[Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[Laurent Desnogues]

[-- Attachment #1: Type: text/plain, Size: 165 bytes --]

Hello,

this patch:

  - fixes argument numbers for div2 and divu2
  - implements TCG not.


Laurent

Signed-off-by: Laurent Desnogues <laurent.desnogues@gmail.com>

[-- Attachment #2: arm-codegen2.patch --]
[-- Type: application/octet-stream, Size: 1692 bytes --]

diff --git a/tcg/arm/tcg-target.c b/tcg/arm/tcg-target.c
index 7ef2b89..53877cf 100644
--- a/tcg/arm/tcg-target.c
+++ b/tcg/arm/tcg-target.c
@@ -1427,6 +1427,10 @@ static inline void tcg_out_op(TCGContext *s, int opc,
     case INDEX_op_neg_i32:
         tcg_out_dat_imm(s, COND_AL, ARITH_RSB, args[0], args[1], 0);
         break;
+    case INDEX_op_not_i32:
+        tcg_out_dat_reg(s, COND_AL,
+                        ARITH_MVN, args[0], 0, args[1], SHIFT_IMM_LSL(0));
+        break;
     case INDEX_op_mul_i32:
         tcg_out_mul32(s, COND_AL, args[0], args[1], args[2]);
         break;
@@ -1555,12 +1559,13 @@ static const TCGTargetOpDef arm_op_defs[] = {
     { INDEX_op_sub_i32, { "r", "r", "rI" } },
     { INDEX_op_mul_i32, { "r", "r", "r" } },
     { INDEX_op_mulu2_i32, { "r", "r", "r", "r" } },
-    { INDEX_op_div2_i32, { "r", "r", "r", "1", "2" } },
-    { INDEX_op_divu2_i32, { "r", "r", "r", "1", "2" } },
+    { INDEX_op_div2_i32, { "r", "r", "r", "0", "1" } },
+    { INDEX_op_divu2_i32, { "r", "r", "r", "0", "1" } },
     { INDEX_op_and_i32, { "r", "r", "rI" } },
     { INDEX_op_or_i32, { "r", "r", "rI" } },
     { INDEX_op_xor_i32, { "r", "r", "rI" } },
     { INDEX_op_neg_i32, { "r", "r" } },
+    { INDEX_op_not_i32, { "r", "r" } },
 
     { INDEX_op_shl_i32, { "r", "r", "ri" } },
     { INDEX_op_shr_i32, { "r", "r", "ri" } },
diff --git a/tcg/arm/tcg-target.h b/tcg/arm/tcg-target.h
index 7ff2928..71e1ec5 100644
--- a/tcg/arm/tcg-target.h
+++ b/tcg/arm/tcg-target.h
@@ -33,6 +33,7 @@
 #define TCG_TARGET_HAS_ext16s_i32
 #define TCG_TARGET_HAS_neg_i32
 #undef TCG_TARGET_HAS_neg_i64
+#define TCG_TARGET_HAS_not_i32
 #undef TCG_TARGET_STACK_GROWSUP
 
 enum {

Re: [Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[Filip Navara]

On Fri, Jul 24, 2009 at 7:23 PM, Laurent
Desnogues<laurent.desnogues@gmail.com> wrote:
> Hello,
>
> this patch:
>
>  - fixes argument numbers for div2 and divu2
>  - implements TCG not.
>
>
> Laurent
>
> Signed-off-by: Laurent Desnogues <laurent.desnogues@gmail.com>
>

Looks good to me.

Acked-by: Filip Navara <filip.navara@gmail.com>

Best regards,
Filip Navara

Re: [Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[Laurent Desnogues]

A quick comment about the div part after a discussion with malc.
The ARM back-end doesn't support div TCG ops, I just fixed a
failure when compiling with TCG debug enabled.  The proper
fix is perhaps to simply remove the two lines.


Laurent

Re: [Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[andrzej zaborowski]

2009/7/24 Laurent Desnogues <laurent.desnogues@gmail.com>:
> Hello,
>
> this patch:
>
>  - fixes argument numbers for div2 and divu2

But "0" is not defined in target_parse_constraint?  Does div2 not work
as is now?

>  - implements TCG not.

Pushed this part.

Cheers

Re: [Qemu-devel] [PATCH] ARM back-end: Fix register numbering for div/divu and add TCG not

[Laurent Desnogues]

On Sat, Aug 22, 2009 at 2:03 PM, andrzej zaborowski <balrogg@gmail.com> wrote:
> 2009/7/24 Laurent Desnogues <laurent.desnogues@gmail.com>:
>> Hello,
>>
>> this patch:
>>
>>  - fixes argument numbers for div2 and divu2
>
> But "0" is not defined in target_parse_constraint?  Does div2 not work
> as is now?

Given Gary Thomas latest question, I guess it's time to
resurrect that discussion :-)

Constraints defined as numbers are handled internally
by TCG in tcg_add_target_add_op_defs.  My understanding
is that they mean that an input and an output arg are aliased.
I don't think it's what the constraint for div2 and divu2 is
trying to express.

Also the div helper looks broken:  it potentially restores less
registers than what it saved, and uses ldrt.

My proposed fix was completely wrong, but I have no time
to make a correct one.


Laurent