[X86] Reduce Store Forward Block issues in HW - Recommit after fixing Bug 36346

If a load follows a store and reloads data that the store has written to memory, Intel microarchitectures can in many cases forward the data directly from the store to the load, This "store forwarding" saves cycles by enabling the load to directly obtain the data instead of accessing the data from cache or memory.
A "store forward block" occurs in cases that a store cannot be forwarded to the load. The most typical case of store forward block on Intel Core microarchiticutre that a small store cannot be forwarded to a large load.
The estimated penalty for a store forward block is ~13 cycles.

This pass tries to recognize and handle cases where "store forward block" is created by the compiler when lowering memcpy calls to a sequence
of a load and a store.

The pass currently only handles cases where memcpy is lowered to XMM/YMM registers, it tries to break the memcpy into smaller copies.
breaking the memcpy should be possible since there is no atomicity guarantee for loads and stores to XMM/YMM.

Differential revision: https://reviews.llvm.org/D41330

Change-Id: Ib48836ccdf6005989f7d4466fa2035b7b04415d9
llvm-svn: 328973
diff --git a/llvm/test/CodeGen/X86/avoid-sfb.ll b/llvm/test/CodeGen/X86/avoid-sfb.ll
new file mode 100644
index 0000000..0e2d292
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avoid-sfb.ll
@@ -0,0 +1,1491 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefix=CHECK
+; RUN: llc < %s -mtriple=x86_64-linux --x86-disable-avoid-SFB | FileCheck %s --check-prefix=DISABLED
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=core-avx2 | FileCheck %s -check-prefix=CHECK-AVX2
+; RUN: llc < %s -mtriple=x86_64-linux -mcpu=skx | FileCheck %s -check-prefix=CHECK-AVX512
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.S = type { i32, i32, i32, i32 }
+
+; Function Attrs: nounwind uwtable
+define void @test_conditional_block(%struct.S* nocapture noalias %s1 , %struct.S* nocapture noalias %s2, i32 %x, %struct.S* nocapture noalias  %s3, %struct.S* nocapture noalias readonly %s4) local_unnamed_addr #0 {
+; CHECK-LABEL: test_conditional_block:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB0_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl %edx, 4(%rdi)
+; CHECK-NEXT:  .LBB0_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl %eax, (%rsi)
+; CHECK-NEXT:    movl 4(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 4(%rsi)
+; CHECK-NEXT:    movq 8(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 8(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_conditional_block:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB0_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl %edx, 4(%rdi)
+; DISABLED-NEXT:  .LBB0_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_conditional_block:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB0_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX2-NEXT:  .LBB0_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX2-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX2-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_conditional_block:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB0_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX512-NEXT:  .LBB0_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX512-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX512-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 1
+  store i32 %x, i32* %b, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S* %s3 to i8*
+  %1 = bitcast %struct.S* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  %2 = bitcast %struct.S* %s2 to i8*
+  %3 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define void @test_imm_store(%struct.S* nocapture noalias %s1, %struct.S* nocapture %s2, i32 %x, %struct.S* nocapture %s3) local_unnamed_addr #0 {
+; CHECK-LABEL: test_imm_store:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl $0, (%rdi)
+; CHECK-NEXT:    movl $1, (%rcx)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl %eax, (%rsi)
+; CHECK-NEXT:    movq 4(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 4(%rsi)
+; CHECK-NEXT:    movl 12(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 12(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_imm_store:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl $0, (%rdi)
+; DISABLED-NEXT:    movl $1, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_imm_store:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl $0, (%rdi)
+; CHECK-AVX2-NEXT:    movl $1, (%rcx)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX2-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 4(%rsi)
+; CHECK-AVX2-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_imm_store:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl $0, (%rdi)
+; CHECK-AVX512-NEXT:    movl $1, (%rcx)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX512-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 4(%rsi)
+; CHECK-AVX512-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %a = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 0
+  store i32 0, i32* %a, align 4
+  %a1 = getelementptr inbounds %struct.S, %struct.S* %s3, i64 0, i32 0
+  store i32 1, i32* %a1, align 4
+  %0 = bitcast %struct.S* %s2 to i8*
+  %1 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define void @test_nondirect_br(%struct.S* nocapture noalias %s1, %struct.S* nocapture %s2, i32 %x, %struct.S* nocapture %s3, %struct.S* nocapture readonly %s4, i32 %x2) local_unnamed_addr #0 {
+; CHECK-LABEL: test_nondirect_br:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB2_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl %edx, 4(%rdi)
+; CHECK-NEXT:  .LBB2_2: # %if.end
+; CHECK-NEXT:    cmpl $14, %r9d
+; CHECK-NEXT:    jl .LBB2_4
+; CHECK-NEXT:  # %bb.3: # %if.then2
+; CHECK-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-NEXT:  .LBB2_4: # %if.end3
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movq (%rdi), %rax
+; CHECK-NEXT:    movq %rax, (%rsi)
+; CHECK-NEXT:    movl 8(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 8(%rsi)
+; CHECK-NEXT:    movl 12(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 12(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_nondirect_br:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB2_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl %edx, 4(%rdi)
+; DISABLED-NEXT:  .LBB2_2: # %if.end
+; DISABLED-NEXT:    cmpl $14, %r9d
+; DISABLED-NEXT:    jl .LBB2_4
+; DISABLED-NEXT:  # %bb.3: # %if.then2
+; DISABLED-NEXT:    movl %r9d, 12(%rdi)
+; DISABLED-NEXT:  .LBB2_4: # %if.end3
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_nondirect_br:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB2_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX2-NEXT:  .LBB2_2: # %if.end
+; CHECK-AVX2-NEXT:    cmpl $14, %r9d
+; CHECK-AVX2-NEXT:    jl .LBB2_4
+; CHECK-AVX2-NEXT:  # %bb.3: # %if.then2
+; CHECK-AVX2-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX2-NEXT:  .LBB2_4: # %if.end3
+; CHECK-AVX2-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX2-NEXT:    movq (%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX2-NEXT:    movl 8(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 8(%rsi)
+; CHECK-AVX2-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_nondirect_br:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB2_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX512-NEXT:  .LBB2_2: # %if.end
+; CHECK-AVX512-NEXT:    cmpl $14, %r9d
+; CHECK-AVX512-NEXT:    jl .LBB2_4
+; CHECK-AVX512-NEXT:  # %bb.3: # %if.then2
+; CHECK-AVX512-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX512-NEXT:  .LBB2_4: # %if.end3
+; CHECK-AVX512-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX512-NEXT:    movq (%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX512-NEXT:    movl 8(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 8(%rsi)
+; CHECK-AVX512-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 1
+  store i32 %x, i32* %b, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %cmp1 = icmp sgt i32 %x2, 13
+  br i1 %cmp1, label %if.then2, label %if.end3
+
+if.then2:                                         ; preds = %if.end
+  %d = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 3
+  store i32 %x2, i32* %d, align 4
+  br label %if.end3
+
+if.end3:                                          ; preds = %if.then2, %if.end
+  %0 = bitcast %struct.S* %s3 to i8*
+  %1 = bitcast %struct.S* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  %2 = bitcast %struct.S* %s2 to i8*
+  %3 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define void @test_2preds_block(%struct.S* nocapture noalias %s1, %struct.S* nocapture %s2, i32 %x, %struct.S* nocapture %s3, %struct.S* nocapture readonly %s4, i32 %x2) local_unnamed_addr #0 {
+; CHECK-LABEL: test_2preds_block:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB3_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl %edx, 4(%rdi)
+; CHECK-NEXT:  .LBB3_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl %eax, (%rsi)
+; CHECK-NEXT:    movl 4(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 4(%rsi)
+; CHECK-NEXT:    movl 8(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 8(%rsi)
+; CHECK-NEXT:    movl 12(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 12(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_2preds_block:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl %r9d, 12(%rdi)
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB3_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl %edx, 4(%rdi)
+; DISABLED-NEXT:  .LBB3_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_2preds_block:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB3_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX2-NEXT:  .LBB3_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX2-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX2-NEXT:    movl 8(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 8(%rsi)
+; CHECK-AVX2-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_2preds_block:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB3_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl %edx, 4(%rdi)
+; CHECK-AVX512-NEXT:  .LBB3_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX512-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX512-NEXT:    movl 8(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 8(%rsi)
+; CHECK-AVX512-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %d = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 3
+  store i32 %x2, i32* %d, align 4
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 1
+  store i32 %x, i32* %b, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S* %s3 to i8*
+  %1 = bitcast %struct.S* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  %2 = bitcast %struct.S* %s2 to i8*
+  %3 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
+  ret void
+}
+%struct.S2 = type { i64, i64 }
+
+; Function Attrs: nounwind uwtable
+define void @test_type64(%struct.S2* nocapture noalias %s1, %struct.S2* nocapture %s2, i32 %x, %struct.S2* nocapture %s3, %struct.S2* nocapture readonly %s4) local_unnamed_addr #0 {
+; CHECK-LABEL: test_type64:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB4_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movslq %edx, %rax
+; CHECK-NEXT:    movq %rax, 8(%rdi)
+; CHECK-NEXT:  .LBB4_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movq (%rdi), %rax
+; CHECK-NEXT:    movq %rax, (%rsi)
+; CHECK-NEXT:    movq 8(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 8(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_type64:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB4_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movslq %edx, %rax
+; DISABLED-NEXT:    movq %rax, 8(%rdi)
+; DISABLED-NEXT:  .LBB4_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_type64:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB4_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movslq %edx, %rax
+; CHECK-AVX2-NEXT:    movq %rax, 8(%rdi)
+; CHECK-AVX2-NEXT:  .LBB4_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX2-NEXT:    movq (%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX2-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_type64:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB4_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movslq %edx, %rax
+; CHECK-AVX512-NEXT:    movq %rax, 8(%rdi)
+; CHECK-AVX512-NEXT:  .LBB4_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX512-NEXT:    movq (%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX512-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %conv = sext i32 %x to i64
+  %b = getelementptr inbounds %struct.S2, %struct.S2* %s1, i64 0, i32 1
+  store i64 %conv, i64* %b, align 8
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S2* %s3 to i8*
+  %1 = bitcast %struct.S2* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 8, i1 false)
+  %2 = bitcast %struct.S2* %s2 to i8*
+  %3 = bitcast %struct.S2* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 8, i1 false)
+  ret void
+}
+%struct.S3 = type { i64, i8, i8, i16, i32 }
+
+; Function Attrs: noinline nounwind uwtable
+define void @test_mixed_type(%struct.S3* nocapture noalias %s1, %struct.S3* nocapture %s2, i32 %x, %struct.S3* nocapture readnone %s3, %struct.S3* nocapture readnone %s4) local_unnamed_addr #0 {
+; CHECK-LABEL: test_mixed_type:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB5_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movslq %edx, %rax
+; CHECK-NEXT:    movq %rax, (%rdi)
+; CHECK-NEXT:    movb %dl, 8(%rdi)
+; CHECK-NEXT:  .LBB5_2: # %if.end
+; CHECK-NEXT:    movq (%rdi), %rax
+; CHECK-NEXT:    movq %rax, (%rsi)
+; CHECK-NEXT:    movb 8(%rdi), %al
+; CHECK-NEXT:    movb %al, 8(%rsi)
+; CHECK-NEXT:    movl 9(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 9(%rsi)
+; CHECK-NEXT:    movzwl 13(%rdi), %eax
+; CHECK-NEXT:    movw %ax, 13(%rsi)
+; CHECK-NEXT:    movb 15(%rdi), %al
+; CHECK-NEXT:    movb %al, 15(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_mixed_type:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB5_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movslq %edx, %rax
+; DISABLED-NEXT:    movq %rax, (%rdi)
+; DISABLED-NEXT:    movb %dl, 8(%rdi)
+; DISABLED-NEXT:  .LBB5_2: # %if.end
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_mixed_type:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB5_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movslq %edx, %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%rdi)
+; CHECK-AVX2-NEXT:    movb %dl, 8(%rdi)
+; CHECK-AVX2-NEXT:  .LBB5_2: # %if.end
+; CHECK-AVX2-NEXT:    movq (%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX2-NEXT:    movb 8(%rdi), %al
+; CHECK-AVX2-NEXT:    movb %al, 8(%rsi)
+; CHECK-AVX2-NEXT:    movl 9(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 9(%rsi)
+; CHECK-AVX2-NEXT:    movzwl 13(%rdi), %eax
+; CHECK-AVX2-NEXT:    movw %ax, 13(%rsi)
+; CHECK-AVX2-NEXT:    movb 15(%rdi), %al
+; CHECK-AVX2-NEXT:    movb %al, 15(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_mixed_type:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB5_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movslq %edx, %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%rdi)
+; CHECK-AVX512-NEXT:    movb %dl, 8(%rdi)
+; CHECK-AVX512-NEXT:  .LBB5_2: # %if.end
+; CHECK-AVX512-NEXT:    movq (%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX512-NEXT:    movb 8(%rdi), %al
+; CHECK-AVX512-NEXT:    movb %al, 8(%rsi)
+; CHECK-AVX512-NEXT:    movl 9(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 9(%rsi)
+; CHECK-AVX512-NEXT:    movzwl 13(%rdi), %eax
+; CHECK-AVX512-NEXT:    movw %ax, 13(%rsi)
+; CHECK-AVX512-NEXT:    movb 15(%rdi), %al
+; CHECK-AVX512-NEXT:    movb %al, 15(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %conv = sext i32 %x to i64
+  %a = getelementptr inbounds %struct.S3, %struct.S3* %s1, i64 0, i32 0
+  store i64 %conv, i64* %a, align 8
+  %conv1 = trunc i32 %x to i8
+  %b = getelementptr inbounds %struct.S3, %struct.S3* %s1, i64 0, i32 1
+  store i8 %conv1, i8* %b, align 8
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S3* %s2 to i8*
+  %1 = bitcast %struct.S3* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 8, i1 false)
+  ret void
+}
+%struct.S4 = type { i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 }
+
+; Function Attrs: nounwind uwtable
+define void @test_multiple_blocks(%struct.S4* nocapture noalias %s1, %struct.S4* nocapture %s2) local_unnamed_addr #0 {
+; CHECK-LABEL: test_multiple_blocks:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl $0, 4(%rdi)
+; CHECK-NEXT:    movl $0, 36(%rdi)
+; CHECK-NEXT:    movups 16(%rdi), %xmm0
+; CHECK-NEXT:    movups %xmm0, 16(%rsi)
+; CHECK-NEXT:    movl 32(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 32(%rsi)
+; CHECK-NEXT:    movl 36(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 36(%rsi)
+; CHECK-NEXT:    movq 40(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 40(%rsi)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl %eax, (%rsi)
+; CHECK-NEXT:    movl 4(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 4(%rsi)
+; CHECK-NEXT:    movq 8(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 8(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_multiple_blocks:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl $0, 4(%rdi)
+; DISABLED-NEXT:    movl $0, 36(%rdi)
+; DISABLED-NEXT:    movups 16(%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, 16(%rsi)
+; DISABLED-NEXT:    movups 32(%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, 32(%rsi)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_multiple_blocks:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl $0, 4(%rdi)
+; CHECK-AVX2-NEXT:    movl $0, 36(%rdi)
+; CHECK-AVX2-NEXT:    vmovups 16(%rdi), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, 16(%rsi)
+; CHECK-AVX2-NEXT:    movl 32(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 32(%rsi)
+; CHECK-AVX2-NEXT:    movl 36(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 36(%rsi)
+; CHECK-AVX2-NEXT:    movq 40(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 40(%rsi)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX2-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX2-NEXT:    vmovups 8(%rdi), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, 8(%rsi)
+; CHECK-AVX2-NEXT:    movq 24(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 24(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_multiple_blocks:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl $0, 4(%rdi)
+; CHECK-AVX512-NEXT:    movl $0, 36(%rdi)
+; CHECK-AVX512-NEXT:    vmovups 16(%rdi), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, 16(%rsi)
+; CHECK-AVX512-NEXT:    movl 32(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 32(%rsi)
+; CHECK-AVX512-NEXT:    movl 36(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 36(%rsi)
+; CHECK-AVX512-NEXT:    movq 40(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 40(%rsi)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX512-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX512-NEXT:    vmovups 8(%rdi), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, 8(%rsi)
+; CHECK-AVX512-NEXT:    movq 24(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 24(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %b = getelementptr inbounds %struct.S4, %struct.S4* %s1, i64 0, i32 1
+  store i32 0, i32* %b, align 4
+  %b3 = getelementptr inbounds %struct.S4, %struct.S4* %s1, i64 0, i32 9
+  store i32 0, i32* %b3, align 4
+  %0 = bitcast %struct.S4* %s2 to i8*
+  %1 = bitcast %struct.S4* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 48, i32 4, i1 false)
+  ret void
+}
+%struct.S5 = type { i16, i16, i16, i16, i16, i16, i16, i16 }
+
+; Function Attrs: nounwind uwtable
+define void @test_type16(%struct.S5* nocapture noalias %s1, %struct.S5* nocapture %s2, i32 %x, %struct.S5* nocapture %s3, %struct.S5* nocapture readonly %s4) local_unnamed_addr #0 {
+; CHECK-LABEL: test_type16:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB7_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movw %dx, 2(%rdi)
+; CHECK-NEXT:  .LBB7_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movzwl (%rdi), %eax
+; CHECK-NEXT:    movw %ax, (%rsi)
+; CHECK-NEXT:    movzwl 2(%rdi), %eax
+; CHECK-NEXT:    movw %ax, 2(%rsi)
+; CHECK-NEXT:    movq 4(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 4(%rsi)
+; CHECK-NEXT:    movl 12(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 12(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_type16:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB7_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movw %dx, 2(%rdi)
+; DISABLED-NEXT:  .LBB7_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_type16:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB7_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movw %dx, 2(%rdi)
+; CHECK-AVX2-NEXT:  .LBB7_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX2-NEXT:    movzwl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movw %ax, (%rsi)
+; CHECK-AVX2-NEXT:    movzwl 2(%rdi), %eax
+; CHECK-AVX2-NEXT:    movw %ax, 2(%rsi)
+; CHECK-AVX2-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 4(%rsi)
+; CHECK-AVX2-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_type16:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB7_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movw %dx, 2(%rdi)
+; CHECK-AVX512-NEXT:  .LBB7_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rcx)
+; CHECK-AVX512-NEXT:    movzwl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movw %ax, (%rsi)
+; CHECK-AVX512-NEXT:    movzwl 2(%rdi), %eax
+; CHECK-AVX512-NEXT:    movw %ax, 2(%rsi)
+; CHECK-AVX512-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 4(%rsi)
+; CHECK-AVX512-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 12(%rsi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %conv = trunc i32 %x to i16
+  %b = getelementptr inbounds %struct.S5, %struct.S5* %s1, i64 0, i32 1
+  store i16 %conv, i16* %b, align 2
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S5* %s3 to i8*
+  %1 = bitcast %struct.S5* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 2, i1 false)
+  %2 = bitcast %struct.S5* %s2 to i8*
+  %3 = bitcast %struct.S5* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 2, i1 false)
+  ret void
+}
+
+%struct.S6 = type { [4 x i32], i32, i32, i32, i32 }
+
+; Function Attrs: nounwind uwtable
+define void @test_stack(%struct.S6* noalias nocapture sret %agg.result, %struct.S6* byval nocapture readnone align 8 %s1, %struct.S6* byval nocapture align 8 %s2, i32 %x) local_unnamed_addr #0 {
+; CHECK-LABEL: test_stack:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl %esi, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%rdi)
+; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-NEXT:    movq %rax, 16(%rdi)
+; CHECK-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl %eax, 24(%rdi)
+; CHECK-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-NEXT:    movl %eax, 28(%rdi)
+; CHECK-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-NEXT:    movl {{[0-9]+}}(%rsp), %ecx
+; CHECK-NEXT:    movl {{[0-9]+}}(%rsp), %edx
+; CHECK-NEXT:    movaps %xmm0, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movq %rax, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movl %ecx, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movl %edx, {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_stack:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl %esi, {{[0-9]+}}(%rsp)
+; DISABLED-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%rdi)
+; DISABLED-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm0
+; DISABLED-NEXT:    movups %xmm0, 16(%rdi)
+; DISABLED-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm0
+; DISABLED-NEXT:    movaps {{[0-9]+}}(%rsp), %xmm1
+; DISABLED-NEXT:    movaps %xmm0, {{[0-9]+}}(%rsp)
+; DISABLED-NEXT:    movaps %xmm1, {{[0-9]+}}(%rsp)
+; DISABLED-NEXT:    movq %rdi, %rax
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_stack:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl %esi, {{[0-9]+}}(%rsp)
+; CHECK-AVX2-NEXT:    vmovups {{[0-9]+}}(%rsp), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%rdi)
+; CHECK-AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 16(%rdi)
+; CHECK-AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 24(%rdi)
+; CHECK-AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 28(%rdi)
+; CHECK-AVX2-NEXT:    vmovups {{[0-9]+}}(%rsp), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, {{[0-9]+}}(%rsp)
+; CHECK-AVX2-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-AVX2-NEXT:    movq %rax, {{[0-9]+}}(%rsp)
+; CHECK-AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX2-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
+; CHECK-AVX2-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX2-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
+; CHECK-AVX2-NEXT:    movq %rdi, %rax
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_stack:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl %esi, {{[0-9]+}}(%rsp)
+; CHECK-AVX512-NEXT:    vmovups {{[0-9]+}}(%rsp), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%rdi)
+; CHECK-AVX512-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 16(%rdi)
+; CHECK-AVX512-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 24(%rdi)
+; CHECK-AVX512-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 28(%rdi)
+; CHECK-AVX512-NEXT:    vmovups {{[0-9]+}}(%rsp), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, {{[0-9]+}}(%rsp)
+; CHECK-AVX512-NEXT:    movq {{[0-9]+}}(%rsp), %rax
+; CHECK-AVX512-NEXT:    movq %rax, {{[0-9]+}}(%rsp)
+; CHECK-AVX512-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX512-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
+; CHECK-AVX512-NEXT:    movl {{[0-9]+}}(%rsp), %eax
+; CHECK-AVX512-NEXT:    movl %eax, {{[0-9]+}}(%rsp)
+; CHECK-AVX512-NEXT:    movq %rdi, %rax
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %s6.sroa.0.0..sroa_cast1 = bitcast %struct.S6* %s2 to i8*
+  %s6.sroa.3.0..sroa_idx4 = getelementptr inbounds %struct.S6, %struct.S6* %s2, i64 0, i32 3
+  store i32 %x, i32* %s6.sroa.3.0..sroa_idx4, align 8
+  %0 = bitcast %struct.S6* %agg.result to i8*
+  %s6.sroa.0.0..sroa_cast2 = bitcast %struct.S6* %s1 to i8*
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* nonnull %s6.sroa.0.0..sroa_cast1, i64 32, i32 4, i1 false)
+  call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull %s6.sroa.0.0..sroa_cast2, i8* nonnull %s6.sroa.0.0..sroa_cast1, i64 32, i32 4, i1 false)
+
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define void @test_limit_all(%struct.S* noalias  %s1, %struct.S* nocapture %s2, i32 %x, %struct.S* nocapture %s3, %struct.S* nocapture readonly %s4, i32 %x2) local_unnamed_addr #0 {
+; CHECK-LABEL: test_limit_all:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    pushq %rbp
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    pushq %r15
+; CHECK-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-NEXT:    pushq %r14
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    pushq %r12
+; CHECK-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset %rbx, -48
+; CHECK-NEXT:    .cfi_offset %r12, -40
+; CHECK-NEXT:    .cfi_offset %r14, -32
+; CHECK-NEXT:    .cfi_offset %r15, -24
+; CHECK-NEXT:    .cfi_offset %rbp, -16
+; CHECK-NEXT:    movq %r8, %r15
+; CHECK-NEXT:    movq %rcx, %r14
+; CHECK-NEXT:    movl %edx, %ebp
+; CHECK-NEXT:    movq %rsi, %r12
+; CHECK-NEXT:    movq %rdi, %rbx
+; CHECK-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:    cmpl $18, %ebp
+; CHECK-NEXT:    jl .LBB9_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl %ebp, 4(%rbx)
+; CHECK-NEXT:    movq %rbx, %rdi
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB9_2: # %if.end
+; CHECK-NEXT:    movups (%r15), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%r14)
+; CHECK-NEXT:    movups (%rbx), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%r12)
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    popq %r12
+; CHECK-NEXT:    popq %r14
+; CHECK-NEXT:    popq %r15
+; CHECK-NEXT:    popq %rbp
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_limit_all:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    pushq %rbp
+; DISABLED-NEXT:    .cfi_def_cfa_offset 16
+; DISABLED-NEXT:    pushq %r15
+; DISABLED-NEXT:    .cfi_def_cfa_offset 24
+; DISABLED-NEXT:    pushq %r14
+; DISABLED-NEXT:    .cfi_def_cfa_offset 32
+; DISABLED-NEXT:    pushq %r12
+; DISABLED-NEXT:    .cfi_def_cfa_offset 40
+; DISABLED-NEXT:    pushq %rbx
+; DISABLED-NEXT:    .cfi_def_cfa_offset 48
+; DISABLED-NEXT:    .cfi_offset %rbx, -48
+; DISABLED-NEXT:    .cfi_offset %r12, -40
+; DISABLED-NEXT:    .cfi_offset %r14, -32
+; DISABLED-NEXT:    .cfi_offset %r15, -24
+; DISABLED-NEXT:    .cfi_offset %rbp, -16
+; DISABLED-NEXT:    movq %r8, %r15
+; DISABLED-NEXT:    movq %rcx, %r14
+; DISABLED-NEXT:    movl %edx, %ebp
+; DISABLED-NEXT:    movq %rsi, %r12
+; DISABLED-NEXT:    movq %rdi, %rbx
+; DISABLED-NEXT:    movl %r9d, 12(%rdi)
+; DISABLED-NEXT:    callq bar
+; DISABLED-NEXT:    cmpl $18, %ebp
+; DISABLED-NEXT:    jl .LBB9_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl %ebp, 4(%rbx)
+; DISABLED-NEXT:    movq %rbx, %rdi
+; DISABLED-NEXT:    callq bar
+; DISABLED-NEXT:  .LBB9_2: # %if.end
+; DISABLED-NEXT:    movups (%r15), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%r14)
+; DISABLED-NEXT:    movups (%rbx), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%r12)
+; DISABLED-NEXT:    popq %rbx
+; DISABLED-NEXT:    popq %r12
+; DISABLED-NEXT:    popq %r14
+; DISABLED-NEXT:    popq %r15
+; DISABLED-NEXT:    popq %rbp
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_limit_all:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    pushq %rbp
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-AVX2-NEXT:    pushq %r15
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-AVX2-NEXT:    pushq %r14
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-AVX2-NEXT:    pushq %r12
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-AVX2-NEXT:    pushq %rbx
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-AVX2-NEXT:    .cfi_offset %rbx, -48
+; CHECK-AVX2-NEXT:    .cfi_offset %r12, -40
+; CHECK-AVX2-NEXT:    .cfi_offset %r14, -32
+; CHECK-AVX2-NEXT:    .cfi_offset %r15, -24
+; CHECK-AVX2-NEXT:    .cfi_offset %rbp, -16
+; CHECK-AVX2-NEXT:    movq %r8, %r15
+; CHECK-AVX2-NEXT:    movq %rcx, %r14
+; CHECK-AVX2-NEXT:    movl %edx, %ebp
+; CHECK-AVX2-NEXT:    movq %rsi, %r12
+; CHECK-AVX2-NEXT:    movq %rdi, %rbx
+; CHECK-AVX2-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX2-NEXT:    callq bar
+; CHECK-AVX2-NEXT:    cmpl $18, %ebp
+; CHECK-AVX2-NEXT:    jl .LBB9_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl %ebp, 4(%rbx)
+; CHECK-AVX2-NEXT:    movq %rbx, %rdi
+; CHECK-AVX2-NEXT:    callq bar
+; CHECK-AVX2-NEXT:  .LBB9_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r15), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%r14)
+; CHECK-AVX2-NEXT:    vmovups (%rbx), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%r12)
+; CHECK-AVX2-NEXT:    popq %rbx
+; CHECK-AVX2-NEXT:    popq %r12
+; CHECK-AVX2-NEXT:    popq %r14
+; CHECK-AVX2-NEXT:    popq %r15
+; CHECK-AVX2-NEXT:    popq %rbp
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_limit_all:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    pushq %rbp
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-AVX512-NEXT:    pushq %r15
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-AVX512-NEXT:    pushq %r14
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-AVX512-NEXT:    pushq %r12
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-AVX512-NEXT:    pushq %rbx
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-AVX512-NEXT:    .cfi_offset %rbx, -48
+; CHECK-AVX512-NEXT:    .cfi_offset %r12, -40
+; CHECK-AVX512-NEXT:    .cfi_offset %r14, -32
+; CHECK-AVX512-NEXT:    .cfi_offset %r15, -24
+; CHECK-AVX512-NEXT:    .cfi_offset %rbp, -16
+; CHECK-AVX512-NEXT:    movq %r8, %r15
+; CHECK-AVX512-NEXT:    movq %rcx, %r14
+; CHECK-AVX512-NEXT:    movl %edx, %ebp
+; CHECK-AVX512-NEXT:    movq %rsi, %r12
+; CHECK-AVX512-NEXT:    movq %rdi, %rbx
+; CHECK-AVX512-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX512-NEXT:    callq bar
+; CHECK-AVX512-NEXT:    cmpl $18, %ebp
+; CHECK-AVX512-NEXT:    jl .LBB9_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl %ebp, 4(%rbx)
+; CHECK-AVX512-NEXT:    movq %rbx, %rdi
+; CHECK-AVX512-NEXT:    callq bar
+; CHECK-AVX512-NEXT:  .LBB9_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r15), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%r14)
+; CHECK-AVX512-NEXT:    vmovups (%rbx), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%r12)
+; CHECK-AVX512-NEXT:    popq %rbx
+; CHECK-AVX512-NEXT:    popq %r12
+; CHECK-AVX512-NEXT:    popq %r14
+; CHECK-AVX512-NEXT:    popq %r15
+; CHECK-AVX512-NEXT:    popq %rbp
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %d = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 3
+  store i32 %x2, i32* %d, align 4
+  tail call void @bar(%struct.S* %s1) #3
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 1
+  store i32 %x, i32* %b, align 4
+  tail call void @bar(%struct.S* nonnull %s1) #3
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S* %s3 to i8*
+  %1 = bitcast %struct.S* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  %2 = bitcast %struct.S* %s2 to i8*
+  %3 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define void @test_limit_one_pred(%struct.S* noalias %s1, %struct.S* nocapture %s2, i32 %x, %struct.S* nocapture %s3, %struct.S* nocapture readonly %s4, i32 %x2) local_unnamed_addr #0 {
+; CHECK-LABEL: test_limit_one_pred:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    pushq %r15
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    pushq %r14
+; CHECK-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-NEXT:    pushq %r12
+; CHECK-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-NEXT:    pushq %rax
+; CHECK-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-NEXT:    .cfi_offset %rbx, -40
+; CHECK-NEXT:    .cfi_offset %r12, -32
+; CHECK-NEXT:    .cfi_offset %r14, -24
+; CHECK-NEXT:    .cfi_offset %r15, -16
+; CHECK-NEXT:    movq %r8, %r12
+; CHECK-NEXT:    movq %rcx, %r15
+; CHECK-NEXT:    movq %rsi, %r14
+; CHECK-NEXT:    movq %rdi, %rbx
+; CHECK-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB10_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl %edx, 4(%rbx)
+; CHECK-NEXT:    movq %rbx, %rdi
+; CHECK-NEXT:    callq bar
+; CHECK-NEXT:  .LBB10_2: # %if.end
+; CHECK-NEXT:    movups (%r12), %xmm0
+; CHECK-NEXT:    movups %xmm0, (%r15)
+; CHECK-NEXT:    movq (%rbx), %rax
+; CHECK-NEXT:    movq %rax, (%r14)
+; CHECK-NEXT:    movl 8(%rbx), %eax
+; CHECK-NEXT:    movl %eax, 8(%r14)
+; CHECK-NEXT:    movl 12(%rbx), %eax
+; CHECK-NEXT:    movl %eax, 12(%r14)
+; CHECK-NEXT:    addq $8, %rsp
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    popq %r12
+; CHECK-NEXT:    popq %r14
+; CHECK-NEXT:    popq %r15
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_limit_one_pred:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    pushq %r15
+; DISABLED-NEXT:    .cfi_def_cfa_offset 16
+; DISABLED-NEXT:    pushq %r14
+; DISABLED-NEXT:    .cfi_def_cfa_offset 24
+; DISABLED-NEXT:    pushq %r12
+; DISABLED-NEXT:    .cfi_def_cfa_offset 32
+; DISABLED-NEXT:    pushq %rbx
+; DISABLED-NEXT:    .cfi_def_cfa_offset 40
+; DISABLED-NEXT:    pushq %rax
+; DISABLED-NEXT:    .cfi_def_cfa_offset 48
+; DISABLED-NEXT:    .cfi_offset %rbx, -40
+; DISABLED-NEXT:    .cfi_offset %r12, -32
+; DISABLED-NEXT:    .cfi_offset %r14, -24
+; DISABLED-NEXT:    .cfi_offset %r15, -16
+; DISABLED-NEXT:    movq %r8, %r15
+; DISABLED-NEXT:    movq %rcx, %r14
+; DISABLED-NEXT:    movq %rsi, %r12
+; DISABLED-NEXT:    movq %rdi, %rbx
+; DISABLED-NEXT:    movl %r9d, 12(%rdi)
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB10_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl %edx, 4(%rbx)
+; DISABLED-NEXT:    movq %rbx, %rdi
+; DISABLED-NEXT:    callq bar
+; DISABLED-NEXT:  .LBB10_2: # %if.end
+; DISABLED-NEXT:    movups (%r15), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%r14)
+; DISABLED-NEXT:    movups (%rbx), %xmm0
+; DISABLED-NEXT:    movups %xmm0, (%r12)
+; DISABLED-NEXT:    addq $8, %rsp
+; DISABLED-NEXT:    popq %rbx
+; DISABLED-NEXT:    popq %r12
+; DISABLED-NEXT:    popq %r14
+; DISABLED-NEXT:    popq %r15
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_limit_one_pred:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    pushq %r15
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-AVX2-NEXT:    pushq %r14
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-AVX2-NEXT:    pushq %r12
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-AVX2-NEXT:    pushq %rbx
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-AVX2-NEXT:    pushq %rax
+; CHECK-AVX2-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-AVX2-NEXT:    .cfi_offset %rbx, -40
+; CHECK-AVX2-NEXT:    .cfi_offset %r12, -32
+; CHECK-AVX2-NEXT:    .cfi_offset %r14, -24
+; CHECK-AVX2-NEXT:    .cfi_offset %r15, -16
+; CHECK-AVX2-NEXT:    movq %r8, %r12
+; CHECK-AVX2-NEXT:    movq %rcx, %r15
+; CHECK-AVX2-NEXT:    movq %rsi, %r14
+; CHECK-AVX2-NEXT:    movq %rdi, %rbx
+; CHECK-AVX2-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB10_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl %edx, 4(%rbx)
+; CHECK-AVX2-NEXT:    movq %rbx, %rdi
+; CHECK-AVX2-NEXT:    callq bar
+; CHECK-AVX2-NEXT:  .LBB10_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r12), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, (%r15)
+; CHECK-AVX2-NEXT:    movq (%rbx), %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%r14)
+; CHECK-AVX2-NEXT:    movl 8(%rbx), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 8(%r14)
+; CHECK-AVX2-NEXT:    movl 12(%rbx), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 12(%r14)
+; CHECK-AVX2-NEXT:    addq $8, %rsp
+; CHECK-AVX2-NEXT:    popq %rbx
+; CHECK-AVX2-NEXT:    popq %r12
+; CHECK-AVX2-NEXT:    popq %r14
+; CHECK-AVX2-NEXT:    popq %r15
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_limit_one_pred:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    pushq %r15
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-AVX512-NEXT:    pushq %r14
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 24
+; CHECK-AVX512-NEXT:    pushq %r12
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 32
+; CHECK-AVX512-NEXT:    pushq %rbx
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 40
+; CHECK-AVX512-NEXT:    pushq %rax
+; CHECK-AVX512-NEXT:    .cfi_def_cfa_offset 48
+; CHECK-AVX512-NEXT:    .cfi_offset %rbx, -40
+; CHECK-AVX512-NEXT:    .cfi_offset %r12, -32
+; CHECK-AVX512-NEXT:    .cfi_offset %r14, -24
+; CHECK-AVX512-NEXT:    .cfi_offset %r15, -16
+; CHECK-AVX512-NEXT:    movq %r8, %r12
+; CHECK-AVX512-NEXT:    movq %rcx, %r15
+; CHECK-AVX512-NEXT:    movq %rsi, %r14
+; CHECK-AVX512-NEXT:    movq %rdi, %rbx
+; CHECK-AVX512-NEXT:    movl %r9d, 12(%rdi)
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB10_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl %edx, 4(%rbx)
+; CHECK-AVX512-NEXT:    movq %rbx, %rdi
+; CHECK-AVX512-NEXT:    callq bar
+; CHECK-AVX512-NEXT:  .LBB10_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r12), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, (%r15)
+; CHECK-AVX512-NEXT:    movq (%rbx), %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%r14)
+; CHECK-AVX512-NEXT:    movl 8(%rbx), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 8(%r14)
+; CHECK-AVX512-NEXT:    movl 12(%rbx), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 12(%r14)
+; CHECK-AVX512-NEXT:    addq $8, %rsp
+; CHECK-AVX512-NEXT:    popq %rbx
+; CHECK-AVX512-NEXT:    popq %r12
+; CHECK-AVX512-NEXT:    popq %r14
+; CHECK-AVX512-NEXT:    popq %r15
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %d = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 3
+  store i32 %x2, i32* %d, align 4
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S, %struct.S* %s1, i64 0, i32 1
+  store i32 %x, i32* %b, align 4
+  tail call void @bar(%struct.S* nonnull %s1) #3
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S* %s3 to i8*
+  %1 = bitcast %struct.S* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 16, i32 4, i1 false)
+  %2 = bitcast %struct.S* %s2 to i8*
+  %3 = bitcast %struct.S* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 16, i32 4, i1 false)
+  ret void
+}
+
+
+declare void @bar(%struct.S*) local_unnamed_addr #1
+
+
+; Function Attrs: argmemonly nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1) #1
+
+attributes #0 = { nounwind uwtable "target-cpu"="x86-64" }
+
+%struct.S7 = type { float, float, float , float, float, float, float, float }
+
+; Function Attrs: nounwind uwtable
+define void @test_conditional_block_float(%struct.S7* nocapture noalias %s1, %struct.S7* nocapture %s2, i32 %x, %struct.S7* nocapture %s3, %struct.S7* nocapture readonly %s4, float %y) local_unnamed_addr #0 {
+; CHECK-LABEL: test_conditional_block_float:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB11_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movl $1065353216, 4(%rdi) # imm = 0x3F800000
+; CHECK-NEXT:  .LBB11_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups 16(%r8), %xmm1
+; CHECK-NEXT:    movups %xmm1, 16(%rcx)
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl 4(%rdi), %ecx
+; CHECK-NEXT:    movq 8(%rdi), %rdx
+; CHECK-NEXT:    movups 16(%rdi), %xmm0
+; CHECK-NEXT:    movups %xmm0, 16(%rsi)
+; CHECK-NEXT:    movl %eax, (%rsi)
+; CHECK-NEXT:    movl %ecx, 4(%rsi)
+; CHECK-NEXT:    movq %rdx, 8(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_conditional_block_float:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB11_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movl $1065353216, 4(%rdi) # imm = 0x3F800000
+; DISABLED-NEXT:  .LBB11_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups 16(%r8), %xmm1
+; DISABLED-NEXT:    movups %xmm1, 16(%rcx)
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups 16(%rdi), %xmm1
+; DISABLED-NEXT:    movups %xmm1, 16(%rsi)
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_conditional_block_float:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB11_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movl $1065353216, 4(%rdi) # imm = 0x3F800000
+; CHECK-AVX2-NEXT:  .LBB11_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %ymm0
+; CHECK-AVX2-NEXT:    vmovups %ymm0, (%rcx)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX2-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX2-NEXT:    vmovups 8(%rdi), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, 8(%rsi)
+; CHECK-AVX2-NEXT:    movq 24(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 24(%rsi)
+; CHECK-AVX2-NEXT:    vzeroupper
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_conditional_block_float:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB11_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movl $1065353216, 4(%rdi) # imm = 0x3F800000
+; CHECK-AVX512-NEXT:  .LBB11_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %ymm0
+; CHECK-AVX512-NEXT:    vmovups %ymm0, (%rcx)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, (%rsi)
+; CHECK-AVX512-NEXT:    movl 4(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 4(%rsi)
+; CHECK-AVX512-NEXT:    vmovups 8(%rdi), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, 8(%rsi)
+; CHECK-AVX512-NEXT:    movq 24(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 24(%rsi)
+; CHECK-AVX512-NEXT:    vzeroupper
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S7, %struct.S7* %s1, i64 0, i32 1
+  store float 1.0, float* %b, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S7* %s3 to i8*
+  %1 = bitcast %struct.S7* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 32, i32 4, i1 false)
+  %2 = bitcast %struct.S7* %s2 to i8*
+  %3 = bitcast %struct.S7* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 32, i32 4, i1 false)
+  ret void
+}
+
+%struct.S8 = type { i64, i64, i64, i64, i64, i64 }
+
+; Function Attrs: nounwind uwtable
+define void @test_conditional_block_ymm(%struct.S8* nocapture noalias %s1, %struct.S8* nocapture %s2, i32 %x, %struct.S8* nocapture %s3, %struct.S8* nocapture readonly %s4) local_unnamed_addr #0 {
+; CHECK-LABEL: test_conditional_block_ymm:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    cmpl $18, %edx
+; CHECK-NEXT:    jl .LBB12_2
+; CHECK-NEXT:  # %bb.1: # %if.then
+; CHECK-NEXT:    movq $1, 8(%rdi)
+; CHECK-NEXT:  .LBB12_2: # %if.end
+; CHECK-NEXT:    movups (%r8), %xmm0
+; CHECK-NEXT:    movups 16(%r8), %xmm1
+; CHECK-NEXT:    movups %xmm1, 16(%rcx)
+; CHECK-NEXT:    movups %xmm0, (%rcx)
+; CHECK-NEXT:    movq (%rdi), %rax
+; CHECK-NEXT:    movq 8(%rdi), %rcx
+; CHECK-NEXT:    movups 16(%rdi), %xmm0
+; CHECK-NEXT:    movups %xmm0, 16(%rsi)
+; CHECK-NEXT:    movq %rax, (%rsi)
+; CHECK-NEXT:    movq %rcx, 8(%rsi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_conditional_block_ymm:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    cmpl $18, %edx
+; DISABLED-NEXT:    jl .LBB12_2
+; DISABLED-NEXT:  # %bb.1: # %if.then
+; DISABLED-NEXT:    movq $1, 8(%rdi)
+; DISABLED-NEXT:  .LBB12_2: # %if.end
+; DISABLED-NEXT:    movups (%r8), %xmm0
+; DISABLED-NEXT:    movups 16(%r8), %xmm1
+; DISABLED-NEXT:    movups %xmm1, 16(%rcx)
+; DISABLED-NEXT:    movups %xmm0, (%rcx)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups 16(%rdi), %xmm1
+; DISABLED-NEXT:    movups %xmm1, 16(%rsi)
+; DISABLED-NEXT:    movups %xmm0, (%rsi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_conditional_block_ymm:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    cmpl $18, %edx
+; CHECK-AVX2-NEXT:    jl .LBB12_2
+; CHECK-AVX2-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX2-NEXT:    movq $1, 8(%rdi)
+; CHECK-AVX2-NEXT:  .LBB12_2: # %if.end
+; CHECK-AVX2-NEXT:    vmovups (%r8), %ymm0
+; CHECK-AVX2-NEXT:    vmovups %ymm0, (%rcx)
+; CHECK-AVX2-NEXT:    movq (%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX2-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX2-NEXT:    vmovups 16(%rdi), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, 16(%rsi)
+; CHECK-AVX2-NEXT:    vzeroupper
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_conditional_block_ymm:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    cmpl $18, %edx
+; CHECK-AVX512-NEXT:    jl .LBB12_2
+; CHECK-AVX512-NEXT:  # %bb.1: # %if.then
+; CHECK-AVX512-NEXT:    movq $1, 8(%rdi)
+; CHECK-AVX512-NEXT:  .LBB12_2: # %if.end
+; CHECK-AVX512-NEXT:    vmovups (%r8), %ymm0
+; CHECK-AVX512-NEXT:    vmovups %ymm0, (%rcx)
+; CHECK-AVX512-NEXT:    movq (%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, (%rsi)
+; CHECK-AVX512-NEXT:    movq 8(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 8(%rsi)
+; CHECK-AVX512-NEXT:    vmovups 16(%rdi), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, 16(%rsi)
+; CHECK-AVX512-NEXT:    vzeroupper
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %cmp = icmp sgt i32 %x, 17
+  br i1 %cmp, label %if.then, label %if.end
+
+if.then:                                          ; preds = %entry
+  %b = getelementptr inbounds %struct.S8, %struct.S8* %s1, i64 0, i32 1
+  store i64 1, i64* %b, align 4
+  br label %if.end
+
+if.end:                                           ; preds = %if.then, %entry
+  %0 = bitcast %struct.S8* %s3 to i8*
+  %1 = bitcast %struct.S8* %s4 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %0, i8* %1, i64 32, i32 4, i1 false)
+  %2 = bitcast %struct.S8* %s2 to i8*
+  %3 = bitcast %struct.S8* %s1 to i8*
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* %2, i8* %3, i64 32, i32 4, i1 false)
+  ret void
+}
+
+define dso_local void @test_alias(i8* nocapture %A, i32 %x) local_unnamed_addr #0 {
+; CHECK-LABEL: test_alias:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl %esi, (%rdi)
+; CHECK-NEXT:    movups (%rdi), %xmm0
+; CHECK-NEXT:    movups %xmm0, 4(%rdi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_alias:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl %esi, (%rdi)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, 4(%rdi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_alias:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl %esi, (%rdi)
+; CHECK-AVX2-NEXT:    vmovups (%rdi), %xmm0
+; CHECK-AVX2-NEXT:    vmovups %xmm0, 4(%rdi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_alias:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl %esi, (%rdi)
+; CHECK-AVX512-NEXT:    vmovups (%rdi), %xmm0
+; CHECK-AVX512-NEXT:    vmovups %xmm0, 4(%rdi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %a = bitcast i8* %A to i32*
+  store i32 %x, i32* %a, align 4
+  %add.ptr = getelementptr inbounds i8, i8* %A, i64 4
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 4 %add.ptr, i8* align 4 %A, i64 16, i32 4, i1 false)
+  ret void
+}
+
+; Function Attrs: nounwind uwtable
+define dso_local void @test_noalias(i8* nocapture %A, i32 %x) local_unnamed_addr #0 {
+; CHECK-LABEL: test_noalias:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movl %esi, (%rdi)
+; CHECK-NEXT:    movl (%rdi), %eax
+; CHECK-NEXT:    movl %eax, 20(%rdi)
+; CHECK-NEXT:    movq 4(%rdi), %rax
+; CHECK-NEXT:    movq %rax, 24(%rdi)
+; CHECK-NEXT:    movl 12(%rdi), %eax
+; CHECK-NEXT:    movl %eax, 32(%rdi)
+; CHECK-NEXT:    retq
+;
+; DISABLED-LABEL: test_noalias:
+; DISABLED:       # %bb.0: # %entry
+; DISABLED-NEXT:    movl %esi, (%rdi)
+; DISABLED-NEXT:    movups (%rdi), %xmm0
+; DISABLED-NEXT:    movups %xmm0, 20(%rdi)
+; DISABLED-NEXT:    retq
+;
+; CHECK-AVX2-LABEL: test_noalias:
+; CHECK-AVX2:       # %bb.0: # %entry
+; CHECK-AVX2-NEXT:    movl %esi, (%rdi)
+; CHECK-AVX2-NEXT:    movl (%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 20(%rdi)
+; CHECK-AVX2-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX2-NEXT:    movq %rax, 24(%rdi)
+; CHECK-AVX2-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX2-NEXT:    movl %eax, 32(%rdi)
+; CHECK-AVX2-NEXT:    retq
+;
+; CHECK-AVX512-LABEL: test_noalias:
+; CHECK-AVX512:       # %bb.0: # %entry
+; CHECK-AVX512-NEXT:    movl %esi, (%rdi)
+; CHECK-AVX512-NEXT:    movl (%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 20(%rdi)
+; CHECK-AVX512-NEXT:    movq 4(%rdi), %rax
+; CHECK-AVX512-NEXT:    movq %rax, 24(%rdi)
+; CHECK-AVX512-NEXT:    movl 12(%rdi), %eax
+; CHECK-AVX512-NEXT:    movl %eax, 32(%rdi)
+; CHECK-AVX512-NEXT:    retq
+entry:
+  %a = bitcast i8* %A to i32*
+  store i32 %x, i32* %a, align 4
+  %add.ptr = getelementptr inbounds i8, i8* %A, i64 20
+  tail call void @llvm.memcpy.p0i8.p0i8.i64(i8* nonnull align 4 %add.ptr, i8* align 4 %A, i64 16, i32 4, i1 false)
+  ret void
+}
+
+
+