1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 19:23:23 +01:00
llvm-mirror/test/CodeGen/X86/memset.ll
Jonas Paulsson 329a30125b [SchedModel] Fix for read advance cycles with implicit pseudo operands.
The SchedModel allows the addition of ReadAdvances to express that certain
operands of the instructions are needed at a later point than the others.

RegAlloc may add pseudo operands that are not part of the instruction
descriptor, and therefore cannot have any read advance entries. This meant
that in some cases the desired read advance was nullified by such a pseudo
operand, which still had the original latency.

This patch fixes this by making sure that such pseudo operands get a zero
latency during DAG construction.

Review: Matthias Braun, Ulrich Weigand.
https://reviews.llvm.org/D49671

llvm-svn: 345606
2018-10-30 15:04:40 +00:00

100 lines
3.2 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mcpu=pentium2 -mtriple=i686-apple-darwin8.8.0 | FileCheck %s --check-prefix=X86
; RUN: llc < %s -mcpu=pentium3 -mtriple=i686-apple-darwin8.8.0 | FileCheck %s --check-prefix=XMM
; RUN: llc < %s -mcpu=bdver1 -mtriple=i686-apple-darwin8.8.0 | FileCheck %s --check-prefix=YMM
%struct.x = type { i16, i16 }
define void @t() nounwind {
; X86-LABEL: t:
; X86: ## %bb.0: ## %entry
; X86-NEXT: subl $44, %esp
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
; X86-NEXT: leal {{[0-9]+}}(%esp), %eax
; X86-NEXT: movl %eax, (%esp)
; X86-NEXT: calll _foo
; X86-NEXT: addl $44, %esp
; X86-NEXT: retl
;
; XMM-LABEL: t:
; XMM: ## %bb.0: ## %entry
; XMM-NEXT: subl $60, %esp
; XMM-NEXT: xorps %xmm0, %xmm0
; XMM-NEXT: movaps %xmm0, {{[0-9]+}}(%esp)
; XMM-NEXT: movaps %xmm0, {{[0-9]+}}(%esp)
; XMM-NEXT: leal {{[0-9]+}}(%esp), %eax
; XMM-NEXT: movl %eax, (%esp)
; XMM-NEXT: calll _foo
; XMM-NEXT: addl $60, %esp
; XMM-NEXT: retl
;
; YMM-LABEL: t:
; YMM: ## %bb.0: ## %entry
; YMM-NEXT: pushl %ebp
; YMM-NEXT: movl %esp, %ebp
; YMM-NEXT: andl $-32, %esp
; YMM-NEXT: subl $96, %esp
; YMM-NEXT: leal {{[0-9]+}}(%esp), %eax
; YMM-NEXT: vxorps %xmm0, %xmm0, %xmm0
; YMM-NEXT: vmovaps %ymm0, {{[0-9]+}}(%esp)
; YMM-NEXT: movl %eax, (%esp)
; YMM-NEXT: vzeroupper
; YMM-NEXT: calll _foo
; YMM-NEXT: movl %ebp, %esp
; YMM-NEXT: popl %ebp
; YMM-NEXT: retl
entry:
%up_mvd = alloca [8 x %struct.x] ; <[8 x %struct.x]*> [#uses=2]
%up_mvd116 = getelementptr [8 x %struct.x], [8 x %struct.x]* %up_mvd, i32 0, i32 0 ; <%struct.x*> [#uses=1]
%tmp110117 = bitcast [8 x %struct.x]* %up_mvd to i8* ; <i8*> [#uses=1]
call void @llvm.memset.p0i8.i64(i8* align 8 %tmp110117, i8 0, i64 32, i1 false)
call void @foo( %struct.x* %up_mvd116 ) nounwind
ret void
}
declare void @foo(%struct.x*)
declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i1) nounwind
; Ensure that alignment of '0' in an @llvm.memset intrinsic results in
; unaligned loads and stores.
define void @PR15348(i8* %a) {
; X86-LABEL: PR15348:
; X86: ## %bb.0:
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
; X86-NEXT: movb $0, 16(%eax)
; X86-NEXT: movl $0, 12(%eax)
; X86-NEXT: movl $0, 8(%eax)
; X86-NEXT: movl $0, 4(%eax)
; X86-NEXT: movl $0, (%eax)
; X86-NEXT: retl
;
; XMM-LABEL: PR15348:
; XMM: ## %bb.0:
; XMM-NEXT: movl {{[0-9]+}}(%esp), %eax
; XMM-NEXT: movb $0, 16(%eax)
; XMM-NEXT: movl $0, 12(%eax)
; XMM-NEXT: movl $0, 8(%eax)
; XMM-NEXT: movl $0, 4(%eax)
; XMM-NEXT: movl $0, (%eax)
; XMM-NEXT: retl
;
; YMM-LABEL: PR15348:
; YMM: ## %bb.0:
; YMM-NEXT: movl {{[0-9]+}}(%esp), %eax
; YMM-NEXT: vxorps %xmm0, %xmm0, %xmm0
; YMM-NEXT: vmovups %xmm0, (%eax)
; YMM-NEXT: movb $0, 16(%eax)
; YMM-NEXT: retl
call void @llvm.memset.p0i8.i64(i8* %a, i8 0, i64 17, i1 false)
ret void
}