shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/X86/lsr-addrecloops.ll
Jeremy Morse e6bf48d110
[X86] Don't request 0x90 nop filling in p2align directives (#110134) 
As of rev ea222be0d, LLVMs assembler will actually try to honour the
"fill value" part of p2align directives. X86 printed these as 0x90, which
isn't actually what it wanted: we want multi-byte nops for .text
padding. Compiling via a textual assembly file produces single-byte
nop padding since ea222be0d but the built-in assembler will produce
multi-byte nops. This divergent behaviour is undesirable.

To fix: don't set the byte padding field for x86, which allows the
assembler to pick multi-byte nops. Test that we get the same multi-byte
padding when compiled via textual assembly or directly to object file.
Added same-align-bytes-with-llasm-llobj.ll to that effect, updated
numerous other tests to not contain check-lines for the explicit padding.
2024-10-02 11:14:05 +01:00

296 lines
14 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
; RUN: llc < %s | FileCheck %s
; Check that the SCEVs produced from the multiple loops don't attempt to get
; combines in invalid ways. The LSR filtering could attempt to combine addrecs
; from different loops.
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
define void @in4dob_(ptr nocapture writeonly %0, ptr nocapture readonly %1, ptr nocapture readonly %2, i64 %3, i1 %min.iters.check840) "target-cpu"="icelake-server" {
; CHECK-LABEL: in4dob_:
; CHECK: # %bb.0: # %.preheader263
; CHECK-NEXT: leaq (,%rcx,4), %r9
; CHECK-NEXT: movl $1, %r10d
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: jmp .LBB0_1
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_20: # in Loop: Header=BB0_1 Depth=1
; CHECK-NEXT: incq %r10
; CHECK-NEXT: addq %r9, %rax
; CHECK-NEXT: cmpq %r10, %rcx
; CHECK-NEXT: je .LBB0_18
; CHECK-NEXT: .LBB0_1: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; CHECK-NEXT: vucomiss %xmm0, %xmm1
; CHECK-NEXT: jne .LBB0_20
; CHECK-NEXT: jp .LBB0_20
; CHECK-NEXT: # %bb.2: # in Loop: Header=BB0_1 Depth=1
; CHECK-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
; CHECK-NEXT: vucomiss %xmm0, %xmm1
; CHECK-NEXT: jne .LBB0_20
; CHECK-NEXT: jp .LBB0_20
; CHECK-NEXT: # %bb.3: # %vector.body807.preheader
; CHECK-NEXT: leaq 1(%rcx), %rdx
; CHECK-NEXT: movl %edx, %esi
; CHECK-NEXT: andl $7, %esi
; CHECK-NEXT: cmpq $7, %rcx
; CHECK-NEXT: jae .LBB0_5
; CHECK-NEXT: # %bb.4:
; CHECK-NEXT: xorl %r9d, %r9d
; CHECK-NEXT: jmp .LBB0_7
; CHECK-NEXT: .LBB0_5: # %vector.body807.preheader.new
; CHECK-NEXT: movq %rdx, %r10
; CHECK-NEXT: andq $-8, %r10
; CHECK-NEXT: xorl %r9d, %r9d
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_6: # %vector.body807
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: leaq (%rdi,%r9), %r11
; CHECK-NEXT: vmovups %ymm0, (%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 1(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 2(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 3(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 4(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 5(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 6(%rax,%r11)
; CHECK-NEXT: vmovups %ymm0, 7(%rax,%r11)
; CHECK-NEXT: addq $8, %r9
; CHECK-NEXT: cmpq %r9, %r10
; CHECK-NEXT: jne .LBB0_6
; CHECK-NEXT: .LBB0_7: # %.lr.ph373.unr-lcssa
; CHECK-NEXT: testq %rsi, %rsi
; CHECK-NEXT: je .LBB0_10
; CHECK-NEXT: # %bb.8: # %vector.body807.epil.preheader
; CHECK-NEXT: addq %rdi, %r9
; CHECK-NEXT: xorl %r10d, %r10d
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_9: # %vector.body807.epil
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: leaq (%r9,%r10), %r11
; CHECK-NEXT: vmovups %ymm0, (%rax,%r11)
; CHECK-NEXT: incq %r10
; CHECK-NEXT: cmpq %r10, %rsi
; CHECK-NEXT: jne .LBB0_9
; CHECK-NEXT: .LBB0_10: # %.lr.ph373
; CHECK-NEXT: testb $1, %r8b
; CHECK-NEXT: je .LBB0_11
; CHECK-NEXT: # %bb.19: # %scalar.ph839.preheader
; CHECK-NEXT: movl $0, (%rdi)
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
; CHECK-NEXT: .LBB0_11: # %vector.body847.preheader
; CHECK-NEXT: movl %edx, %esi
; CHECK-NEXT: andl $7, %esi
; CHECK-NEXT: cmpq $7, %rcx
; CHECK-NEXT: jae .LBB0_13
; CHECK-NEXT: # %bb.12:
; CHECK-NEXT: xorl %ecx, %ecx
; CHECK-NEXT: jmp .LBB0_15
; CHECK-NEXT: .LBB0_13: # %vector.body847.preheader.new
; CHECK-NEXT: andq $-8, %rdx
; CHECK-NEXT: xorl %ecx, %ecx
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_14: # %vector.body847
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: leaq (%rdi,%rcx), %r8
; CHECK-NEXT: vmovups %ymm0, 96(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 97(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 98(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 99(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 100(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 101(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 102(%rax,%r8)
; CHECK-NEXT: vmovups %ymm0, 103(%rax,%r8)
; CHECK-NEXT: addq $8, %rcx
; CHECK-NEXT: cmpq %rcx, %rdx
; CHECK-NEXT: jne .LBB0_14
; CHECK-NEXT: .LBB0_15: # %common.ret.loopexit.unr-lcssa
; CHECK-NEXT: testq %rsi, %rsi
; CHECK-NEXT: je .LBB0_18
; CHECK-NEXT: # %bb.16: # %vector.body847.epil.preheader
; CHECK-NEXT: leaq 96(%rcx,%rdi), %rcx
; CHECK-NEXT: xorl %edx, %edx
; CHECK-NEXT: vxorps %xmm0, %xmm0, %xmm0
; CHECK-NEXT: .p2align 4
; CHECK-NEXT: .LBB0_17: # %vector.body847.epil
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
; CHECK-NEXT: leaq (%rcx,%rdx), %rdi
; CHECK-NEXT: vmovups %ymm0, (%rax,%rdi)
; CHECK-NEXT: incq %rdx
; CHECK-NEXT: cmpq %rdx, %rsi
; CHECK-NEXT: jne .LBB0_17
; CHECK-NEXT: .LBB0_18: # %common.ret
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
.preheader263:
%4 = shl i64 %3, 2
br label %5
5: ; preds = %16, %.preheader263
%lsr.iv1135 = phi ptr [ %0, %.preheader263 ], [ %uglygep1136, %16 ]
%indvars.iv487 = phi i64 [ 1, %.preheader263 ], [ %indvars.iv.next488, %16 ]
%6 = getelementptr float, ptr %1, i64 %indvars.iv487
%7 = load float, ptr %6, align 4
%8 = fcmp oeq float %7, 0.000000e+00
%9 = getelementptr float, ptr %2, i64 %indvars.iv487
%10 = load float, ptr %9, align 4
%11 = fcmp oeq float %10, 0.000000e+00
%12 = and i1 %8, %11
br i1 %12, label %vector.body807.preheader, label %16
vector.body807.preheader: ; preds = %5
%13 = add i64 %3, 1
%xtraiter = and i64 %13, 7
%14 = icmp ult i64 %3, 7
br i1 %14, label %.lr.ph373.unr-lcssa, label %vector.body807.preheader.new
vector.body807.preheader.new: ; preds = %vector.body807.preheader
%unroll_iter = and i64 %13, -8
br label %vector.body807
vector.body807: ; preds = %vector.body807, %vector.body807.preheader.new
%lsr.iv1194 = phi i64 [ 0, %vector.body807.preheader.new ], [ %lsr.iv.next1195.7, %vector.body807 ]
%niter = phi i64 [ 0, %vector.body807.preheader.new ], [ %niter.next.7, %vector.body807 ]
%uglygep1197 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv1194
store <8 x float> zeroinitializer, ptr %uglygep1197, align 4
%lsr.iv.next1195 = or disjoint i64 %lsr.iv1194, 1
%uglygep1197.1 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195
store <8 x float> zeroinitializer, ptr %uglygep1197.1, align 4
%lsr.iv.next1195.1 = or disjoint i64 %lsr.iv1194, 2
%uglygep1197.2 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.1
store <8 x float> zeroinitializer, ptr %uglygep1197.2, align 4
%lsr.iv.next1195.2 = or disjoint i64 %lsr.iv1194, 3
%uglygep1197.3 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.2
store <8 x float> zeroinitializer, ptr %uglygep1197.3, align 4
%lsr.iv.next1195.3 = or disjoint i64 %lsr.iv1194, 4
%uglygep1197.4 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.3
store <8 x float> zeroinitializer, ptr %uglygep1197.4, align 4
%lsr.iv.next1195.4 = or disjoint i64 %lsr.iv1194, 5
%uglygep1197.5 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.4
store <8 x float> zeroinitializer, ptr %uglygep1197.5, align 4
%lsr.iv.next1195.5 = or disjoint i64 %lsr.iv1194, 6
%uglygep1197.6 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.5
store <8 x float> zeroinitializer, ptr %uglygep1197.6, align 4
%lsr.iv.next1195.6 = or disjoint i64 %lsr.iv1194, 7
%uglygep1197.7 = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv.next1195.6
store <8 x float> zeroinitializer, ptr %uglygep1197.7, align 4
%lsr.iv.next1195.7 = add i64 %lsr.iv1194, 8
%niter.next.7 = add i64 %niter, 8
%niter.ncmp.7 = icmp eq i64 %niter.next.7, %unroll_iter
br i1 %niter.ncmp.7, label %.lr.ph373.unr-lcssa.loopexit, label %vector.body807
.lr.ph373.unr-lcssa.loopexit: ; preds = %vector.body807
br label %.lr.ph373.unr-lcssa
.lr.ph373.unr-lcssa: ; preds = %.lr.ph373.unr-lcssa.loopexit, %vector.body807.preheader
%lsr.iv1194.unr = phi i64 [ 0, %vector.body807.preheader ], [ %lsr.iv.next1195.7, %.lr.ph373.unr-lcssa.loopexit ]
%lcmp.mod.not = icmp eq i64 %xtraiter, 0
br i1 %lcmp.mod.not, label %.lr.ph373, label %vector.body807.epil.preheader
vector.body807.epil.preheader: ; preds = %.lr.ph373.unr-lcssa
br label %vector.body807.epil
vector.body807.epil: ; preds = %vector.body807.epil.preheader, %vector.body807.epil
%lsr.iv1194.epil = phi i64 [ %lsr.iv.next1195.epil, %vector.body807.epil ], [ %lsr.iv1194.unr, %vector.body807.epil.preheader ]
%epil.iter = phi i64 [ %epil.iter.next, %vector.body807.epil ], [ 0, %vector.body807.epil.preheader ]
%uglygep1197.epil = getelementptr i8, ptr %lsr.iv1135, i64 %lsr.iv1194.epil
store <8 x float> zeroinitializer, ptr %uglygep1197.epil, align 4
%lsr.iv.next1195.epil = add i64 %lsr.iv1194.epil, 1
%epil.iter.next = add i64 %epil.iter, 1
%epil.iter.cmp.not = icmp eq i64 %epil.iter.next, %xtraiter
br i1 %epil.iter.cmp.not, label %.lr.ph373.loopexit, label %vector.body807.epil
.lr.ph373.loopexit: ; preds = %vector.body807.epil
br label %.lr.ph373
.lr.ph373: ; preds = %.lr.ph373.loopexit, %.lr.ph373.unr-lcssa
br i1 %min.iters.check840, label %scalar.ph839.preheader, label %vector.body847.preheader
vector.body847.preheader: ; preds = %.lr.ph373
%uglygep11551 = getelementptr i8, ptr %lsr.iv1135, i64 96
%xtraiter12 = and i64 %13, 7
%15 = icmp ult i64 %3, 7
br i1 %15, label %common.ret.loopexit.unr-lcssa, label %vector.body847.preheader.new
vector.body847.preheader.new: ; preds = %vector.body847.preheader
%unroll_iter15 = and i64 %13, -8
br label %vector.body847
vector.body847: ; preds = %vector.body847, %vector.body847.preheader.new
%lsr.iv1152 = phi i64 [ 0, %vector.body847.preheader.new ], [ %lsr.iv.next1153.7, %vector.body847 ]
%niter16 = phi i64 [ 0, %vector.body847.preheader.new ], [ %niter16.next.7, %vector.body847 ]
%uglygep1156 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv1152
store <8 x float> zeroinitializer, ptr %uglygep1156, align 4
%lsr.iv.next1153 = or disjoint i64 %lsr.iv1152, 1
%uglygep1156.1 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153
store <8 x float> zeroinitializer, ptr %uglygep1156.1, align 4
%lsr.iv.next1153.1 = or disjoint i64 %lsr.iv1152, 2
%uglygep1156.2 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.1
store <8 x float> zeroinitializer, ptr %uglygep1156.2, align 4
%lsr.iv.next1153.2 = or disjoint i64 %lsr.iv1152, 3
%uglygep1156.3 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.2
store <8 x float> zeroinitializer, ptr %uglygep1156.3, align 4
%lsr.iv.next1153.3 = or disjoint i64 %lsr.iv1152, 4
%uglygep1156.4 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.3
store <8 x float> zeroinitializer, ptr %uglygep1156.4, align 4
%lsr.iv.next1153.4 = or disjoint i64 %lsr.iv1152, 5
%uglygep1156.5 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.4
store <8 x float> zeroinitializer, ptr %uglygep1156.5, align 4
%lsr.iv.next1153.5 = or disjoint i64 %lsr.iv1152, 6
%uglygep1156.6 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.5
store <8 x float> zeroinitializer, ptr %uglygep1156.6, align 4
%lsr.iv.next1153.6 = or disjoint i64 %lsr.iv1152, 7
%uglygep1156.7 = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv.next1153.6
store <8 x float> zeroinitializer, ptr %uglygep1156.7, align 4
%lsr.iv.next1153.7 = add i64 %lsr.iv1152, 8
%niter16.next.7 = add i64 %niter16, 8
%niter16.ncmp.7 = icmp eq i64 %niter16.next.7, %unroll_iter15
br i1 %niter16.ncmp.7, label %common.ret.loopexit.unr-lcssa.loopexit, label %vector.body847
common.ret.loopexit.unr-lcssa.loopexit: ; preds = %vector.body847
br label %common.ret.loopexit.unr-lcssa
common.ret.loopexit.unr-lcssa: ; preds = %common.ret.loopexit.unr-lcssa.loopexit, %vector.body847.preheader
%lsr.iv1152.unr = phi i64 [ 0, %vector.body847.preheader ], [ %lsr.iv.next1153.7, %common.ret.loopexit.unr-lcssa.loopexit ]
%lcmp.mod14.not = icmp eq i64 %xtraiter12, 0
br i1 %lcmp.mod14.not, label %common.ret, label %vector.body847.epil.preheader
vector.body847.epil.preheader: ; preds = %common.ret.loopexit.unr-lcssa
br label %vector.body847.epil
vector.body847.epil: ; preds = %vector.body847.epil.preheader, %vector.body847.epil
%lsr.iv1152.epil = phi i64 [ %lsr.iv.next1153.epil, %vector.body847.epil ], [ %lsr.iv1152.unr, %vector.body847.epil.preheader ]
%epil.iter13 = phi i64 [ %epil.iter13.next, %vector.body847.epil ], [ 0, %vector.body847.epil.preheader ]
%uglygep1156.epil = getelementptr i8, ptr %uglygep11551, i64 %lsr.iv1152.epil
store <8 x float> zeroinitializer, ptr %uglygep1156.epil, align 4
%lsr.iv.next1153.epil = add i64 %lsr.iv1152.epil, 1
%epil.iter13.next = add i64 %epil.iter13, 1
%epil.iter13.cmp.not = icmp eq i64 %epil.iter13.next, %xtraiter12
br i1 %epil.iter13.cmp.not, label %common.ret.loopexit, label %vector.body847.epil
common.ret.loopexit: ; preds = %vector.body847.epil
br label %common.ret
common.ret.loopexit1: ; preds = %16
br label %common.ret
common.ret: ; preds = %common.ret.loopexit1, %common.ret.loopexit, %scalar.ph839.preheader, %common.ret.loopexit.unr-lcssa
ret void
scalar.ph839.preheader: ; preds = %.lr.ph373
store float 0.000000e+00, ptr %0, align 4
br label %common.ret
16: ; preds = %5
%indvars.iv.next488 = add i64 %indvars.iv487, 1
%exitcond492.not = icmp eq i64 %indvars.iv.next488, %3
%uglygep1136 = getelementptr i8, ptr %lsr.iv1135, i64 %4
br i1 %exitcond492.not, label %common.ret.loopexit1, label %5
}
Reference in New Issue View Git Blame Copy Permalink