[Bug middle-end/70159] missed CSE optimization

[anton at samba dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #18 from Anton Blanchard  ---
Urgh too early in the morning for me. PR71866 created, with the correct
backtrace.

[Bug middle-end/70159] missed CSE optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #17 from Andrew Pinski  ---
(In reply to Anton Blanchard from comment #16)
> I'm seeing a lockup in gcc with this patch on ppc64le. Run as:
> 
> gcc -O2 -c testcase.i

Can you file a new bug for this?  Also your backtrace is just for the driver.

[Bug middle-end/70159] missed CSE optimization

[anton at samba dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #16 from Anton Blanchard  ---
I'm seeing a lockup in gcc with this patch on ppc64le. Run as:

gcc -O2 -c testcase.i

It gets stuck in:

#0  0x3fffb7e5e3e8 in __waitpid_nocancel ()
at ../sysdeps/unix/syscall-template.S:84
#1  0x10088c58 in pex_wait (obj=, time=0x0, 
status=0x1015f080, pid=)
at ../../gcc/libiberty/pex-unix.c:134
#2  pex_unix_wait (obj=, pid=, 
status=0x1015f080, time=0x0, done=, errmsg=0x3fffd750, 
err=0x3fffd74c) at ../../gcc/libiberty/pex-unix.c:738
#3  0x1008762c in pex_get_status_and_time (obj=, 
done=, errmsg=, err=)
at ../../gcc/libiberty/pex-common.c:534
#4  0x10088630 in pex_get_status (obj=0x1015efc0, 
count=, vector=0x3fffd7e0)
at ../../gcc/libiberty/pex-common.c:554
#5  0x100101dc in execute () at ../../gcc/gcc/gcc.c:3107
#6  0x10010f04 in do_spec_1 (
spec=0x1015ec20 "-o %|.s |\n as %(asm_options) %m.s %A", inswitch=0, 
soft_matched_part=0x0) at ../../gcc/gcc/gcc.c:5145
#7  0x100139e4 in process_brace_body (matched=, 
starred=0, end_atom=0x1015ed3c ":-o %|.s |\n as %(asm_options) %m.s %A }", 
atom=0x1015ed3b "S:-o %|.s |\n as %(asm_options) %m.s %A }", 
p=0x1015ed62 "}") at ../../gcc/gcc/gcc.c:6431
#8  handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#9  0x10011604 in do_spec_1 (
spec=0x1015ecf0 "  
%{fcompare-debug=*|fdump-final-insns=*:%:compare-debug-dump-opt()}   %{!S:-o
%|.s |\n as %(asm_options) %m.s %A }", inswitch=0, 
soft_matched_part=0x0) at ../../gcc/gcc/gcc.c:5802
#10 0x100139e4 in process_brace_body (matched=, 
starred=1, 
end_atom=0x100a877f "*:  
%{fcompare-debug=*|fdump-final-insns=*:%:compare-debug-dump-opt()}   %{!S:-o
%|.s |\n as %(asm_options) %m.s %A }  }", 
atom=0x100a877b "fwpa*:  
%{fcompare-debug=*|fdump-final-insns=*:%:compare-debug-dump-opt()}   %{!S:-o
%|.s |\n as %(asm_options) %m.s %A }  }", 
p=0x100a87f6 "}") at ../../gcc/gcc/gcc.c:6431
#11 handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#12 0x10011604 in do_spec_1 (
spec=0x100a8778 "%{!fwpa*:  
%{fcompare-debug=*|fdump-final-insns=*:%:compare-debug-dump-opt()}   %{!S:-o
%|.s |\n as %(asm_options) %m.s %A }  }", 
inswitch=0, soft_matched_part=0x0) at ../../gcc/gcc/gcc.c:5802
#13 0x10011228 in do_spec_1 (spec=0x1015ec90 "%(invoke_as)", 
inswitch=0, soft_matched_part=0x0) at ../../gcc/gcc/gcc.c:5917
#14 0x100139e4 in process_brace_body (matched=, 
starred=0, end_atom=0x1015ec04 ":%(invoke_as)}", 
atom=0x1015ebf8 "fsyntax-only:%(invoke_as)}", p=0x1015ec11 "}")
at ../../gcc/gcc/gcc.c:6431
#15 handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#16 0x10011604 in do_spec_1 (
spec=0x1015ebd0 "cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}", inswitch=0, soft_matched_part=0x0) at
../../gcc/gcc/gcc.c:5802
#17 0x100139e4 in process_brace_body (matched=, 
starred=0, 
end_atom=0x1015eb74 ":cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}", 
atom=0x1015eb73 "E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}", p=0x1015ebb7 "}") at ../../gcc/gcc/gcc.c:6431
#18 handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#19 0x10011604 in do_spec_1 (
spec=0x1015eb70 "%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}", inswitch=0, soft_matched_part=0x0) at
../../gcc/gcc/gcc.c:5802
#20 0x100139e4 in process_brace_body (matched=, 
starred=0, 
end_atom=0x1015eb15 ":%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}}", 
atom=0x1015eb13 "MM:%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}}", p=0x1015eb5e "}") at ../../gcc/gcc/gcc.c:6431
#21 handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#22 0x10011604 in do_spec_1 (
spec=0x1015eb10 "%{!MM:%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)}}}", inswitch=0, soft_matched_part=0x0)
at ../../gcc/gcc/gcc.c:5802
#23 0x100139e4 in process_brace_body (matched=, 
starred=0, 
end_atom=0x100a789c ":%{!MM:%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)", 
atom=0x100a789b "M:%{!MM:%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)", p=0x100a78ec "}") at
../../gcc/gcc/gcc.c:6431
#24 handle_braces (p=) at ../../gcc/gcc/gcc.c:6345
#25 0x10011604 in do_spec_1 (
spec=0x100a7898 "%{!M:%{!MM:%{!E:cc1 -fpreprocessed %i %(cc1_options)
%{!fsyntax-only:%(invoke_as)", inswitch=0, soft_matched_part=0x0)
at ../../gcc/gcc/gcc.c:5802
#26 0x10012878 in do_spec_2 (spec=)
at ../../gcc/gcc/gcc.c:4841
#27 0x10014514 in do_spec (spec=)
at ../../gcc/gcc/gcc.c:4808
#28 0x1001479c in driver::do_spec_on_infiles (this=0x3108)
at ../../gcc/gcc/gcc.c:8076
#29 0x10003138 in driver::main 

[Bug middle-end/70159] missed CSE optimization

[anton at samba dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

Anton Blanchard  changed:

   What|Removed |Added

 CC||anton at samba dot org

--- Comment #15 from Anton Blanchard  ---
Created attachment 38890
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38890=edit
test case

[Bug middle-end/70159] missed CSE optimization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

Richard Biener  changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |FIXED
   Target Milestone|--- |7.0

--- Comment #14 from Richard Biener  ---
Fixed.

[Bug middle-end/70159] missed CSE optimization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159
Bug 70159 depends on bug 23286, which changed state.

Bug 23286 Summary: Missed code hoisting optimization
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=23286

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |FIXED

[Bug middle-end/70159] missed CSE optimization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #13 from Richard Biener  ---
Author: rguenth
Date: Tue Jul 12 13:32:04 2016
New Revision: 238242

URL: https://gcc.gnu.org/viewcvs?rev=238242=gcc=rev
Log:
2016-07-12  Steven Bosscher  
Richard Biener  

PR tree-optimization/23286
PR tree-optimization/70159
* doc/invoke.texi: Document -fcode-hoisting.
* common.opt (fcode-hoisting): New flag.
* opts.c (default_options_table): Enable -fcode-hoisting at -O2+.
* tree-ssa-pre.c (pre_stats): Add hoist_insert.
(do_regular_insertion): Rename to ...
(do_pre_regular_insertion): ... this and amend general comments
on insertion strathegy.
(do_partial_partial_insertion): Rename to ...
(do_pre_partial_partial_insertion): ... this.
(do_hoist_insertion): New function.
(insert_aux): Take flags on whether to do PRE and/or hoist insertion
and call do_hoist_insertion properly.
(insert): Adjust.
(pass_pre::gate): Enable also if -fcode-hoisting is enabled.
(pass_pre::execute): Register hoist_insert stats.

* gcc.dg/tree-ssa/ssa-pre-11.c: Disable code hosting.
* gcc.dg/tree-ssa/ssa-pre-27.c: Likewise.
* gcc.dg/tree-ssa/ssa-pre-28.c: Likewise.
* gcc.dg/tree-ssa/ssa-pre-2.c: Likewise.
* gcc.dg/tree-ssa/pr35286.c: Likewise.
* gcc.dg/tree-ssa/pr35287.c: Likewise.
* gcc.dg/hoist-register-pressure-1.c: Likewise.
* gcc.dg/hoist-register-pressure-2.c: Likewise.
* gcc.dg/hoist-register-pressure-3.c: Likewise.
* gcc.dg/pr51879-12.c: Likewise.
* gcc.dg/strlenopt-9.c: Likewise.
* gcc.dg/tree-ssa/pr47392.c: Likewise.
* gcc.dg/tree-ssa/pr68619-4.c: Likewise.
* gcc.dg/tree-ssa/split-path-5.c: Likewise.
* gcc.dg/tree-ssa/slsr-35.c: Likewise.
* gcc.dg/tree-ssa/slsr-36.c: Likewise.
* gcc.dg/tree-ssa/loadpre3.c: Adjust so hosting doesn't apply.
* gcc.dg/tree-ssa/pr43491.c: Scan optimized dump for desired result.
* gcc.dg/tree-ssa/ssa-pre-31.c: Adjust expected outcome for hoisting.
* gcc.dg/tree-ssa/ssa-hoist-1.c: New testcase.
* gcc.dg/tree-ssa/ssa-hoist-2.c: New testcase.
* gcc.dg/tree-ssa/ssa-hoist-3.c: New testcase.
* gcc.dg/tree-ssa/ssa-hoist-4.c: New testcase.
* gcc.dg/tree-ssa/ssa-hoist-5.c: New testcase.
* gcc.dg/tree-ssa/ssa-hoist-6.c: New testcase.
* gfortran.dg/pr43984.f90: Adjust expected outcome.

Added:
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-hoist-1.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-hoist-2.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-hoist-3.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-hoist-4.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-hoist-6.c
Modified:
trunk/gcc/ChangeLog
trunk/gcc/common.opt
trunk/gcc/doc/invoke.texi
trunk/gcc/opts.c
trunk/gcc/testsuite/ChangeLog
trunk/gcc/testsuite/gcc.dg/hoist-register-pressure-1.c
trunk/gcc/testsuite/gcc.dg/hoist-register-pressure-2.c
trunk/gcc/testsuite/gcc.dg/hoist-register-pressure-3.c
trunk/gcc/testsuite/gcc.dg/pr51879-12.c
trunk/gcc/testsuite/gcc.dg/strlenopt-9.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/loadpre3.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/pr35286.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/pr35287.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/pr43491.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/pr47392.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/pr68619-4.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/slsr-35.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/slsr-36.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-pre-11.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-pre-2.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-pre-27.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-pre-28.c
trunk/gcc/testsuite/gcc.dg/tree-ssa/ssa-pre-31.c
trunk/gcc/testsuite/gfortran.dg/pr43984.f90
trunk/gcc/tree-ssa-pre.c

[Bug middle-end/70159] missed CSE optimization

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #12 from rguenther at suse dot de  ---
On Sat, 2 Jul 2016, hiraditya at msn dot com wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159
> 
> AK  changed:
> 
>What|Removed |Added
> 
>  CC||hiraditya at msn dot com
> 
> --- Comment #11 from AK  ---
> Just as an update, the new gvn-hoist pass in llvm hoists the common
> computations:
> 
> @cat test.c
> float foo_p(float d, float min, float max, float a)
> {
>   float tmin;
>   float tmax;
> 
>   float inv = 1.0f / d;
>   if (inv >= 0) {
> tmin = (min - a) * inv;
> tmax = (max - a) * inv;
>   } else {
> tmin = (max - a) * inv;
> tmax = (min - a) * inv;
>   }
> 
>   return tmax + tmin;
> }
> 
> 
> clang -c -Ofast test.c -mllvm -print-after-all
> 
> 
> *** IR Dump Before Early GVN Hoisting of Expressions ***
> ; Function Attrs: nounwind uwtable
> define float @_Z5foo_p(float %d, float %min, float %max, float %a) #0 {
> entry:
>   %div = fdiv fast float 1.00e+00, %d
>   %cmp = fcmp fast oge float %div, 0.00e+00
>   br i1 %cmp, label %if.then, label %if.else
> 
> if.then:  ; preds = %entry
>   %sub = fsub fast float %min, %a
>   %mul = fmul fast float %sub, %div
>   %sub1 = fsub fast float %max, %a
>   %mul2 = fmul fast float %sub1, %div
>   br label %if.end
> 
> if.else:  ; preds = %entry
>   %sub3 = fsub fast float %max, %a
>   %mul4 = fmul fast float %sub3, %div
>   %sub5 = fsub fast float %min, %a
>   %mul6 = fmul fast float %sub5, %div
>   br label %if.end
> 
> if.end:   ; preds = %if.else, %if.then
>   %tmax.0 = phi float [ %mul2, %if.then ], [ %mul6, %if.else ]
>   %tmin.0 = phi float [ %mul, %if.then ], [ %mul4, %if.else ]
>   %add = fadd fast float %tmax.0, %tmin.0
>   ret float %add
> }
> 
> 
> *** IR Dump After Early GVN Hoisting of Expressions ***
> ; Function Attrs: nounwind uwtable
> define float @_Z5foo_p(float %d, float %min, float %max, float %a) #0 {
> entry:
>   %div = fdiv fast float 1.00e+00, %d
>   %cmp = fcmp fast oge float %div, 0.00e+00
>   %sub = fsub fast float %min, %a
>   %mul = fmul fast float %sub, %div
>   %sub1 = fsub fast float %max, %a
>   %mul2 = fmul fast float %sub1, %div
>   br i1 %cmp, label %if.then, label %if.else
> 
> if.then:  ; preds = %entry
>   br label %if.end
> 
> if.else:  ; preds = %entry
>   br label %if.end
> 
> if.end:   ; preds = %if.else, %if.then
>   %tmax.0 = phi float [ %mul2, %if.then ], [ %mul, %if.else ]
>   %tmin.0 = phi float [ %mul, %if.then ], [ %mul2, %if.else ]
>   %add = fadd fast float %tmax.0, %tmin.0
>   ret float %add
> }

Same if you forward-port the patch from PR23286 (just did that).

> cat t.c.126t.pre
...
foo_p (float d, float min, float max, float a)
{
  float inv;
  float tmax;
  float tmin;
  float _16;
  float _18;
  float _19;
  float _20;
  float _21;

  :
  inv_8 = 1.0e+0 / d_7(D);
  _18 = min_9(D) - a_10(D);
  _19 = inv_8 * _18;
  _20 = max_12(D) - a_10(D);
  _21 = inv_8 * _20;
  if (inv_8 >= 0.0)
goto ;
  else
goto ;

  :

  :
  # tmin_5 = PHI <_19(2), _21(3)>
  # tmax_6 = PHI <_21(2), _19(3)>
  _16 = tmin_5 + tmax_6;
  return _16;

[Bug middle-end/70159] missed CSE optimization

[hiraditya at msn dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

AK  changed:

   What|Removed |Added

 CC||hiraditya at msn dot com

--- Comment #11 from AK  ---
Just as an update, the new gvn-hoist pass in llvm hoists the common
computations:

@cat test.c
float foo_p(float d, float min, float max, float a)
{
  float tmin;
  float tmax;

  float inv = 1.0f / d;
  if (inv >= 0) {
tmin = (min - a) * inv;
tmax = (max - a) * inv;
  } else {
tmin = (max - a) * inv;
tmax = (min - a) * inv;
  }

  return tmax + tmin;
}


clang -c -Ofast test.c -mllvm -print-after-all


*** IR Dump Before Early GVN Hoisting of Expressions ***
; Function Attrs: nounwind uwtable
define float @_Z5foo_p(float %d, float %min, float %max, float %a) #0 {
entry:
  %div = fdiv fast float 1.00e+00, %d
  %cmp = fcmp fast oge float %div, 0.00e+00
  br i1 %cmp, label %if.then, label %if.else

if.then:  ; preds = %entry
  %sub = fsub fast float %min, %a
  %mul = fmul fast float %sub, %div
  %sub1 = fsub fast float %max, %a
  %mul2 = fmul fast float %sub1, %div
  br label %if.end

if.else:  ; preds = %entry
  %sub3 = fsub fast float %max, %a
  %mul4 = fmul fast float %sub3, %div
  %sub5 = fsub fast float %min, %a
  %mul6 = fmul fast float %sub5, %div
  br label %if.end

if.end:   ; preds = %if.else, %if.then
  %tmax.0 = phi float [ %mul2, %if.then ], [ %mul6, %if.else ]
  %tmin.0 = phi float [ %mul, %if.then ], [ %mul4, %if.else ]
  %add = fadd fast float %tmax.0, %tmin.0
  ret float %add
}


*** IR Dump After Early GVN Hoisting of Expressions ***
; Function Attrs: nounwind uwtable
define float @_Z5foo_p(float %d, float %min, float %max, float %a) #0 {
entry:
  %div = fdiv fast float 1.00e+00, %d
  %cmp = fcmp fast oge float %div, 0.00e+00
  %sub = fsub fast float %min, %a
  %mul = fmul fast float %sub, %div
  %sub1 = fsub fast float %max, %a
  %mul2 = fmul fast float %sub1, %div
  br i1 %cmp, label %if.then, label %if.else

if.then:  ; preds = %entry
  br label %if.end

if.else:  ; preds = %entry
  br label %if.end

if.end:   ; preds = %if.else, %if.then
  %tmax.0 = phi float [ %mul2, %if.then ], [ %mul, %if.else ]
  %tmin.0 = phi float [ %mul, %if.then ], [ %mul2, %if.else ]
  %add = fadd fast float %tmax.0, %tmin.0
  ret float %add
}

[Bug middle-end/70159] missed CSE optimization

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #10 from rguenther at suse dot de  ---
On Thu, 10 Mar 2016, spop at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159
> 
> --- Comment #9 from Sebastian Pop  ---
> Created attachment 37927
>   --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37927=edit
> patch for hoisting expressions
> 
> Updated the patch from PR23286 to hoist the redundant expressions:
> 
>   :
>   inv_4 = 1.0e+0 / d_3(D);
>   _18 = min_5(D) - a_6(D);
>   _19 = _18 / inv_4;
>   _20 = max_9(D) - a_6(D);
>   _21 = _20 / inv_4;
>   if (inv_4 >= 0.0)
> goto ;
>   else
> goto ;
> 
>   :
> 
>   :
>   # tmin_1 = PHI <_19(2), _21(3)>
>   # tmax_2 = PHI <_21(2), _19(3)>
>   _16 = tmin_1 + tmax_2;
>   return _16;
> 
> The attached patch does not pass make check and causes some infinite 
> recursion.

Yes, the patch has some issues still.  As for every stage1 I plan
to try picking it up again ...

I notice from the above output that it doesn't see the CSE
opportunity but inserts both expressions - it shouldn't do that,
so there's something "new wrong" with the patch (or your update
to it)

[Bug middle-end/70159] missed CSE optimization

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #9 from Sebastian Pop  ---
Created attachment 37927
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37927=edit
patch for hoisting expressions

Updated the patch from PR23286 to hoist the redundant expressions:

  :
  inv_4 = 1.0e+0 / d_3(D);
  _18 = min_5(D) - a_6(D);
  _19 = _18 / inv_4;
  _20 = max_9(D) - a_6(D);
  _21 = _20 / inv_4;
  if (inv_4 >= 0.0)
goto ;
  else
goto ;

  :

  :
  # tmin_1 = PHI <_19(2), _21(3)>
  # tmax_2 = PHI <_21(2), _19(3)>
  _16 = tmin_1 + tmax_2;
  return _16;

The attached patch does not pass make check and causes some infinite recursion.

[Bug middle-end/70159] missed CSE optimization

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

Richard Biener  changed:

   What|Removed |Added

   Keywords||missed-optimization
 Status|UNCONFIRMED |NEW
   Last reconfirmed||2016-03-10
 CC||rguenth at gcc dot gnu.org
Version|unknown |6.0
 Ever confirmed|0   |1

--- Comment #8 from Richard Biener  ---
We already compute nearly everything that is necessary:

Value numbering tmin_13 stmt = tmin_13 = inv_4 * _12;
Match-and-simplified inv_4 * _12 to tmax_11
RHS inv_4 * _12 simplified to tmax_11
Setting value number of tmin_13 to tmax_11 (changed)
...
Value numbering tmax_15 stmt = tmax_15 = inv_4 * _14;
Match-and-simplified inv_4 * _14 to tmin_8
RHS inv_4 * _14 simplified to tmin_8
Setting value number of tmax_15 to tmin_8 (changed)

so we know the equivalency but can't do sth sensible with it yet.

  :
  # tmin_1 = PHI 
  # tmax_2 = PHI 
  _16 = tmin_1 + tmax_2;

so value-wise this is

  :
  # tmin_1 = PHI 
  # tmax_2 = PHI 
  _16 = tmin_1 + tmax_2;

which we could transform by pattern matching this case changing the
if (inv >= 0) to always take the path which otherwise has no side-effects.

I'm not sure it really fits into FRE/PRE elimination phase but at least
it may be easier to do sth like phi-opt on the more complex cases in the
VN framework (to avoid the need to compute equivalencies this complex).

Hoisting would surely help as well but it still would need sth to detect
the commutatively redundant swap.  That is,

float foo_p(float d, float min, float max, float a)
{
  float tmin;
  float tmax;
  float inv = 1.0f / d;
  if (inv >= 0) {
  tmin = min;
  tmax = max;
  } else {
  tmin = max;
  tmax = min;
  }
  return tmax + tmin;
}

is not optimized either and we retain

  if (inv_4 >= 0.0)
goto ;
  else
goto ;

  :

  :
  # tmin_1 = PHI 
  # tmax_2 = PHI 
  _7 = tmin_1 + tmax_2;

until .optimized.  So that part of this PR is independent of the hoisting
issue we have other PRs for.

(simplify
 (plus (cond @0 @1 @2) (cond @0 @2 @1))
 (plus @1 @2))

would fix it if we'd match PHIs for conds as well.  If we "help" PRE with

float foo_p(float d, float min, float max, float a)
{
  float tmin;
  float tmax;
  float inv = 1.0f / d;
  float bar = min + max;
  if (inv >= 0) {
  tmin = min;
  tmax = max;
  } else {
  tmin = max;
  tmax = min;
  }
  return tmax + tmin + bar;
}

it figures that tmax + tmin is equal to bar and optimizes this.  Sth
that hoisting (as implemented in PRE) should then do in one step
hopefully.  Catching this somewhat earlier would be nice though.

[Bug middle-end/70159] missed CSE optimization

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #7 from Sebastian Pop  ---
(In reply to Andrew Pinski from comment #6)
> Note this is both a hoisting and a sinking issue.
> Hoisting should happen before sinking.
> LLVM looks like it only implements sinking.

You are right: LLVM does sinking very early as part of instcombine: it
transforms the phi nodes after the if into selects over the operands and sinks
the sub and mul after the select.  By the time other redundancy elimination
passes are executed the shape of the code is more difficult to optimize.

[Bug middle-end/70159] missed CSE optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #6 from Andrew Pinski  ---
Note this is both a hoisting and a sinking issue.
Hoisting should happen before sinking.
LLVM looks like it only implements sinking.

[Bug middle-end/70159] missed CSE optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #3 from Andrew Pinski  ---
Related to 5738.

Try with -Os and you will see the RTL non PRE based GCSE does most of the job:
foo_p:
fmovs4, 1.0e+0
fdivs0, s4, s0
fsubs1, s1, s3
fsubs2, s2, s3
fcmpe   s0, #0.0
blt .L6
fmuls4, s1, s0
fmuls0, s2, s0
.L4:
fadds0, s4, s0
ret
.L6:
fmuls4, s2, s0
fmuls0, s1, s0
b   .L4

[Bug middle-end/70159] missed CSE optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #5 from Andrew Pinski  ---
Oh and bug 23286.

[Bug middle-end/70159] missed CSE optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #4 from Andrew Pinski  ---
"Related to bug 5738."

[Bug middle-end/70159] missed CSE optimization

[spop at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

--- Comment #2 from Sebastian Pop  ---
Right, with -Ofast it be able to optimize away the branch or selects.
The original benchmark had something more complex than fadd to use the tmin and
tmax results. Here is one more test using the results in a non commutative
operation:

bool foo_p(float d, float min, float max, float a)
{
  float tmin;
  float tmax;

  float inv = 1.0f / d;
  if (inv >= 0) {
tmin = (min - a) * inv;
tmax = (max - a) * inv;
  } else {
tmin = (max - a) * inv;
tmax = (min - a) * inv;
  }

  return tmax > tmin;
}

[Bug middle-end/70159] missed CSE optimization

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70159

Jakub Jelinek  changed:

   What|Removed |Added

 CC||jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek  ---
Well, as tmax + tmin is commutative, this one actually should be optimized into
just (min - a) * inv + (max - a) * inv or with -Ofast perhaps to (min - a + max
- a) * inv.