llvm/test/CodeGen/X86/bool-simplify.ll
Simon Pilgrim 71589ff672 [X86] Handle COPYs of physregs better (regalloc hints)
Enable enableMultipleCopyHints() on X86.

Original Patch by @jonpa:

While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling.

Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates.

Differential Revision: https://reviews.llvm.org/D38128

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@342578 91177308-0d34-0410-b5e6-96231b3b80d8
2018-09-19 18:59:08 +00:00

162 lines
4.7 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse4.1,-avx,+rdrnd,+rdseed | FileCheck %s
define i32 @foo(<2 x i64> %c, i32 %a, i32 %b) {
; CHECK-LABEL: foo:
; CHECK: # %bb.0:
; CHECK-NEXT: movl %edi, %eax
; CHECK-NEXT: ptest %xmm0, %xmm0
; CHECK-NEXT: cmovnel %esi, %eax
; CHECK-NEXT: retq
%t1 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %c, <2 x i64> %c)
%t2 = icmp ne i32 %t1, 0
%t3 = select i1 %t2, i32 %a, i32 %b
ret i32 %t3
}
define i32 @bar(<2 x i64> %c) {
; CHECK-LABEL: bar:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: ptest %xmm0, %xmm0
; CHECK-NEXT: jne .LBB1_2
; CHECK-NEXT: # %bb.1: # %if-true-block
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: retq
; CHECK-NEXT: .LBB1_2: # %endif-block
; CHECK-NEXT: movl $1, %eax
; CHECK-NEXT: retq
entry:
%0 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %c, <2 x i64> %c)
%1 = icmp ne i32 %0, 0
br i1 %1, label %if-true-block, label %endif-block
if-true-block:
ret i32 0
endif-block:
ret i32 1
}
define i32 @bax(<2 x i64> %c) {
; CHECK-LABEL: bax:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: ptest %xmm0, %xmm0
; CHECK-NEXT: sete %al
; CHECK-NEXT: retq
%t1 = call i32 @llvm.x86.sse41.ptestz(<2 x i64> %c, <2 x i64> %c)
%t2 = icmp eq i32 %t1, 1
%t3 = zext i1 %t2 to i32
ret i32 %t3
}
define i16 @rnd16(i16 %arg) nounwind {
; CHECK-LABEL: rnd16:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdrandw %cx
; CHECK-NEXT: cmovbl %edi, %eax
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%1 = tail call { i16, i32 } @llvm.x86.rdrand.16() nounwind
%2 = extractvalue { i16, i32 } %1, 0
%3 = extractvalue { i16, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i16 0, i16 %arg
%6 = add i16 %5, %2
ret i16 %6
}
define i32 @rnd32(i32 %arg) nounwind {
; CHECK-LABEL: rnd32:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdrandl %ecx
; CHECK-NEXT: cmovbl %edi, %eax
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: retq
%1 = tail call { i32, i32 } @llvm.x86.rdrand.32() nounwind
%2 = extractvalue { i32, i32 } %1, 0
%3 = extractvalue { i32, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i32 0, i32 %arg
%6 = add i32 %5, %2
ret i32 %6
}
define i64 @rnd64(i64 %arg) nounwind {
; CHECK-LABEL: rnd64:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdrandq %rcx
; CHECK-NEXT: cmovbq %rdi, %rax
; CHECK-NEXT: addq %rcx, %rax
; CHECK-NEXT: retq
%1 = tail call { i64, i32 } @llvm.x86.rdrand.64() nounwind
%2 = extractvalue { i64, i32 } %1, 0
%3 = extractvalue { i64, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i64 0, i64 %arg
%6 = add i64 %5, %2
ret i64 %6
}
define i16 @seed16(i16 %arg) nounwind {
; CHECK-LABEL: seed16:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdseedw %cx
; CHECK-NEXT: cmovbl %edi, %eax
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: # kill: def $ax killed $ax killed $eax
; CHECK-NEXT: retq
%1 = tail call { i16, i32 } @llvm.x86.rdseed.16() nounwind
%2 = extractvalue { i16, i32 } %1, 0
%3 = extractvalue { i16, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i16 0, i16 %arg
%6 = add i16 %5, %2
ret i16 %6
}
define i32 @seed32(i32 %arg) nounwind {
; CHECK-LABEL: seed32:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdseedl %ecx
; CHECK-NEXT: cmovbl %edi, %eax
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: retq
%1 = tail call { i32, i32 } @llvm.x86.rdseed.32() nounwind
%2 = extractvalue { i32, i32 } %1, 0
%3 = extractvalue { i32, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i32 0, i32 %arg
%6 = add i32 %5, %2
ret i32 %6
}
define i64 @seed64(i64 %arg) nounwind {
; CHECK-LABEL: seed64:
; CHECK: # %bb.0:
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: rdseedq %rcx
; CHECK-NEXT: cmovbq %rdi, %rax
; CHECK-NEXT: addq %rcx, %rax
; CHECK-NEXT: retq
%1 = tail call { i64, i32 } @llvm.x86.rdseed.64() nounwind
%2 = extractvalue { i64, i32 } %1, 0
%3 = extractvalue { i64, i32 } %1, 1
%4 = icmp eq i32 %3, 0
%5 = select i1 %4, i64 0, i64 %arg
%6 = add i64 %5, %2
ret i64 %6
}
declare i32 @llvm.x86.sse41.ptestz(<2 x i64>, <2 x i64>) nounwind readnone
declare { i16, i32 } @llvm.x86.rdrand.16() nounwind
declare { i32, i32 } @llvm.x86.rdrand.32() nounwind
declare { i64, i32 } @llvm.x86.rdrand.64() nounwind
declare { i16, i32 } @llvm.x86.rdseed.16() nounwind
declare { i32, i32 } @llvm.x86.rdseed.32() nounwind
declare { i64, i32 } @llvm.x86.rdseed.64() nounwind