Bug 95097 - Missed optimization with bitfield value ranges
Summary: Missed optimization with bitfield value ranges
Status: NEW
Alias: None
Product: gcc
Classification: Unclassified
Component: tree-optimization (show other bugs)
Version: 11.0
: P3 enhancement
Target Milestone: ---
Assignee: Not yet assigned to anyone
URL:
Keywords: missed-optimization
Depends on:
Blocks: bitfield VRP
  Show dependency treegraph
 
Reported: 2020-05-13 03:00 UTC by Rich Felker
Modified: 2024-07-30 22:35 UTC (History)
1 user (show)

See Also:
Host:
Target:
Build:
Known to work:
Known to fail:
Last reconfirmed: 2020-05-13 00:00:00


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Rich Felker 2020-05-13 03:00:52 UTC
#include <stdint.h>
struct foo {
	uint32_t x:20;
};
int bar(struct foo f)
{
	if (f.x) {
		uint32_t y = (uint32_t)f.x*4096;
		if (y<200) return 1;
		else return 2;
	}
	return 3;
}

Here, truth of the condition f.x implies y>=4096, but GCC does not DCE the y<200 test and return 1 codepath.

I actually had this come up in real world code, where I was considering use of an inline function with nontrivial low size cases when a "page count" bitfield is zero, where I expected these nontrivial cases to be optimized out based on already having tested that the page count being nonzero, but GCC was unable to do it. LLVM/clang does it.
Comment 1 Eric Gallager 2020-05-13 04:05:05 UTC
Confirmed via godbolt: https://godbolt.org/z/FexRgJ
Comment 2 Andrew Pinski 2020-05-13 06:08:51 UTC
The problem is if (f.x) gets lowered too early.
This is not directly a VPR issue either.
Changing the code slightly:
#include <stdint.h>
struct foo {
        uint32_t x:20;
};
int bar(struct foo f)
{

       uint32_t y = (uint32_t)f.x;
        if (y) {
                y *= 4096;
                if (y<200) return 1;
                else return 2;
        }
        return 3;
}

GCC is able to optimize it.
Comment 3 Richard Biener 2020-05-13 06:31:14 UTC
Just to quote EVRP sees

  <bb 2> :
  _1 = VIEW_CONVERT_EXPR<unsigned int>(f);
  _2 = _1 & 1048575;
  if (_2 != 0)
    goto <bb 3>; [INV]
  else
    goto <bb 6>; [INV]

  <bb 3> :
  _3 = f.x;
  _4 = (unsigned int) _3;
  y_8 = _4 * 4096;
  if (y_8 <= 199)

thus the f.x != 0 test has been folded by one of those $?%&! permature
fold-const transforms to

  if ((BIT_FIELD_REF <f, 32, 0> & 1048575) != 0)

the fix is to get rid of those (and fix the "fallout").
Comment 4 Andrew Pinski 2021-08-22 00:32:08 UTC
Mine for GCC 13.
Comment 5 Andrew Pinski 2024-07-30 22:35:51 UTC
No longer working on bitfields optimization. Maybe in a few years I will be again but not today.