https://llvm.org/bugs/show_bug.cgi?id=26374
Bug ID: 26374
Summary: Unused stackslots generated in reg spills
Product: libraries
Version: trunk
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: Register Allocator
Assignee: unassignedb...@nondot.org
Reporter: w...@google.com
CC: llvm-bugs@lists.llvm.org
Classification: Unclassified
Created attachment 15754
--> https://llvm.org/bugs/attachment.cgi?id=15754&action=edit
testcase 1.c
The problem was originally described in
https://llvm.org/bugs/show_bug.cgi?id=25776 #comment3. Reposted it here to give
it a separate bug number so the testcase of the patch to fix it can use the
number.
$ ~/workarea/llvm-r254800/build/bin/clang -O2 -S 1.c
$ grep 'rsp' 1.s
subq $408, %rsp # imm = 0x198
movl $0, 124(%rsp) # 4-byte Folded Spill
movq %rdx, 72(%rsp) # 8-byte Spill
movq %rax, 128(%rsp) # 8-byte Spill
movq %rax, 88(%rsp) # 8-byte Spill
movq %rdx, 96(%rsp) # 8-byte Spill
movq %rcx, 80(%rsp) # 8-byte Spill
leaq 240(%rsp), %r15
movq %r15, 104(%rsp) # 8-byte Spill
movq %r14, 112(%rsp) # 8-byte Spill
movq %rsi, 136(%rsp) # 8-byte Spill
movq 96(%rsp), %rcx # 8-byte Reload
movq 80(%rsp), %r13 # 8-byte Reload
movl $0, (%rsp)
addl %eax, 124(%rsp) # 4-byte Folded Spill
movq 136(%rsp), %rsi # 8-byte Reload
cmpq 128(%rsp), %r13 # 8-byte Folded Reload
movq 112(%rsp), %r14 # 8-byte Reload
movq 104(%rsp), %r15 # 8-byte Reload
cmpq 88(%rsp), %rsi # 8-byte Folded Reload
movaps 144(%rsp), %xmm0
movaps 160(%rsp), %xmm1
leaq 176(%rsp), %rax
leaq 208(%rsp), %rax
leaq 240(%rsp), %rax
leaq 272(%rsp), %rax
leaq 304(%rsp), %rax
leaq 336(%rsp), %rax
leaq 368(%rsp), %rax
addl %eax, 124(%rsp) # 4-byte Folded Spill
movq 72(%rsp), %rdx # 8-byte Reload
movl 124(%rsp), %eax # 4-byte Reload
addq $408, %rsp # imm = 0x198
88(%rsp) and 96(%rsp) are both used as Spill stack slots above. 90(%rsp) is a
hole which is never used.
A part of -debug-only=regalloc dump result:
********** REGISTER MAP **********
[%vreg0 -> fi#1] GR64
[%vreg1 -> fi#2] GR64
[%vreg2 -> fi#3] GR64
[%vreg3 -> fi#4] GR64
[%vreg4 -> fi#5] GR64
...
fi#1,fi#2,fi#3,fi#4,fi#5 have no references in the IR, which means there will
be a lot of unused holes in the stack allocated.
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
