Re: Combine or peephole?
On Mon, Apr 19, 2010 at 5:54 PM, Jeff Law l...@redhat.com wrote: combine requires a data dependency, so for this situation, combine isn't going to help. The easy solution is to create a peephole. You can also create a machine dependent reorg pass to detect more of these opportunities. Jeff Hi Jeff, et al, Thank you for your reply. Two more questions: 1. Is it possible to add a machine dependent reorg pass at backend level without changing the standard infrastructure? If so, can you please point me such example? If no, may the new plugin architecture help here? 2. A peephole for such case just repeats instruction definition pattern. As all information already available for such peephole, wouldn’t it be useful to implement the pass to be a part of the standard infrastructure? Thank you, Frank
Re: Combine or peephole?
Hi Ian, On Wed, Apr 21, 2010 at 5:42 PM, Ian Lance Taylor i...@google.com wrote: Frank Isamov frank.isa...@gmail.com writes: 2. A peephole for such case just repeats instruction definition pattern. As all information already available for such peephole, wouldn’t it be useful to implement the pass to be a part of the standard infrastructure? See define_peephole and define_peephole2. If that doesn't answer your question, can you rephrase it? Ian I think I understood the points from Jeff’s reply and I am going to look at PA implementation now. Just for this email thread completeness, I’ll try to rephrase the initial question: Instructions which manipulate with data in parallel and have no data dependency automatically require peephole2 definition or/and machine dependent reorg pass. (Please see an example at the bottom of this email). Peephole2 pattern, in this case, just repeats instruction’s RTL pattern. As such instructions can appear in SIMD architectures, I just thought that it would be profitable to have this pass to be a part of the common infrastructure. (define_insn assi6 [(parallel [ (set (match_operand:SI 0 register_operand =r) (minus:SI (match_operand:SI 1 register_operand r) (match_operand:SI 2 register_operand r))) (set (match_operand:SI 3 register_operand =r) (plus:SI (match_operand:SI 4 register_operand r) (match_operand:SI 5 register_operand r))) ])] as\t%5, %4, %3, %2, %1, %0 %! ) (define_peephole2 [ (set (match_operand:SI 0 register_operand) (minus:SI (match_operand:SI 1 register_operand) (match_operand:SI 2 register_operand))) (set (match_operand:SI 3 register_operand) (plus:SI (match_operand:SI 4 register_operand) (match_operand:SI 5 register_operand))) ] [(parallel [ (set (match_dup 0) (minus:SI (match_dup 1) (match_dup 2))) (set (match_dup 3) (plus:SI (match_dup 4) (match_dup 5))) ])] )
Combine or peephole?
Hi, My architecture supports instructions with two parallel side effects. For example, addition and subtraction can be done in parallel: (define_insn assi6 [(parallel [ (set (match_operand:SI 0 register_operand =r) (minus:SI (match_operand:SI 1 register_operand r) (match_operand:SI 2 register_operand r))) (set (match_operand:SI 3 register_operand =r) (plus:SI (match_operand:SI 4 register_operand r) (match_operand:SI 5 register_operand r))) ])] as\t%5, %4, %3, %2, %1, %0 %! ) This instruction is not chosen at ‘combine’ time even if ‘plus’ and ‘minus’ instructions are located one after other. That it is, probably, there is no data dependency between them. In attempt to resolve the problem, I am providing a peephole optimization: (define_peephole2 [ (set (match_operand:SI 0 register_operand) (minus:SI (match_operand:SI 1 register_operand) (match_operand:SI 2 register_operand))) (set (match_operand:SI 3 register_operand) (plus:SI (match_operand:SI 4 register_operand) (match_operand:SI 5 register_operand))) ] [(parallel [ (set (match_dup 0) (minus:SI (match_dup 1) (match_dup 2))) (set (match_dup 3) (plus:SI (match_dup 4) (match_dup 5))) ])] ) This works for some cases, but I wanted to ask experts whether this is the way to go. Repeating the same pattern for a peephole might be not the best way to resolve the problem. Did you observe something similar? What would be the best way to resolve such situation? Thank you, Frank
Coloring problem - Pass 0 for finding allocno costs
Hi,
In my backend, I have a problem with the pass which determines the
best register class for a virtual register (Pass 0 for finding allocno
costs).
In all insns in this example both R_REGS and D_REGS register classes
are applicable (but all registers in an insn should be from the same
register class).
This is asmcons output:
;; Function mul (mul)
(note 1 0 5 NOTE_INSN_DELETED)
(note 5 1 2 2 [bb 2] NOTE_INSN_BASIC_BLOCK)
(note 2 5 3 2 NOTE_INSN_DELETED)
(insn 3 2 4 2 a.c:2 (set (reg/v:SI 97 [ b ])
(reg:SI 49 r1 [ b ])) 43 {movsi_regs} (expr_list:REG_DEAD
(reg:SI 49 r1 [ b ])
(nil)))
(note 4 3 7 2 NOTE_INSN_FUNCTION_BEG)
(insn 7 4 9 2 a.c:5 (set (reg/v:SI 94 [ __a_11 ])
(plus:SI (reg:SI 48 r0 [ a ])
(reg/v:SI 97 [ b ]))) 0 {*addsi3_1} (expr_list:REG_DEAD
(reg:SI 48 r0 [ a ])
(nil)))
(insn 9 7 10 2 a.c:5 (parallel [
(set (reg:SI 100)
(ashift:SI (reg/v:SI 97 [ b ])
(const_int 3 [0x3])))
(clobber (reg:CC 88 cc))
]) 36 {ashlsi3} (expr_list:REG_UNUSED (reg:CC 88 cc)
(expr_list:REG_EQUAL (ashift:SI (reg/v:SI 97 [ b ])
(const_int 3 [0x3]))
(nil
(insn 10 9 11 2 a.c:5 (set (reg:SI 101)
(plus:SI (reg:SI 100)
(reg/v:SI 97 [ b ]))) 0 {*addsi3_1} (expr_list:REG_DEAD (reg:SI 100)
(expr_list:REG_DEAD (reg/v:SI 97 [ b ])
(expr_list:REG_EQUAL (mult:SI (reg/v:SI 97 [ b ])
(const_int 9 [0x9]))
(nil)
(insn 11 10 12 2 a.c:5 (set (reg:SI 102)
(plus:SI (reg:SI 101)
(reg/v:SI 94 [ __a_11 ]))) 0 {*addsi3_1}
(expr_list:REG_DEAD (reg:SI 101)
(expr_list:REG_DEAD (reg/v:SI 94 [ __a_11 ])
(nil
(note 12 11 17 2 NOTE_INSN_DELETED)
(insn 17 12 23 2 a.c:7 (parallel [
(set (reg/i:SI 48 r0)
(ashift:SI (reg:SI 102)
(const_int 10 [0xa])))
(clobber (reg:CC 88 cc))
]) 36 {ashlsi3} (expr_list:REG_UNUSED (reg:CC 88 cc)
(expr_list:REG_DEAD (reg:SI 102)
(nil
(insn 23 17 0 2 a.c:7 (use (reg/i:SI 48 r0)) -1 (nil))
The problem I see is that for registers 100,101 I get best register
class D instead of R – actually they get the same cost and D is chosen
(maybe because it is first).
But they should not get the same cost since choosing D_REGS causes two
additional copies.
To my understanding the algorithm checks every insn, and since
register 100 appears only in insns in which all registers are still
virtual, any register class fits without additional cost. But later
when coloring it with d register, we need copies from r to d and back
to r.
Can someone please help with this?
Is there any reading material about this part of the IRA?
Thanks, Frank.
Re: Coloring problem - Pass 0 for finding allocno costs
-- Forwarded message --
From: Frank Isamov frank.isa...@gmail.com
Date: Thu, Mar 18, 2010 at 4:28 PM
Subject: Re: Coloring problem - Pass 0 for finding allocno costs
To: Ian Bolton bol...@icerasemi.com
On Thu, Mar 18, 2010 at 3:51 PM, Ian Bolton bol...@icerasemi.com wrote:
The problem I see is that for registers 100,101 I get best register
class D instead of R - actually they get the same cost and D is chosen
(maybe because it is first).
Hi Frank.
Do D and R overlap? It would be useful to know which regs are in
which class, before trying to understand what is going on.
Can you paste an example of your define_insn from your MD file to show
how operands from D or R are both valid? I ask this because it is
possible to express that D is more expensive than R with operand
constraints.
For general IRA info, you might like to look over my long thread on
here called Understanding IRA.
Cheers,
Ian
Hi Ian,
Thank you very much for your prompt reply.
D and R are not overlap. Please see fragments of .md and .h files below:
From the md:
(define_register_constraint a R_REGS )
(define_register_constraint d D_REGS )
(define_predicate a_operand
(match_operand 0 register_operand)
{
unsigned int regno;
if (GET_CODE (op) == SUBREG)
op = SUBREG_REG (op);
regno = REGNO (op);
return (regno = FIRST_PSEUDO_REGISTER ||
REGNO_REG_CLASS(regno) == R_REGS);
}
)
(define_predicate d_operand
(match_operand 0 register_operand)
{
unsigned int regno;
if (GET_CODE (op) == SUBREG)
op = SUBREG_REG (op);
regno = REGNO (op);
return (regno = FIRST_PSEUDO_REGISTER ||
REGNO_REG_CLASS(regno) == D_REGS);
}
)
(define_predicate a_d_operand
(ior (match_operand 0 a_operand)
(match_operand 0 d_operand)))
(define_predicate i_a_d_operand
(ior (match_operand 0 immediate_operand)
(match_operand 0 a_d_operand)))
(define_insn movmode_regs
[(set (match_operand:SISFM 0 a_d_operand =a, a, a, d, d, d)
(match_operand:SISFM 1 i_a_d_operand a, i, d, a, i, d))]
move\t%1, %0
)
(define_insn *addsi3_1
[(set (match_operand:SI 0 a_d_operand =a, a, a,d,d)
(plus:SI (match_operand:SI 1 a_d_operand %a, a, a,d,d)
(match_operand:SI 2 nonmemory_operand
U05,S16,a,d,U05)))]
adda\t%2, %1, %0
)
;; Arithmetic Left and Right Shift Instructions
(define_insn shPatmode3
[(set (match_operand:SCIM 0 register_operand =a,d,d)
(sh_oprnd:SCIM (match_operand:SCIM 1 register_operand a,d,d)
(match_operand:SI 2
nonmemory_operand U05,d,U05)))
(clobber (reg:CC CC_REGNUM))]
shIsa\t%2, %1, %0
)
From the h file:
#define REG_CLASS_CONTENTS \
{
\
{0x, 0x, 0x}, /* NO_REGS*/ \
{0x, 0x, 0x}, /* D_REGS*/ \
{0x, 0x, 0x}, /* R_REGS*/ \
ABI requires use of R registers for arguments and return value. Other
than that all of these instructions are more or less symmetrical in
sense of using D or R. So, an optimal choice would be use of R for
this example. And if D register is chosen, it involves additional copy
from R to D and back to R.
Thank you, Frank
Advancing SP on a call
We have a problem with arguments passing in memory. The caller puts the arguments in memory relative to the sp: add sp, 4 // allocate space for the argument. stack grows up store r1, (sp-4) // store the argument on the stack call xxx// call the function. In xxx the result code looks like: load (sp-4), r1 // load the argument from the stack. The problem is that the 'call' instruction pushes the return address to the stack and increments the sp by 4 so when the callee tries to access the memory it does not get to the correct location. How can I tell GCC that that the callee should load from the original offset + 4? Thanks.
Re: How to make 'long int' type be a PDImode?
On Mon, Mar 8, 2010 at 8:29 AM, Joern Rennecke joern.renne...@embecosm.com wrote: Quoting Frank Isamov frank.isa...@gmail.com: Hi, I'd like to make a backend which would have 48 bits for 'long' type. (32 for int and 64 for long long). I have tried to define: #define LONG_TYPE_SIZE 48 That's not a partial integer mode; PDImode would have the same size as DImode, just not all bits would be significant. and one of: INT_MODE (PDI, 6); And that wouldn't be PDImode, more like THImode (three-halves integer mode, going by the precedent of TQFmode - three-quarter float mode - of the 1750a port in GCC prior to version 3.1) I am sorry, I still can conclude how to make the backend use a certain mode for 'long' type. My architecture implements 32- and 48- bit registers and instructions operating on these registers. That it why my intention is to use 'int' for 32 and 'long' for 48 bit operations. My attempts lead to the backend ignore 48 bits path inspite of the LONG_TYPE_SIZE definition and 'long' is still processed as 32 bit value. I think I am missing something, so I am applying for help. Any piece of information would be useful. Thank you, Frank
Re: How to make 'long int' type be a PDImode?
On Mon, Mar 8, 2010 at 4:27 PM, Frank Isamov frank.isa...@gmail.com wrote: On Mon, Mar 8, 2010 at 8:29 AM, Joern Rennecke joern.renne...@embecosm.com wrote: Quoting Frank Isamov frank.isa...@gmail.com: Hi, I'd like to make a backend which would have 48 bits for 'long' type. (32 for int and 64 for long long). I have tried to define: #define LONG_TYPE_SIZE 48 That's not a partial integer mode; PDImode would have the same size as DImode, just not all bits would be significant. and one of: INT_MODE (PDI, 6); And that wouldn't be PDImode, more like THImode (three-halves integer mode, going by the precedent of TQFmode - three-quarter float mode - of the 1750a port in GCC prior to version 3.1) I am sorry, I still can conclude how to make the backend use a certain mode for 'long' type. My architecture implements 32- and 48- bit registers and instructions operating on these registers. That it why my intention is to use 'int' for 32 and 'long' for 48 bit operations. My attempts lead to the backend ignore 48 bits path inspite of the LONG_TYPE_SIZE definition and 'long' is still processed as 32 bit value. I think I am missing something, so I am applying for help. Any piece of information would be useful. Thank you, Frank Correction: I still can conclude should be read as I still cannot conclude
How to make 'long int' type be a PDImode?
Hi, I'd like to make a backend which would have 48 bits for 'long' type. (32 for int and 64 for long long). I have tried to define: #define LONG_TYPE_SIZE 48 and one of: INT_MODE (PDI, 6); PARTIAL_INT_MODE (DI); Unfortunately, trying to compile a program, I see that the backend still uses SImode for 'long'. I could not find an example for target doing similar work. Could you please advise or show the location of implementation I can refer? Thank you, Frank
