X86DAGToDAGISel::matchBitExtract() with truncation (PR36419)

[lebedev.ri]

Summary

Previously in D54095 i have added support for extraction of lshr from X if we are to produce BEXTR.
That was good, but the fix was partial, there was still PR36419.

That pattern can also appear, roughly, when you have a large (64-bit) storage, and the consume bits from it.
It will not be unexpected if you will be doing further computations in 32-bit width.
And then the current code breaks, as the tests show.

The basic idea/pattern here is following:

1. We have i64 input
2. We perform i64 right-shift on it.
3. We truncate that shifted value
4. We do all further work (masking) in i32

Since we see truncation and not lshr, we give up, and stop trying to extract that right-shift.
BUT. The mask is i32, therefore we can extend both of the operands of the masking (and) to i64
and truncate the result after masking: https://rise4fun.com/Alive/K4B

Name: @bextr64_32_b1 -> @bextr64_32_b0  
  %shiftedval = lshr i64 %val, %numskipbits
  %truncshiftedval = trunc i64 %shiftedval to i32
  %widenumlowbits1 = zext i8 %numlowbits to i32
  %notmask1 = shl nsw i32 -1, %widenumlowbits1
  %mask1 = xor i32 %notmask1, -1
  %res = and i32 %truncshiftedval, %mask1
=>
  %shiftedval = lshr i64 %val, %numskipbits
  %widenumlowbits = zext i8 %numlowbits to i64
  %notmask = shl nsw i64 -1, %widenumlowbits
  %mask = xor i64 %notmask, -1
  %wideres = and i64 %shiftedval, %mask
  %res = trunc i64 %wideres to i32

Thus, we are again able to extract that lshr into BEXTR's control.

Now, the perf (via llvm-exegesis) of the snippet suggests that it is not a good idea:

$ cat /tmp/old.s 
# bextr64_32_b1
# LLVM-EXEGESIS-LIVEIN RSI
# LLVM-EXEGESIS-LIVEIN EDX
# LLVM-EXEGESIS-LIVEIN RDI
movq %rsi, %rcx
shrq %cl, %rdi
shll $8, %edx
bextrl %edx, %edi, %eax
$ cat /tmp/old.s | ./bin/llvm-exegesis -mode=latency -snippets-file=-
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-1e0082.o
---
mode:            latency
key:             
  instructions:    
    - 'MOV64rr RCX RSI'
    - 'SHR64rCL RDI RDI'
    - 'SHL32ri EDX EDX i_0x8'
    - 'BEXTR32rr EAX EDI EDX'
  config:          ''
  register_initial_values: []
cpu_name:        bdver2
llvm_triple:     x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:    
  - { key: latency, value: 0.6638, per_snippet_value: 2.6552 }
error:           ''
info:            ''
assembled_snippet: 4889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C7C3
...
$ cat /tmp/old.s | ./bin/llvm-exegesis -mode=uops -snippets-file=-
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-43e346.o
---
mode:            uops
key:             
  instructions:    
    - 'MOV64rr RCX RSI'
    - 'SHR64rCL RDI RDI'
    - 'SHL32ri EDX EDX i_0x8'
    - 'BEXTR32rr EAX EDI EDX'
  config:          ''
  register_initial_values: []
cpu_name:        bdver2
llvm_triple:     x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:    
  - { key: PdFPU0, value: 0, per_snippet_value: 0 }
  - { key: PdFPU1, value: 0, per_snippet_value: 0 }
  - { key: PdFPU2, value: 0, per_snippet_value: 0 }
  - { key: PdFPU3, value: 0, per_snippet_value: 0 }
  - { key: NumMicroOps, value: 1.2571, per_snippet_value: 5.0284 }
error:           ''
info:            ''
assembled_snippet: 4889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C74889F148D3EFC1E208C4E268F7C7C3
...

vs

$ cat /tmp/new.s 
# bextr64_32_b1
# LLVM-EXEGESIS-LIVEIN RDX
# LLVM-EXEGESIS-LIVEIN SIL
# LLVM-EXEGESIS-LIVEIN RDI
shlq $8, %rdx
movzbl %sil, %eax
orq %rdx, %rax
bextrq %rax, %rdi, %rax
$ cat /tmp/new.s | ./bin/llvm-exegesis -mode=latency -snippets-file=-
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-8944f1.o
---
mode:            latency
key:             
  instructions:    
    - 'SHL64ri RDX RDX i_0x8'
    - 'MOVZX32rr8 EAX SIL'
    - 'OR64rr RAX RAX RDX'
    - 'BEXTR64rr RAX RDI RAX'
  config:          ''
  register_initial_values: []
cpu_name:        bdver2
llvm_triple:     x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:    
  - { key: latency, value: 0.7454, per_snippet_value: 2.9816 }
error:           ''
info:            ''
assembled_snippet: 48C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C7C3
...
$ cat /tmp/new.s | ./bin/llvm-exegesis -mode=uops -snippets-file=-
Check generated assembly with: /usr/bin/objdump -d /tmp/snippet-da403c.o
---
mode:            uops
key:             
  instructions:    
    - 'SHL64ri RDX RDX i_0x8'
    - 'MOVZX32rr8 EAX SIL'
    - 'OR64rr RAX RAX RDX'
    - 'BEXTR64rr RAX RDI RAX'
  config:          ''
  register_initial_values: []
cpu_name:        bdver2
llvm_triple:     x86_64-unknown-linux-gnu
num_repetitions: 10000
measurements:    
  - { key: PdFPU0, value: 0, per_snippet_value: 0 }
  - { key: PdFPU1, value: 0, per_snippet_value: 0 }
  - { key: PdFPU2, value: 0, per_snippet_value: 0 }
  - { key: PdFPU3, value: 0, per_snippet_value: 0 }
  - { key: NumMicroOps, value: 1.2571, per_snippet_value: 5.0284 }
error:           ''
info:            ''
assembled_snippet: 48C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C748C1E208400FB6C64809D0C4E2F8F7C7C3
...

^ latency increased (worse).

Except maybe not really.
Like with all synthetic benchmarks, they may be misleading.

Let's take a look on some actual real-world hotpath.
In this case it's 'my' RawSpeed's BitStream<>::peekBitsNoFill(), in GoPro VC5 decompressor:

raw.pixls.us-unique/GoPro/HERO6 Black$ /usr/src/googlebenchmark/tools/compare.py -a benchmarks ~/rawspeed/build-clangs1-{old,new}/src/utilities/rsbench/rsbench --benchmark_counters_tabular=true --benchmark_min_time=0.00000001 --benchmark_repetitions=128 GOPR9172.GPR 
RUNNING: /home/lebedevri/rawspeed/build-clangs1-old/src/utilities/rsbench/rsbench --benchmark_counters_tabular=true --benchmark_min_time=0.00000001 --benchmark_repetitions=128 GOPR9172.GPR --benchmark_display_aggregates_only=true --benchmark_out=/tmp/tmplwbKEM
2018-12-22 21:23:03
Running /home/lebedevri/rawspeed/build-clangs1-old/src/utilities/rsbench/rsbench
Run on (8 X 4012.81 MHz CPU s)
CPU Caches:
  L1 Data 16K (x8)
  L1 Instruction 64K (x4)
  L2 Unified 2048K (x4)
  L3 Unified 8192K (x1)
Load Average: 3.41, 2.41, 2.03
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Benchmark                                        Time           CPU Iterations  CPUTime,s CPUTime/WallTime     Pixels Pixels/CPUTime Pixels/WallTime Raws/CPUTime Raws/WallTime WallTime,s
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
GOPR9172.GPR/threads:8/real_time_mean           40 ms         40 ms        128   0.322244          7.96974        12M       37.4457M        298.534M      3.12047       24.8778   0.040465
GOPR9172.GPR/threads:8/real_time_median         39 ms         39 ms        128   0.312606          7.99155        12M        38.387M        306.788M      3.19891       25.5656   0.039115
GOPR9172.GPR/threads:8/real_time_stddev          4 ms          3 ms        128  0.0271557         0.130575          0        2.4941M        21.3909M     0.207842       1.78257   3.81081m
RUNNING: /home/lebedevri/rawspeed/build-clangs1-new/src/utilities/rsbench/rsbench --benchmark_counters_tabular=true --benchmark_min_time=0.00000001 --benchmark_repetitions=128 GOPR9172.GPR --benchmark_display_aggregates_only=true --benchmark_out=/tmp/tmpWAkan9
2018-12-22 21:23:08
Running /home/lebedevri/rawspeed/build-clangs1-new/src/utilities/rsbench/rsbench
Run on (8 X 4013.1 MHz CPU s)
CPU Caches:
  L1 Data 16K (x8)
  L1 Instruction 64K (x4)
  L2 Unified 2048K (x4)
  L3 Unified 8192K (x1)
Load Average: 3.78, 2.50, 2.06
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Benchmark                                        Time           CPU Iterations  CPUTime,s CPUTime/WallTime     Pixels Pixels/CPUTime Pixels/WallTime Raws/CPUTime Raws/WallTime WallTime,s
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
GOPR9172.GPR/threads:8/real_time_mean           39 ms         39 ms        128   0.311533          7.97323        12M       38.6828M        308.471M      3.22356        25.706  0.0390928
GOPR9172.GPR/threads:8/real_time_median         38 ms         38 ms        128   0.304231          7.99005        12M       39.4437M        315.527M      3.28698        26.294  0.0380316
GOPR9172.GPR/threads:8/real_time_stddev          3 ms          3 ms        128  0.0229149         0.133814          0       2.26225M        19.1421M     0.188521       1.59517   3.13671m
Comparing /home/lebedevri/rawspeed/build-clangs1-old/src/utilities/rsbench/rsbench to /home/lebedevri/rawspeed/build-clangs1-new/src/utilities/rsbench/rsbench
Benchmark                                                 Time             CPU      Time Old      Time New       CPU Old       CPU New
--------------------------------------------------------------------------------------------------------------------------------------
GOPR9172.GPR/threads:8/real_time_pvalue                 0.0000          0.0000      U Test, Repetitions: 128 vs 128
GOPR9172.GPR/threads:8/real_time_mean                  -0.0339         -0.0316            40            39            40            39
GOPR9172.GPR/threads:8/real_time_median                -0.0277         -0.0274            39            38            39            38
GOPR9172.GPR/threads:8/real_time_stddev                -0.1769         -0.1267             4             3             3             3

I.e. this results in roughly -3% improvements in perf.

While this will help PR36419, it won't address it fully.

[lebedev.ri]

low-key ping

[lebedev.ri]

Hm, ping.
I'm guessing everyone is busy with funnel-related stuff, on the eve of the 8.0
Or would it be easier if i dropped all the TLDR in the description?

[craig.topper]

lib/Target/X86/X86ISelDAGToDAG.cpp

2822 ↗

(On Diff #179429)

Should we ensure the truncate input has only one use?

[lebedev.ri]

lib/Target/X86/X86ISelDAGToDAG.cpp

2822 ↗

(On Diff #179429)

Hm, since we only take that branch in BMI1's case, it would be consistent with the rest of the one-use checks,
so i guess we should, yes. Will do.

test/CodeGen/X86/extract-bits.ll

1832 ↗

(On Diff #179429)

One more thing is that 'control' being calculated as i64, while we only care about low 32 bits.
Will take a look, but not in this patch.

[lebedev.ri]

Rebased ontop of patch with extra-use trunk.
Did not add the use check yet..

[lebedev.ri]

Only skip one-use truncations.

[craig.topper]

lib/Target/X86/X86ISelDAGToDAG.cpp

2954 ↗

(On Diff #181757)

Just use X.hasOneUse()? There's a special case in checkOneUse for hasBMI2, but that doesnt' apply here. So let's just write the most direct thing

[lebedev.ri]

Address review feedback - use X.hasOneUse() directly.

lib/Target/X86/X86ISelDAGToDAG.cpp

2954 ↗

(On Diff #181757)

Right.
I was going for consistency so one would not wonder why this would be using hasOneUse(),
while others would be using those helper lambdas.

[craig.topper]

LGTM

[lebedev.ri]

In D56052#1358320, @craig.topper wrote:

LGTM

Thank you for the review!

test/CodeGen/X86/extract-bits.ll

1832 ↗

(On Diff #179429)

Submitted as D56715.

llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp

if (matchLowBitMask(Mask)) {
       if (!matchLowBitMask(Mask))
         return false;
     }
   } else if (!matchPatternD(Node))
     return false;
 
   SDLoc DL(Node);
 
+  // If we do *NOT* have BMI2, let's find out if the if the 'X' is *logically*
+  // shifted (potentially with one-use trunc inbetween),
+  // and if so look past one-use truncation.
+  MVT XVT = NVT;
+  if (!Subtarget->hasBMI2() && X.getOpcode() == ISD::TRUNCATE &&
+      X.hasOneUse() && X.getOperand(0).getOpcode() == ISD::SRL) {
+    assert(NVT == MVT::i32 && "Expected target valuetype to be i32");
+    X = X.getOperand(0);
+    XVT = X.getSimpleValueType();
+    assert(XVT == MVT::i64 && "Expected truncation from i64");
+  }
+
   SDValue OrigNBits = NBits;
-  if (NBits.getValueType() != NVT) {
+  if (NBits.getValueType() != XVT) {
     // Truncate the shift amount.
     NBits = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, NBits);
     insertDAGNode(*CurDAG, OrigNBits, NBits);
 
-    // Insert 8-bit NBits into lowest 8 bits of NVT-sized (32 or 64-bit)
+    // Insert 8-bit NBits into lowest 8 bits of XVT-sized (32 or 64-bit)
     // register. All the other bits are undefined, we do not care about them.
     SDValue ImplDef =
-        SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, NVT), 0);
+        SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, XVT), 0);
     insertDAGNode(*CurDAG, OrigNBits, ImplDef);
     NBits =
-        CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, NVT, ImplDef, NBits);
+        CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, XVT, ImplDef, NBits);
     insertDAGNode(*CurDAG, OrigNBits, NBits);
   }
 
   if (Subtarget->hasBMI2()) {
     // Great, just emit the the BZHI..
-    SDValue Extract = CurDAG->getNode(X86ISD::BZHI, DL, NVT, X, NBits);
+    SDValue Extract = CurDAG->getNode(X86ISD::BZHI, DL, XVT, X, NBits);
     ReplaceNode(Node, Extract.getNode());
     SelectCode(Extract.getNode());
     return true;
   }
 
   // Else, emitting BEXTR requires one more step.
   // The 'control' of BEXTR has the pattern of:
   // [15...8 bit][ 7...0 bit] location
   // [ bit count][     shift] name
   // I.e. 0b000000011'00000001 means  (x >> 0b1) & 0b11
 
   // Shift NBits left by 8 bits, thus producing 'control'.
   // This makes the low 8 bits to be zero.
   SDValue C8 = CurDAG->getConstant(8, DL, MVT::i8);
-  SDValue Control = CurDAG->getNode(ISD::SHL, DL, NVT, NBits, C8);
+  SDValue Control = CurDAG->getNode(ISD::SHL, DL, XVT, NBits, C8);
   insertDAGNode(*CurDAG, OrigNBits, Control);
 
   // If the 'X' is *logically* shifted, we can fold that shift into 'control'.
   if (X.getOpcode() == ISD::SRL) {
     SDValue ShiftAmt = X.getOperand(1);
     X = X.getOperand(0);
 
     assert(ShiftAmt.getValueType() == MVT::i8 &&
            "Expected shift amount to be i8");
 
     // Now, *zero*-extend the shift amount. The bits 8...15 *must* be zero!
     SDValue OrigShiftAmt = ShiftAmt;
-    ShiftAmt = CurDAG->getNode(ISD::ZERO_EXTEND, DL, NVT, ShiftAmt);
+    ShiftAmt = CurDAG->getNode(ISD::ZERO_EXTEND, DL, XVT, ShiftAmt);
     insertDAGNode(*CurDAG, OrigShiftAmt, ShiftAmt);
 
     // And now 'or' these low 8 bits of shift amount into the 'control'.
-    Control = CurDAG->getNode(ISD::OR, DL, NVT, Control, ShiftAmt);
+    Control = CurDAG->getNode(ISD::OR, DL, XVT, Control, ShiftAmt);
     insertDAGNode(*CurDAG, OrigNBits, Control);
   }
 
   // And finally, form the BEXTR itself.
-  SDValue Extract = CurDAG->getNode(X86ISD::BEXTR, DL, NVT, X, Control);
+  SDValue Extract = CurDAG->getNode(X86ISD::BEXTR, DL, XVT, X, Control);
+
+  // The 'X' was originally truncated. Do that now.
+  if (XVT != NVT) {
+    insertDAGNode(*CurDAG, OrigNBits, Extract);
+    Extract = CurDAG->getNode(ISD::TRUNCATE, DL, NVT, Extract);
+  }
+
   ReplaceNode(Node, Extract.getNode());
   SelectCode(Extract.getNode());
 
   return true;
 }
 
 // See if this is an (X >> C1) & C2 that we can match to BEXTR/BEXTRI.
 MachineSDNode *X86DAGToDAGISel::matchBEXTRFromAndImm(SDNode *Node) {

llvm/trunk/test/CodeGen/X86/extract-bits.ll

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    decl %eax
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_a1:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_a1:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shifted = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    decl %eax
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_a2:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_a2:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shifted = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    notl %eax
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_b1:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_b1:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shiftedval = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    notl %eax
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_b2:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_b2:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shiftedval = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_c1:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_c1:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shifted = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_c2:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_c2:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shifted = lshr i64 %val, %numskipbits

 ; X64-NOBMI-NEXT:    movl %edx, %ecx
 ; X64-NOBMI-NEXT:    shll %cl, %eax
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-NOBMI-NEXT:    retq
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_32_d1:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movzbl %sil, %eax
+; X64-BMI1NOTBM-NEXT:    orq %rdx, %rax
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $eax killed $eax killed $rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_32_d1:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
 ; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
 ; X64-BMI1BMI2-NEXT:    retq
   %shifted = lshr i64 %val, %numskipbits