All of lore.kernel.org
 help / color / mirror / Atom feed
From: James Hogan <james.hogan@imgtec.com>
To: Richard Henderson <rth@twiddle.net>
Cc: qemu-devel@nongnu.org, aurelien@aurel32.net
Subject: Re: [Qemu-devel] [PATCH 11/15] tcg-mips: Use mips64r6 instructions in tcg_out_movi
Date: Wed, 10 Feb 2016 00:32:51 +0000	[thread overview]
Message-ID: <20160210003251.GE3678@jhogan-linux.le.imgtec.org> (raw)
In-Reply-To: <20160209165052.GC3678@jhogan-linux.le.imgtec.org>

[-- Attachment #1: Type: text/plain, Size: 3254 bytes --]

Hi Richard,

On Tue, Feb 09, 2016 at 04:50:52PM +0000, James Hogan wrote:
> > @@ -589,6 +608,50 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
> >      }
> >      if (TCG_TARGET_REG_BITS == 32 || arg == (int32_t)arg) {
> >          tcg_out_opc_imm(s, OPC_LUI, ret, TCG_REG_ZERO, arg >> 16);
> > +    } else if (use_mips32r6_instructions) {
> > +        tcg_target_long disp = arg - (intptr_t)s->code_ptr;
> > +        if (disp == sextract32(disp, 2, 19) * 4) {
> > +            tcg_out_opc_pc19(s, OPC_ADDIUPC, ret, disp >> 2);
> > +            return;
> > +        } else if ((disp & ~(tcg_target_long)0xffff)
> > +                   == sextract32(disp, 16, 16) * 0x10000) {
> > +            tcg_out_opc_imm(s, OPC_ALUIPC, ret, 0, disp >> 16);
> 
> I think ret and 0 are the wrong way around here. You're putting 0 in rs
> (the destination register), which causes a seg fault.
> 
> OUT: [size=56] 
> 0xfff30b0064:  lw       s1,-8(s0) 
> 0xfff30b0068:  bnezalc  zero,s1,0xfff30b0090 
> 0xfff30b006c:  nop 
> 0xfff30b0070:  j        0xfff0000000 
> 0xfff30b0074:  nop 
> 0xfff30b0078:  lui      s1,0xbfc0 
> 0xfff30b007c:  ori      s1,s1,0x580 
> 0xfff30b0080:  sd       s1,256(s0) 
> 0xfff30b0084:  aluipc   zero,0xfeb7 
> 0xfff30b0088:  j        0xfff30b0034 
> 0xfff30b008c:  ori      v0,v0,0x4010 
> 0xfff30b0090:  aluipc   zero,0xfeb7 
> 0xfff30b0094:  j        0xfff30b0034 
> 0xfff30b0098:  ori      v0,v0,0x4013

Actually, still not quite right.

ALUIPC does
dest <- ~0xffff & (PC + sign_extend(imm16<<16))

which is effectively
dest <- PC & ~0xffff + sign_extend(imm16<<16)

so disp should be between arg and code_ptr & ~0xffff, i.e. something
like this I think:

diff --git a/tcg/mips/tcg-target.c b/tcg/mips/tcg-target.c
index 8205ea4e159f..9a5d31478797 100644
--- a/tcg/mips/tcg-target.c
+++ b/tcg/mips/tcg-target.c
@@ -666,12 +666,13 @@ static void tcg_out_movi(TCGContext *s, TCGType type,
         tcg_out_opc_imm(s, OPC_LUI, ret, TCG_REG_ZERO, arg >> 16);
     } else if (use_mips32r6_instructions) {
         tcg_target_long disp = arg - (intptr_t)s->code_ptr;
+        tcg_target_long disphi = arg - ((intptr_t)s->code_ptr & ~(tcg_target_long)0xffff);
         if (disp == sextract32(disp, 2, 19) * 4) {
             tcg_out_opc_pc19(s, OPC_ADDIUPC, ret, disp >> 2);
             return;
-        } else if ((disp & ~(tcg_target_long)0xffff)
-                   == sextract32(disp, 16, 16) * 0x10000) {
-            tcg_out_opc_imm(s, OPC_ALUIPC, 0, ret, disp >> 16);
+        } else if ((disphi & ~(tcg_target_long)0xffff)
+                   == sextract32(disphi, 16, 16) * 0x10000) {
+            tcg_out_opc_imm(s, OPC_ALUIPC, 0, ret, disphi >> 16);
         } else {
             TCGReg in = TCG_REG_ZERO;
             tcg_target_long tmp = (int16_t)arg;

Otherwise, in this case its trying to load the immediate 0xfff1c30000
relative to 0xfff30b0084, and calculates a disp of FEB7FF7C, which is
truncated to 0xFEB7. The result is then:
0xfff30b0000 + (int)0xfeb70000 = 0xfff1c20000
which is off by 64KiB.

With the above change we get:
disphi = 0xfeb80000
and the result is then:
0xfff30b0000 + (int)0xfeb80000 = 0xfff1c30000

Cheers
James

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

  parent reply	other threads:[~2016-02-10  0:32 UTC|newest]

Thread overview: 29+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-02-09 10:39 [Qemu-devel] [PATCH 00/15] tcg mips64 and mipsr6 improvements Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 01/15] tcg-mips: Add mips64 opcodes Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 02/15] tcg-mips: Support 64-bit opcodes Richard Henderson
2016-02-09 15:24   ` James Hogan
2016-02-09 17:16     ` Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 03/15] tcg-mips: Adjust move functions for mips64 Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 04/15] tcg-mips: Adjust load/store " Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 05/15] tcg-mips: Adjust prologue " Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 06/15] tcg-mips: Add tcg unwind info Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 07/15] tcg-mips: Adjust qemu_ld/st for mips64 Richard Henderson
2016-02-10 16:34   ` James Hogan
2016-02-10 17:35     ` Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 08/15] tcg-mips: Adjust calling conventions " Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 09/15] tcg-mips: Fix exit_tb " Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 10/15] tcg-mips: Move bswap code to subroutines Richard Henderson
2016-02-09 10:39 ` [Qemu-devel] [PATCH 11/15] tcg-mips: Use mips64r6 instructions in tcg_out_movi Richard Henderson
2016-02-09 16:50   ` James Hogan
2016-02-09 17:20     ` Richard Henderson
2016-02-09 17:25     ` Richard Henderson
2016-02-10  0:32     ` James Hogan [this message]
2016-02-09 10:40 ` [Qemu-devel] [PATCH 12/15] tcg-mips: Use mips64r6 instructions in tcg_out_ldst Richard Henderson
2016-02-09 10:40 ` [Qemu-devel] [PATCH 13/15] tcg-mips: Use mips64r6 instructions in constant addition Richard Henderson
2016-02-09 10:40 ` [Qemu-devel] [PATCH 14/15] tcg-mips: Use mipsr6 instructions in branches Richard Henderson
2016-02-09 16:22   ` James Hogan
2016-02-09 17:13     ` Richard Henderson
2016-02-09 18:46       ` Maciej W. Rozycki
2016-02-10  0:20     ` James Hogan
2016-02-09 10:40 ` [Qemu-devel] [PATCH 15/15] tcg-mips: Use mipsr6 instructions in calls Richard Henderson
2016-02-10 12:49   ` James Hogan

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20160210003251.GE3678@jhogan-linux.le.imgtec.org \
    --to=james.hogan@imgtec.com \
    --cc=aurelien@aurel32.net \
    --cc=qemu-devel@nongnu.org \
    --cc=rth@twiddle.net \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.