[X86] Fix bug in legalize vector types - Split large loads

[aymanmus]

Summary

When splitting a large load to smaller legally-typed loads, the last load should be padded to reach the size of the previous one so a CONCAT_VECTORS node could reunite them again.
The code currently pads the last load to reach the size of the first load (instead of the previous).

[RKSimon]

Does this fix PR27708 as well?

[aymanmus]

The bug was filed for Clang's 3.8 version, it passed with clang-4.0 and clang-ToT (top of trunc).
Seems like the issue was fixed while ago.

[RKSimon]

In D38495#891561, @aymanmus wrote:

The bug was filed for Clang's 3.8 version, it passed with clang-4.0 and clang-ToT (top of trunc).
Seems like the issue was fixed while ago.

Confirmed I added a test case at rL315186

[RKSimon]

Possibly add a test case for PR34657 as well? @zvi has already reduced much of it

test/CodeGen/X86/pr34653.ll

1 ↗

(On Diff #117482)

Add a triple, also I realise its an assertion test but it doesn't usually hurt to try and create FileChecks for a test

[aymanmus]

Adding triple & a test case for PR23657

[RKSimon]

test/CodeGen/X86/pr34653-pr34657.ll

314

I have a feeling this will reduce further - bugpoint typically stops a little early.

[aymanmus]

Manually reducing "bugpointed" test.

[RKSimon]

test/CodeGen/X86/pr34653-pr34657.ll

4

Why the difference in attributes?

[aymanmus]

test/CodeGen/X86/pr34653-pr34657.ll

4

To reproduce both of the bugs as they were filed.

[RKSimon]

LGTM once the tests have been split up

test/CodeGen/X86/pr34653-pr34657.ll

4

OK - please can you just split these into 2 test files. pr34653.ll and pr34657.ll

[aymanmus]

test/CodeGen/X86/pr34653-pr34657.ll

4

Done. Thanks

lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

while (LdWidth > 0) {
     } else {
       L = DAG.getLoad(NewVT, dl, Chain, BasePtr,
                       LD->getPointerInfo().getWithOffset(Offset),
                       MinAlign(Align, Increment), MMOFlags, AAInfo);
       LdChain.push_back(L.getValue(1));
     }
 
     LdOps.push_back(L);
-
+    LdOp = L;
 
     LdWidth -= NewVTWidth;
   }
 
   // Build the vector from the load operations.
   unsigned End = LdOps.size();
   if (!LdOps[0].getValueType().isVector())
     // All the loads are scalar loads.

test/CodeGen/X86/pr34653-pr34657.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc %s -O0 -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s -check-prefix=CHECKO0
+; RUN: llc %s -O2 -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s -check-prefix=CHECKO2
+
+declare fastcc <38 x double> @test()
+
+define void @pr34653() {
+; CHECKO0-LABEL: pr34653:
+; CHECKO0:       # BB#0: # %entry
+; CHECKO0-NEXT:    pushq %rbp
+; CHECKO0-NEXT:    .cfi_def_cfa_offset 16
+; CHECKO0-NEXT:    .cfi_offset %rbp, -16
+; CHECKO0-NEXT:    movq %rsp, %rbp
+; CHECKO0-NEXT:    .cfi_def_cfa_register %rbp
+; CHECKO0-NEXT:    andq $-512, %rsp # imm = 0xFE00
+; CHECKO0-NEXT:    subq $2048, %rsp # imm = 0x800
+; CHECKO0-NEXT:    leaq {{[0-9]+}}(%rsp), %rdi
+; CHECKO0-NEXT:    callq test
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, %xmm1
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %ymm2
+; CHECKO0-NEXT:    vextractf128 $1, %ymm2, %xmm3
+; CHECKO0-NEXT:    vmovaps %xmm3, %xmm4
+; CHECKO0-NEXT:    vmovaps %xmm2, %xmm5
+; CHECKO0-NEXT:    vmovaps %xmm5, %xmm6
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %zmm7
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %zmm8
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %zmm9
+; CHECKO0-NEXT:    vmovupd {{[0-9]+}}(%rsp), %zmm10
+; CHECKO0-NEXT:    vextractf32x4 $3, %zmm10, %xmm11
+; CHECKO0-NEXT:    vmovaps %xmm11, %xmm12
+; CHECKO0-NEXT:    vextractf32x4 $2, %zmm10, %xmm13
+; CHECKO0-NEXT:    vmovaps %xmm13, %xmm14
+; CHECKO0-NEXT:    vmovaps %xmm10, %xmm15
+; CHECKO0-NEXT:    vmovaps %xmm15, %xmm2
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $3, %zmm9, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $2, %zmm9, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm9, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $3, %zmm8, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $2, %zmm8, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm8, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $3, %zmm7, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vextractf32x4 $2, %zmm7, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm7, %xmm0
+; CHECKO0-NEXT:    vmovaps %xmm0, {{[0-9]+}}(%rsp) # 16-byte Spill
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm3 = xmm3[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm5 = xmm5[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm11 = xmm11[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm13 = xmm13[1,0]
+; CHECKO0-NEXT:    # kill: %YMM10<def> %YMM10<kill> %ZMM10<kill>
+; CHECKO0-NEXT:    vextractf128 $1, %ymm10, %xmm10
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm10, %xmm0
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm15 = xmm15[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    # kill: %YMM9<def> %YMM9<kill> %ZMM9<kill>
+; CHECKO0-NEXT:    vextractf128 $1, %ymm9, %xmm9
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm9, %xmm0
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    # kill: %YMM8<def> %YMM8<kill> %ZMM8<kill>
+; CHECKO0-NEXT:    vextractf128 $1, %ymm8, %xmm8
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm8, %xmm0
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    # kill: %YMM7<def> %YMM7<kill> %ZMM7<kill>
+; CHECKO0-NEXT:    vextractf128 $1, %ymm7, %xmm7
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps %xmm7, %xmm0
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovaps {{[0-9]+}}(%rsp), %xmm0 # 16-byte Reload
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm10 = xmm10[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm9 = xmm9[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm8 = xmm8[1,0]
+; CHECKO0-NEXT:    vpermilpd {{.*#+}} xmm7 = xmm7[1,0]
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd {{[0-9]+}}(%rsp), %xmm0 # 8-byte Reload
+; CHECKO0-NEXT:    # xmm0 = mem[0],zero
+; CHECKO0-NEXT:    vmovsd %xmm8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm13, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm1, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm14, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm2, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm4, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm10, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm15, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm11, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm3, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm6, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm5, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm12, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm0, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    vmovsd %xmm7, {{[0-9]+}}(%rsp) # 8-byte Spill
+; CHECKO0-NEXT:    movq %rbp, %rsp
+; CHECKO0-NEXT:    popq %rbp
+; CHECKO0-NEXT:    vzeroupper
+; CHECKO0-NEXT:    ret{{[l|q]}}
+;
+; CHECKO2-LABEL: pr34653:
+; CHECKO2:       # BB#0: # %entry
+; CHECKO2-NEXT:    pushq %rbp
+; CHECKO2-NEXT:    .cfi_def_cfa_offset 16
+; CHECKO2-NEXT:    .cfi_offset %rbp, -16
+; CHECKO2-NEXT:    movq %rsp, %rbp
+; CHECKO2-NEXT:    .cfi_def_cfa_register %rbp
+; CHECKO2-NEXT:    andq $-512, %rsp # imm = 0xFE00
+; CHECKO2-NEXT:    subq $1024, %rsp # imm = 0x400
+; CHECKO2-NEXT:    movq %rsp, %rdi
+; CHECKO2-NEXT:    callq test
+; CHECKO2-NEXT:    movq %rbp, %rsp
+; CHECKO2-NEXT:    popq %rbp
+; CHECKO2-NEXT:    ret{{[l|q]}}
+entry:
+  %v = call fastcc <38 x double> @test()
+  %v.0 = extractelement <38 x double> %v, i32 0
+  ret void
+}
+
+
+@u = external local_unnamed_addr global [20 x i32], align 16
+
+define void @pr34657() local_unnamed_addr {
+; CHECKO0-LABEL: pr34657:
+; CHECKO0:       # BB#0: # %entry
+; CHECKO0-NEXT:    # implicit-def: %RAX
+; CHECKO0-NEXT:    vmovups (%rax), %ymm0
+; CHECKO0-NEXT:    # implicit-def: %RAX
+; CHECKO0-NEXT:    vmovdqu (%rax), %xmm1
+; CHECKO0-NEXT:    # implicit-def: %YMM2
+; CHECKO0-NEXT:    vmovaps %xmm1, %xmm2
+; CHECKO0-NEXT:    vmovaps {{.*#+}} ymm3 = <255,u,0,255,u,u,255,255,u,0,255,u,u,255,255,u,u,255,u,u,255,255,u,u,255,u,u,255,255,u,u,255>
+; CHECKO0-NEXT:    vpblendvb %ymm3, %ymm0, %ymm2, %ymm0
+; CHECKO0-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; CHECKO0-NEXT:    vmovdqa {{.*#+}} xmm4 = [128,128,128,5,12,128,128,1,8,15,128,128,4,11,128,128]
+; CHECKO0-NEXT:    vpshufb %xmm4, %xmm1, %xmm1
+; CHECKO0-NEXT:    vmovaps %xmm0, %xmm4
+; CHECKO0-NEXT:    vmovdqa {{.*#+}} xmm5 = [0,7,14,128,128,3,10,128,128,128,6,13,128,128,2,9]
+; CHECKO0-NEXT:    vpshufb %xmm5, %xmm4, %xmm4
+; CHECKO0-NEXT:    vpor %xmm1, %xmm4, %xmm1
+; CHECKO0-NEXT:    vpmovzxbd {{.*#+}} zmm6 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero,xmm1[8],zero,zero,zero,xmm1[9],zero,zero,zero,xmm1[10],zero,zero,zero,xmm1[11],zero,zero,zero,xmm1[12],zero,zero,zero,xmm1[13],zero,zero,zero,xmm1[14],zero,zero,zero,xmm1[15],zero,zero,zero
+; CHECKO0-NEXT:    vmovaps %xmm6, %xmm1
+; CHECKO0-NEXT:    vmovd %xmm1, %ecx
+; CHECKO0-NEXT:    addl $0, %ecx
+; CHECKO0-NEXT:    movl %ecx, u
+; CHECKO0-NEXT:    vzeroupper
+; CHECKO0-NEXT:    ret{{[l|q]}}
+;
+; CHECKO2-LABEL: pr34657:
+; CHECKO2:       # BB#0: # %entry
+; CHECKO2-NEXT:    vmovdqu (%rax), %xmm0
+; CHECKO2-NEXT:    vmovdqu (%rax), %ymm1
+; CHECKO2-NEXT:    vmovdqu64 (%rax), %zmm2
+; CHECKO2-NEXT:    vpextrb $7, %xmm2, %eax
+; CHECKO2-NEXT:    vpextrb $0, %xmm2, %ecx
+; CHECKO2-NEXT:    vmovd %ecx, %xmm3
+; CHECKO2-NEXT:    vpinsrb $1, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vpextrb $14, %xmm2, %eax
+; CHECKO2-NEXT:    vpinsrb $2, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vextracti128 $1, %ymm2, %xmm4
+; CHECKO2-NEXT:    vpextrb $5, %xmm4, %eax
+; CHECKO2-NEXT:    vpinsrb $3, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vpextrb $12, %xmm4, %eax
+; CHECKO2-NEXT:    vpinsrb $4, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vextracti32x4 $2, %zmm2, %xmm4
+; CHECKO2-NEXT:    vpextrb $3, %xmm4, %eax
+; CHECKO2-NEXT:    vpinsrb $5, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vpextrb $10, %xmm4, %eax
+; CHECKO2-NEXT:    vpinsrb $6, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vextracti32x4 $3, %zmm2, %xmm2
+; CHECKO2-NEXT:    vpextrb $1, %xmm2, %eax
+; CHECKO2-NEXT:    vpinsrb $7, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vpextrb $8, %xmm2, %eax
+; CHECKO2-NEXT:    vpinsrb $8, %eax, %xmm3, %xmm3
+; CHECKO2-NEXT:    vpextrb $15, %xmm2, %eax
+; CHECKO2-NEXT:    vpinsrb $9, %eax, %xmm3, %xmm2
+; CHECKO2-NEXT:    vpextrb $6, %xmm1, %eax
+; CHECKO2-NEXT:    vpinsrb $10, %eax, %xmm2, %xmm2
+; CHECKO2-NEXT:    vpextrb $13, %xmm1, %eax
+; CHECKO2-NEXT:    vpinsrb $11, %eax, %xmm2, %xmm2
+; CHECKO2-NEXT:    vextracti128 $1, %ymm1, %xmm1
+; CHECKO2-NEXT:    vpextrb $4, %xmm1, %eax
+; CHECKO2-NEXT:    vpinsrb $12, %eax, %xmm2, %xmm2
+; CHECKO2-NEXT:    vpextrb $11, %xmm1, %eax
+; CHECKO2-NEXT:    vpinsrb $13, %eax, %xmm2, %xmm1
+; CHECKO2-NEXT:    vpextrb $2, %xmm0, %eax
+; CHECKO2-NEXT:    vpinsrb $14, %eax, %xmm1, %xmm1
+; CHECKO2-NEXT:    vpextrb $9, %xmm0, %eax
+; CHECKO2-NEXT:    vpinsrb $15, %eax, %xmm1, %xmm0
+; CHECKO2-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
+; CHECKO2-NEXT:    vmovd %xmm0, {{.*}}(%rip)
+; CHECKO2-NEXT:    vzeroupper
+; CHECKO2-NEXT:    ret{{[l|q]}}
+entry:
+  %wide.vec51 = load <112 x i8>, <112 x i8>* undef, align 2
+  %strided.vec54 = shufflevector <112 x i8> %wide.vec51, <112 x i8> undef, <16 x i32> <i32 0, i32 7, i32 14, i32 21, i32 28, i32 35, i32 42, i32 49, i32 56, i32 63, i32 70, i32 77, i32 84, i32 91, i32 98, i32 105>
+  %0 = zext <16 x i8> %strided.vec54 to <16 x i32>
+  %1 = add <16 x i32> zeroinitializer, %0
+  %bin.rdx56 = add <16 x i32> %1, zeroinitializer
+  %bin.rdx57 = add <16 x i32> zeroinitializer, %bin.rdx56
+  %bin.rdx58 = add <16 x i32> %bin.rdx57, zeroinitializer
+  %bin.rdx60 = add <16 x i32> %bin.rdx58, zeroinitializer
+  %bin.rdx62 = add <16 x i32> %bin.rdx60, zeroinitializer
+  %bin.rdx64 = add <16 x i32> %bin.rdx62, zeroinitializer
+  %2 = extractelement <16 x i32> %bin.rdx64, i32 0
+  %add18 = add i32 %2, 0
+  store i32 %add18, i32* getelementptr inbounds ([20 x i32], [20 x i32]* @u, i64 0, i64 0), align 16
+  ret void
+}
+