Use getMaxBits in comparisons

E.g.

(i32.lt_u
 (i32.ne
  (local.get $0)
  (i32.const 0)
 )
 (i32.const 128)
)
 =->
(i32.const 1)

i32.ne returns 0 or 1, so it is max 1 bit, and that is always less than 128.

Found by the superoptimizer #4994 (for comparison to other findings: rule #20, benefit 13760).

generalize to:

u32(x)  < C      ->      1,    iff  maxBits(x) < maxBits(C)
u32(x) <= C      ->      1,    iff  maxBits(x) < maxBits(C)

Also

x == C   =>   0   if maxBits(x) < bits(C)
x != C   =>   1   if maxBits(x) < bits(C)

(to be equal, the number of bits must be equal, so if that differs, the values must differ).

Special case:

x == -1   =>   0   if maxBits(x) != bits(-1)

OptimizeInstructions: Use min/max bits in comparisons (#5035)

This should be complete except for the last part in #5010 (comment) That pattern is basically where the right side is -1 (or more generally a mask of lower bits), so we know more than just the min/max bits.

Navigation Menu

Repository navigation

Use getMaxBits in comparisons #5010

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Use getMaxBits in comparisons #5010

Description

Activity

MaxGraey commented on Sep 3, 2022

kripken commented on Sep 8, 2022

MaxGraey commented on Sep 9, 2022

kripken commented on Sep 13, 2022

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions