Index: lib/Target/X86/X86ISelDAGToDAG.cpp
===================================================================
--- lib/Target/X86/X86ISelDAGToDAG.cpp
+++ lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -2861,7 +2861,23 @@
   SDValue C8 = CurDAG->getConstant(8, DL, MVT::i8);
   SDValue Control = CurDAG->getNode(ISD::SHL, DL, NVT, NBits, C8);
   insertDAGNode(*CurDAG, OrigNBits, Control);
-  // NOTE: could also try to extract  start  from  (x >> start)
+
+  // 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);
+
+    if (ShiftAmt.getValueType() != MVT::i8) {
+      // Truncate the ShiftAmt amount.
+      SDValue OrigShiftAmt = ShiftAmt;
+      ShiftAmt = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, ShiftAmt);
+      insertDAGNode(*CurDAG, OrigShiftAmt, ShiftAmt);
+    }
+
+    // Insert 8-bit ShiftAmt into lowest 8 bits of 'control'.
+    Control = CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, NVT, Control,
+                                            ShiftAmt);
+  }
 
   // And finally, form the BEXTR itself.
   SDValue Extract = CurDAG->getNode(X86ISD::BEXTR, DL, NVT, X, Control);
Index: test/CodeGen/X86/extract-bits.ll
===================================================================
--- test/CodeGen/X86/extract-bits.ll
+++ test/CodeGen/X86/extract-bits.ll
@@ -50,10 +50,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a0:
@@ -78,10 +77,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -184,10 +181,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a1_indexzext:
@@ -212,10 +208,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -252,13 +246,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_a2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a2_load:
@@ -285,12 +278,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_a2_load:
@@ -325,13 +315,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
@@ -358,12 +347,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
@@ -401,10 +387,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_a4_commutative:
@@ -429,10 +414,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_a4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -476,13 +459,12 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
-; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
 ; X86-BMI1NOTBM-NEXT:    addl $8, %esp
@@ -523,9 +505,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_a5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -677,10 +658,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a0:
 ; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -967,10 +946,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1118,12 +1095,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_a2_load:
@@ -1268,12 +1242,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_a3_load_indexzext:
@@ -1417,10 +1388,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_a4_commutative:
 ; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1605,9 +1574,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
@@ -1657,10 +1625,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b0:
@@ -1685,10 +1652,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1724,10 +1689,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b1_indexzext:
@@ -1752,10 +1716,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -1792,13 +1754,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1825,12 +1786,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b2_load:
@@ -1865,13 +1823,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1898,12 +1855,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
@@ -1941,10 +1895,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_b4_commutative:
@@ -1969,10 +1922,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_b4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2016,13 +1967,12 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
-; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
 ; X86-BMI1NOTBM-NEXT:    addl $8, %esp
@@ -2063,9 +2013,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -2214,10 +2163,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b0:
 ; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2355,10 +2302,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2503,12 +2448,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b2_load:
@@ -2650,12 +2592,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_b3_load_indexzext:
@@ -2796,10 +2735,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_b4_commutative:
 ; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -2979,9 +2916,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
@@ -5164,12 +5100,11 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d0:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d0:
@@ -5194,10 +5129,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5230,10 +5163,9 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d1_indexzext:
@@ -5258,10 +5190,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5298,10 +5228,9 @@
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -5327,12 +5256,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -5364,13 +5290,12 @@
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    shll $8, %edx
+; X86-BMI1NOTBM-NEXT:    movb %cl, %dl
+; X86-BMI1NOTBM-NEXT:    bextrl %edx, (%eax), %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -5396,12 +5321,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrl %edx, (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -5444,13 +5366,12 @@
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
-; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
+; X86-BMI1NOTBM-NEXT:    movb %al, %cl
+; X86-BMI1NOTBM-NEXT:    bextrl %ecx, {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
 ; X86-BMI1NOTBM-NEXT:    addl $8, %esp
@@ -5491,9 +5412,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr32_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
 ; X64-BMI1NOTBM-NEXT:    shll $8, %edx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
@@ -5673,10 +5593,8 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d0:
 ; 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:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -5845,10 +5763,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
@@ -6023,12 +5939,9 @@
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d2_load:
@@ -6200,12 +6113,9 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
-; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
-; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
+; X64-BMI1NOTBM-NEXT:    bextrq %rdx, (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d3_load_indexzext:
@@ -6419,9 +6329,8 @@
 ; X64-BMI1NOTBM-LABEL: bextr64_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
 ; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
+; X64-BMI1NOTBM-NEXT:    movb %sil, %dl
 ; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64