llvm/test/CodeGen/ARM/eh-dispcont.ll
Andrew Trick 6ea2b9608a Allocate local registers in order for optimal coloring.
Also avoid locals evicting locals just because they want a cheaper register.

Problem: MI Sched knows exactly how many registers we have and assumes
they can be colored. In cases where we have large blocks, usually from
unrolled loops, greedy coloring fails. This is a source of
"regressions" from the MI Scheduler on x86. I noticed this issue on
x86 where we have long chains of two-address defs in the same live
range. It's easy to see this in matrix multiplication benchmarks like
IRSmk and even the unit test misched-matmul.ll.

A fundamental difference between the LLVM register allocator and
conventional graph coloring is that in our model a live range can't
discover its neighbors, it can only verify its neighbors. That's why
we initially went for greedy coloring and added eviction to deal with
the hard cases. However, for singly defined and two-address live
ranges, we can optimally color without visiting neighbors simply by
processing the live ranges in instruction order.

Other beneficial side effects:

It is much easier to understand and debug regalloc for large blocks
when the live ranges are allocated in order. Yes, global allocation is
still very confusing, but it's nice to be able to comprehend what
happened locally.

Heuristics could be added to bias register assignment based on
instruction locality (think late register pairing, banks...).

Intuituvely this will make some test cases that are on the threshold
of register pressure more stable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187139 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-25 18:35:14 +00:00

90 lines
3.1 KiB
LLVM

; RUN: llc -mtriple armv7-apple-ios -relocation-model=pic -o - %s | FileCheck %s -check-prefix=ARM-PIC
; RUN: llc -mtriple armv7-apple-ios -relocation-model=static -o - %s | FileCheck %s -check-prefix=ARM-NOPIC
; RUN: llc -mtriple armv7-apple-ios -relocation-model=dynamic-no-pic -o - %s | FileCheck %s -check-prefix=ARM-NOPIC
; RUN: llc -mtriple thumbv6-apple-ios -relocation-model=pic -o - %s | FileCheck %s -check-prefix=THUMB1-PIC
; RUN: llc -mtriple thumbv6-apple-ios -relocation-model=static -o - %s | FileCheck %s -check-prefix=THUMB1-NOPIC
; RUN: llc -mtriple thumbv6-apple-ios -relocation-model=dynamic-no-pic -o - %s | FileCheck %s -check-prefix=THUMB1-NOPIC
@_ZTIi = external constant i8*
define i32 @main() #0 {
entry:
%exception = tail call i8* @__cxa_allocate_exception(i32 4) #1
%0 = bitcast i8* %exception to i32*
store i32 1, i32* %0, align 4
invoke void @__cxa_throw(i8* %exception, i8* bitcast (i8** @_ZTIi to i8*), i8* null) #2
to label %unreachable unwind label %lpad
lpad: ; preds = %entry
%1 = landingpad { i8*, i32 } personality i8* bitcast (i32 (...)* @__gxx_personality_sj0 to i8*)
catch i8* null
%2 = extractvalue { i8*, i32 } %1, 0
%3 = tail call i8* @__cxa_begin_catch(i8* %2) #1
tail call void @__cxa_end_catch()
ret i32 0
unreachable: ; preds = %entry
unreachable
}
declare i8* @__cxa_allocate_exception(i32)
declare void @__cxa_throw(i8*, i8*, i8*)
declare i8* @__cxa_begin_catch(i8*)
declare void @__cxa_end_catch()
declare i32 @__gxx_personality_sj0(...)
attributes #0 = { ssp }
attributes #1 = { nounwind }
attributes #2 = { noreturn }
; ARM-PIC: cxa_throw
; ARM-PIC: trap
; ARM-PIC: adr [[REG1:r[0-9]+]], [[LJTI:.*]]
; ARM-PIC: ldr [[REG0:r[0-9]+]], [r{{[0-9]+}}, [[REG1]]]
; ARM-PIC: add pc, [[REG0]], [[REG1]]
; ARM-PIC: [[LJTI]]
; ARM-PIC: .data_region jt32
; ARM-PIC: .long [[LABEL:LBB0_[0-9]]]-[[LJTI]]
; ARM-PIC: .end_data_region
; ARM-PIC: [[LABEL]]
; ARM-NOPIC: cxa_throw
; ARM-NOPIC: trap
; ARM-NOPIC: adr [[REG1:r[0-9]+]], [[LJTI:.*]]
; ARM-NOPIC: ldr [[REG0:r[0-9]+]], [r{{[0-9]+}}, [[REG1]]]
; ARM-NOPIC: mov pc, [[REG0]]
; ARM-NOPIC: [[LJTI]]
; ARM-NOPIC: .data_region jt32
; ARM-NOPIC: .long [[LABEL:LBB0_[0-9]]]
; ARM-NOPIC: .end_data_region
; ARM-NOPIC: [[LABEL]]
; THUMB1-PIC: cxa_throw
; THUMB1-PIC: trap
; THUMB1-PIC: adr [[REG1:r[0-9]+]], [[LJTI:.*]]
; THUMB1-PIC: adds [[REG0:r[0-9]+]], [[REG0]], [[REG1]]
; THUMB1-PIC: ldr [[REG0]]
; THUMB1-PIC: adds [[REG0]], [[REG0]], [[REG1]]
; THUMB1-PIC: mov pc, [[REG0]]
; THUMB1-PIC: [[LJTI]]
; THUMB1-PIC: .data_region jt32
; THUMB1-PIC: .long [[LABEL:LBB0_[0-9]]]-[[LJTI]]
; THUMB1-PIC: .end_data_region
; THUMB1-PIC: [[LABEL]]
; THUMB1-NOPIC: cxa_throw
; THUMB1-NOPIC: trap
; THUMB1-NOPIC: adr [[REG1:r[0-9]+]], [[LJTI:.*]]
; THUMB1-NOPIC: adds [[REG0:r[0-9]+]], [[REG0]], [[REG1]]
; THUMB1-NOPIC: ldr [[REG0]]
; THUMB1-NOPIC: mov pc, [[REG0]]
; THUMB1-NOPIC: [[LJTI]]
; THUMB1-NOPIC: .data_region jt32
; THUMB1-NOPIC: .long [[LABEL:LBB0_[0-9]]]+1
; THUMB1-NOPIC: .end_data_region
; THUMB1-NOPIC: [[LABEL]]