tree 0b674505b93ce75be9d04000fbaf546be5822877
parent d85a7a27b27269741a779a72f33436b83d98687c
author Nicolas Capens <capn@google.com> 1612589738 -0500
committer Nicolas Capens <nicolascapens@google.com> 1614018302 +0000

Implement propagation of stores to loads in single basic blocks

For loads following stores in the same basic block, replace their result
with the data that was stored.

It transforms the output of the StoresInMultipleBlocks test from:

 sub         rsp,38h
 mov         dword ptr [rsp],0Dh
 cmp         ecx,0
 je          a
 mov         dword ptr [rsp],4
 add         dword ptr [rsp],3
b:
 mov         eax,dword ptr [rsp]
 add         rsp,38h
 ret
a:
 mov         dword ptr [rsp],6
 add         dword ptr [rsp],5
 jmp         b

Into:

 sub         rsp,38h
 mov         dword ptr [rsp],0Dh
 cmp         ecx,0
 je          a
 mov         dword ptr [rsp],4
 mov         eax,4
 add         eax,3
 mov         dword ptr [rsp],eax
b:
 mov         eax,dword ptr [rsp]
 add         rsp,38h
 ret
a:
 mov         dword ptr [rsp],6
 mov         eax,6
 add         eax,5
 mov         dword ptr [rsp],eax
 jmp         b

While at first this might seem like a regression, note that
`add [rsp],3` performs both a load and a store. The optimization pass
eliminated two load operations from this test. The redundant stores will
be eliminated by a subsequent change.

Also adds a unit test for the case where store-to-load propagation
should not be performed due to an indirect store through a pointer
happening in between.

Bug: b/179668593
Change-Id: I6ca133ac4e77bbbc3efd517dff6e8dee6d2dc147
Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/52533
Kokoro-Result: kokoro <noreply+kokoro@google.com>
Tested-by: Nicolas Capens <nicolascapens@google.com>
Presubmit-Ready: Nicolas Capens <nicolascapens@google.com>
Reviewed-by: Antonio Maiorano <amaiorano@google.com>