mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-09-01 19:29:22 +00:00
This changes the SelectionDAG scheduling preference to source order. Soon, the SelectionDAG scheduler can be bypassed saving a nice chunk of compile time. Performance differences that result from this change are often a consequence of register coalescing. The register coalescer is far from perfect. Bugs can be filed for deficiencies. On x86 SandyBridge/Haswell, the source order schedule is often preserved, particularly for small blocks. Register pressure is generally improved over the SD scheduler's ILP mode. However, we are still able to handle large blocks that require latency hiding, unlike the SD scheduler's BURR mode. MI scheduler also attempts to discover the critical path in single-block loops and adjust heuristics accordingly. The MI scheduler relies on the new machine model. This is currently unimplemented for AVX, so we may not be generating the best code yet. Unit tests are updated so they don't depend on SD scheduling heuristics. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192750 91177308-0d34-0410-b5e6-96231b3b80d8
63 lines
1.8 KiB
LLVM
63 lines
1.8 KiB
LLVM
; The register allocator can commute two-address instructions to avoid
|
|
; insertion of register-register copies.
|
|
|
|
; Make sure there are only 3 mov's for each testcase
|
|
; RUN: llc < %s -mtriple=i686-pc-linux-gnu -mcpu=corei7 | FileCheck %s -check-prefix=LINUX
|
|
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7 | FileCheck %s -check-prefix=DARWIN
|
|
|
|
|
|
@G = external global i32 ; <i32*> [#uses=2]
|
|
|
|
declare void @ext(i32)
|
|
|
|
define i32 @t1(i32 %X, i32 %Y) nounwind {
|
|
; LINUX-LABEL: t1:
|
|
; LINUX: movl 4(%esp), %eax
|
|
; LINUX: movl 8(%esp), %ecx
|
|
; LINUX: addl %eax, %ecx
|
|
; LINUX: movl %ecx, G
|
|
%Z = add i32 %X, %Y ; <i32> [#uses=1]
|
|
store i32 %Z, i32* @G
|
|
ret i32 %X
|
|
}
|
|
|
|
define i32 @t2(i32 %X, i32 %Y) nounwind {
|
|
; LINUX-LABEL: t2:
|
|
; LINUX: movl 4(%esp), %eax
|
|
; LINUX: movl 8(%esp), %ecx
|
|
; LINUX: xorl %eax, %ecx
|
|
; LINUX: movl %ecx, G
|
|
%Z = xor i32 %X, %Y ; <i32> [#uses=1]
|
|
store i32 %Z, i32* @G
|
|
ret i32 %X
|
|
}
|
|
|
|
; rdar://8762995
|
|
%0 = type { i64, i32 }
|
|
|
|
define %0 @t3(i32 %lb, i8 zeroext %has_lb, i8 zeroext %lb_inclusive, i32 %ub, i8 zeroext %has_ub, i8 zeroext %ub_inclusive) nounwind {
|
|
entry:
|
|
; DARWIN-LABEL: t3:
|
|
; DARWIN: shlq $32, %rcx
|
|
; DARWIN-NEXT: orq %rcx, %rax
|
|
; DARWIN-NEXT: shll $8
|
|
; DARWIN-NOT: leaq
|
|
%tmp21 = zext i32 %lb to i64
|
|
%tmp23 = zext i32 %ub to i64
|
|
%tmp24 = shl i64 %tmp23, 32
|
|
%ins26 = or i64 %tmp24, %tmp21
|
|
%tmp28 = zext i8 %has_lb to i32
|
|
%tmp33 = zext i8 %has_ub to i32
|
|
%tmp34 = shl i32 %tmp33, 8
|
|
%tmp38 = zext i8 %lb_inclusive to i32
|
|
%tmp39 = shl i32 %tmp38, 16
|
|
%tmp43 = zext i8 %ub_inclusive to i32
|
|
%tmp44 = shl i32 %tmp43, 24
|
|
%ins31 = or i32 %tmp39, %tmp28
|
|
%ins36 = or i32 %ins31, %tmp34
|
|
%ins46 = or i32 %ins36, %tmp44
|
|
%tmp16 = insertvalue %0 undef, i64 %ins26, 0
|
|
%tmp19 = insertvalue %0 %tmp16, i32 %ins46, 1
|
|
ret %0 %tmp19
|
|
}
|