count is 0, and we get that for free.  */
+        if (opsize == OS_LONG) {
+            tcg_gen_extr_i64_i32(QREG_CC_N, QREG_CC_C, t64);
+            /* Note that C=0 if shift count is 0, and we get that for
free.  */
+        } else {
+            TCGv zero = tcg_const_i32(0);
+            tcg_gen_extrl_i64_i32(QREG_CC_N, t64);
+            tcg_gen_shri_i64(t64, t64, bits);
+            tcg_gen_extrl_i64_i32(QREG_CC_C, t64);
+            tcg_gen_movcond_i32(TCG_COND_EQ, QREG_CC_C,
+                                s32, zero, zero, QREG_CC_C);
+            tcg_temp_free(zero);
+        }

Do you have a better idea?

thanks,
Laurent