[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH] mips: Fix the 64-bit case for microMIPS MOVE16
From: |
Leon Alrae |
Subject: |
Re: [Qemu-devel] [PATCH] mips: Fix the 64-bit case for microMIPS MOVE16 and MOVEP |
Date: |
Mon, 24 Nov 2014 11:31:45 +0000 |
User-agent: |
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:24.0) Gecko/20100101 Thunderbird/24.5.0 |
On 12/11/2014 15:21, Maciej W. Rozycki wrote:
> Fix microMIPS MOVE16 and MOVEP instructions on 64-bit processors by
> using register addition operations.
>
> This copies the approach taken with MIPS16 MOVE instructions (I8_MOV32R
> and I8_MOVR32 opcodes) and follows the observation that OPC_ADDU expands
> to tcg_gen_mov_tl whenever `rt' is 0 and `rs' is not, therefore copying
> `rs' to `rd' verbatim. This is not the case with OPC_ADDIU where a
> sign-extension from bit #31 is made, unless in the uninteresting case of
> `rs' being 0, losing the upper 32 bits of the value copied for any
> proper 64-bit values.
>
> This also serves as an optimization as one op is produced in generated
> code rather than two (again, unless `rs' is 0, where it doesn't change
> anything).
>
> Signed-off-by: Maciej W. Rozycki <address@hidden>
> ---
> This is rather obvious, but I also pushed it through full bare-iron GCC
> regression testing with an o32 big-endian microMIPS multilib. That
> includes several instances of both instructions. No changes in results
> were observed with the patch applied compared to the original version.
>
> I wonder if all these move operations shouldn't actually be switched to
> using OPC_OR that is agnostic to the machine word size regardless of the
> operand selection. But that is something to consider separately.
>
> So meanwhile, please apply.
>
> Maciej
>
> qemu-umips-move.diff
> Index: qemu-git-trunk/target-mips/translate.c
> ===================================================================
> --- qemu-git-trunk.orig/target-mips/translate.c 2014-11-02
> 17:57:16.998924336 +0000
> +++ qemu-git-trunk/target-mips/translate.c 2014-11-02 17:57:19.498930155
> +0000
> @@ -13492,8 +13492,8 @@ static int decode_micromips_opc (CPUMIPS
> rs = rs_rt_enc[enc_rs];
> rt = rs_rt_enc[enc_rt];
>
> - gen_arith_imm(ctx, OPC_ADDIU, rd, rs, 0);
> - gen_arith_imm(ctx, OPC_ADDIU, re, rt, 0);
> + gen_arith(ctx, OPC_ADDU, rd, rs, 0);
> + gen_arith(ctx, OPC_ADDU, re, rt, 0);
> }
> break;
> case LBU16:
> @@ -13574,7 +13574,7 @@ static int decode_micromips_opc (CPUMIPS
> int rd = uMIPS_RD5(ctx->opcode);
> int rs = uMIPS_RS5(ctx->opcode);
>
> - gen_arith_imm(ctx, OPC_ADDIU, rd, rs, 0);
> + gen_arith(ctx, OPC_ADDU, rd, rs, 0);
> }
> break;
> case ANDI16:
Reviewed-by: Leon Alrae <address@hidden>
All the patches up to this one have been applied to mips-next branch
(available at git://github.com/lalrae/qemu.git), thanks. I'll go through
the remaining soon.
Regards,
Leon