PATCH: S/390: Fix instruction type of efpc and sfpc - take 2

PATCH: S/390: Fix instruction type of efpc and sfpc - take 2

[Andreas Krebbel]

Hi,

>> the "extract fpc" and "set fpc" instructions take only one register
>> > argument.  Fixed with the attached patch.
>> > 
>> > OK?
>
>Technically correct but this will break existing applications. In
>particular the glibc won't compile anymore. I'm not sure if we want to
>do that.

So someone indeed added a pointless dummy parameter to glibc inline assembly
instructions instead of fixing a typo in binutils - mmmh.

As you've proposed offline, I've made the extra operand optional so that
now at least the correct versions of the instructions are accepted as well.
objdump still prints efpc and sfpc with the dummy operand.  That's a bit
annoying but at least backward compatible.

OK?

-Andreas-


2007-02-16  Andreas Krebbel  <krebbel1@de.ibm.com>

	* opcodes/s390-opc.txt ("efpc", "sfpc"): Set to RRE_RR_OPT instruction
	type.
	* opcodes/s390-opc.c (s390_operands): Add RO_28 as optional gpr.
	(INSTR_RRE_RR_OPT, MASK_RRE_RR_OPT): New instruction type for efpc
	and sfpc.


Index: opcodes/s390-opc.txt
===================================================================
*** opcodes/s390-opc.txt.orig	2007-02-16 09:49:57.000000000 +0100
--- opcodes/s390-opc.txt	2007-02-16 09:50:38.000000000 +0100
*************** b30d debr RRE_FF "divide short bfp" g5 e
*** 354,360 ****
  ed000000000d deb RXE_FRRD "divide short bfp" g5 esa,zarch
  b35b didbr RRF_FUFF "divide to integer long bfp" g5 esa,zarch
  b353 diebr RRF_FUFF "divide to integer short bfp" g5 esa,zarch
! b38c efpc RRE_RR "extract fpc" g5 esa,zarch
  b342 ltxbr RRE_FF "load and test extended bfp" g5 esa,zarch
  b312 ltdbr RRE_FF "load and test long bfp" g5 esa,zarch
  b302 ltebr RRE_FF "load and test short bfp" g5 esa,zarch
--- 354,360 ----
  ed000000000d deb RXE_FRRD "divide short bfp" g5 esa,zarch
  b35b didbr RRF_FUFF "divide to integer long bfp" g5 esa,zarch
  b353 diebr RRF_FUFF "divide to integer short bfp" g5 esa,zarch
! b38c efpc RRE_RR_OPT "extract fpc" g5 esa,zarch
  b342 ltxbr RRE_FF "load and test extended bfp" g5 esa,zarch
  b312 ltdbr RRE_FF "load and test long bfp" g5 esa,zarch
  b302 ltebr RRE_FF "load and test short bfp" g5 esa,zarch
*************** b31f msdbr RRF_F0FF "multiply and subtra
*** 397,403 ****
  ed000000001f msdb RXF_FRRDF "multiply and subtract long bfp" g5 esa,zarch
  b30f msebr RRF_F0FF "multiply and subtract short bfp" g5 esa,zarch
  ed000000000f mseb RXF_FRRDF "multiply and subtract short bfp" g5 esa,zarch
! b384 sfpc RRE_RR "set fpc" g5 esa,zarch
  b299 srnm S_RD "set rounding mode" g5 esa,zarch
  b316 sqxbr RRE_FF "square root extended bfp" g5 esa,zarch
  b315 sqdbr RRE_FF "square root long bfp" g5 esa,zarch
--- 397,403 ----
  ed000000001f msdb RXF_FRRDF "multiply and subtract long bfp" g5 esa,zarch
  b30f msebr RRF_F0FF "multiply and subtract short bfp" g5 esa,zarch
  ed000000000f mseb RXF_FRRDF "multiply and subtract short bfp" g5 esa,zarch
! b384 sfpc RRE_RR_OPT "set fpc" g5 esa,zarch
  b299 srnm S_RD "set rounding mode" g5 esa,zarch
  b316 sqxbr RRE_FF "square root extended bfp" g5 esa,zarch
  b315 sqdbr RRE_FF "square root long bfp" g5 esa,zarch
Index: opcodes/s390-opc.c
===================================================================
*** opcodes/s390-opc.c.orig	2007-02-16 09:49:57.000000000 +0100
--- opcodes/s390-opc.c	2007-02-16 15:17:39.000000000 +0100
*************** const struct s390_operand s390_operands[
*** 133,139 ****
  #define U32_16 41		  /* 32 bit unsigned value starting at 16 */
    { 32, 16, 0 },
  #define M_16   42                 /* 4 bit optional mask starting at 16 */
!   { 4, 16, S390_OPERAND_OPTIONAL }
  };
  
  
--- 133,142 ----
  #define U32_16 41		  /* 32 bit unsigned value starting at 16 */
    { 32, 16, 0 },
  #define M_16   42                 /* 4 bit optional mask starting at 16 */
!   { 4, 16, S390_OPERAND_OPTIONAL },
! #define RO_28  43                 /* optional GPR starting at position 28 */
!   { 4, 28, (S390_OPERAND_GPR | S390_OPERAND_OPTIONAL) }
! 
  };
  
  
*************** const struct s390_operand s390_operands[
*** 202,207 ****
--- 205,213 ----
  #define INSTR_RRE_RA     4, { R_24,A_28,0,0,0,0 }              /* e.g. ear   */
  #define INSTR_RRE_RF     4, { R_24,F_28,0,0,0,0 }              /* e.g. cefbr */
  #define INSTR_RRE_RR     4, { R_24,R_28,0,0,0,0 }              /* e.g. lura  */
+ /* Actually efpc and sfpc do not take an optional operand.
+    This is just a workaround for existing code e.g. glibc.  */
+ #define INSTR_RRE_RR_OPT 4, { R_24,RO_28,0,0,0,0 }             /* efpc, sfpc */
  #define INSTR_RRF_F0FF   4, { F_16,F_24,F_28,0,0,0 }           /* e.g. madbr */
  #define INSTR_RRF_FUFF   4, { F_24,F_16,F_28,U4_20,0,0 }       /* e.g. didbr */
  #define INSTR_RRF_RURR   4, { R_24,R_28,R_16,U4_20,0,0 }       /* e.g. .insn */
*************** const struct s390_operand s390_operands[
*** 275,280 ****
--- 281,287 ----
  #define MASK_RRE_RA      { 0xff, 0xff, 0xff, 0x00, 0x00, 0x00 }
  #define MASK_RRE_RF      { 0xff, 0xff, 0xff, 0x00, 0x00, 0x00 }
  #define MASK_RRE_RR      { 0xff, 0xff, 0xff, 0x00, 0x00, 0x00 }
+ #define MASK_RRE_RR_OPT  { 0xff, 0xff, 0xff, 0x00, 0x00, 0x00 }
  #define MASK_RRF_F0FF    { 0xff, 0xff, 0x0f, 0x00, 0x00, 0x00 }
  #define MASK_RRF_FUFF    { 0xff, 0xff, 0x00, 0x00, 0x00, 0x00 }
  #define MASK_RRF_RURR    { 0xff, 0xff, 0x00, 0x00, 0x00, 0x00 }

Re: PATCH: S/390: Fix instruction type of efpc and sfpc - take 2

[Martin Schwidefsky]

On Fri, 2007-02-16 at 15:40 +0100, Andreas Krebbel wrote:
> So someone indeed added a pointless dummy parameter to glibc inline assembly
> instructions instead of fixing a typo in binutils - mmmh.
> 
> As you've proposed offline, I've made the extra operand optional so that
> now at least the correct versions of the instructions are accepted as well.
> objdump still prints efpc and sfpc with the dummy operand.  That's a bit
> annoying but at least backward compatible.
> 
> OK?

objdump ignores the S390_OPERAND_OPTIONAL flag and prints the full
instruction format. We could think about skipping optional operands in
the objdump output if the value of the operand is 0.

Anyway, I've committed the patch.

-- 
blue skies,              IBM Deutschland Entwicklung GmbH
   Martin                Vorsitzender des Aufsichtsrats: Johann Weihen
                         Geschäftsführung: Herbert Kircher
Martin Schwidefsky       Sitz der Gesellschaft: Böblingen
Linux on zSeries         Registergericht: Amtsgericht Stuttgart,
   Development           HRB 243294

"Reality continues to ruin my life." - Calvin.