[llvm] r342623 - [NFC][x86][AArch64] Add BEXTR-like test patterns.

Roman Lebedev via llvm-commits llvm-commits at lists.llvm.org
Thu Sep 20 00:54:49 PDT 2018


Author: lebedevri
Date: Thu Sep 20 00:54:49 2018
New Revision: 342623

URL: http://llvm.org/viewvc/llvm-project?rev=342623&view=rev
Log:
[NFC][x86][AArch64] Add BEXTR-like test patterns.

Summary: Also, adjust the check prefixes so that we actually get to check the BMI1-only-case.

Reviewers: craig.topper, RKSimon, spatel, javed.absar

Reviewed By: RKSimon

Subscribers: kristof.beyls, llvm-commits

Differential Revision: https://reviews.llvm.org/D48490

Added:
    llvm/trunk/test/CodeGen/AArch64/extract-bits.ll
    llvm/trunk/test/CodeGen/X86/extract-bits.ll
Modified:
    llvm/trunk/test/CodeGen/X86/extract-lowbits.ll

Added: llvm/trunk/test/CodeGen/AArch64/extract-bits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/extract-bits.ll?rev=342623&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/extract-bits.ll (added)
+++ llvm/trunk/test/CodeGen/AArch64/extract-bits.ll Thu Sep 20 00:54:49 2018
@@ -0,0 +1,717 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
+
+; *Please* keep in sync with test/CodeGen/X86/extract-bits.ll
+
+; https://bugs.llvm.org/show_bug.cgi?id=36419
+; https://bugs.llvm.org/show_bug.cgi?id=37603
+; https://bugs.llvm.org/show_bug.cgi?id=37610
+
+; Patterns:
+;   a) (x >> start) &  (1 << nbits) - 1
+;   b) (x >> start) & ~(-1 << nbits)
+;   c) (x >> start) &  (-1 >> (32 - y))
+;   d) (x >> start) << (32 - y) >> (32 - y)
+; are equivalent.
+
+; ---------------------------------------------------------------------------- ;
+; Pattern a. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_a0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_a0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    sub w9, w9, #1 // =1
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_a1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    sub w9, w9, #1 // =1
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %onebit = shl i32 1, %conv
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_a2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    sub w9, w9, #1 // =1
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_a3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    sub w9, w9, #1 // =1
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %onebit = shl i32 1, %conv
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_a4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    sub w9, w9, #1 // =1
+; CHECK-NEXT:    and w0, w8, w9
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_a0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    sub x9, x9, #1 // =1
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_a1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    sub x9, x9, #1 // =1
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %onebit = shl i64 1, %conv
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_a2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    sub x9, x9, #1 // =1
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_a3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    sub x9, x9, #1 // =1
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %onebit = shl i64 1, %conv
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_a4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    sub x9, x9, #1 // =1
+; CHECK-NEXT:    and x0, x8, x9
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern b. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_b0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_b0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w9, #-1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    bic w0, w8, w9
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_b1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w9, #-1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    bic w0, w8, w9
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %notmask = shl i32 -1, %conv
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_b2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    mov w9, #-1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    bic w0, w8, w9
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_b3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    mov w9, #-1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    bic w0, w8, w9
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %notmask = shl i32 -1, %conv
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_b4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov w9, #-1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsl w9, w9, w2
+; CHECK-NEXT:    bic w0, w8, w9
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_b0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_b0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov x9, #-1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    bic x0, x8, x9
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_b1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov x9, #-1
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    bic x0, x8, x9
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %notmask = shl i64 -1, %conv
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_b2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    mov x9, #-1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    bic x0, x8, x9
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_b3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    mov x9, #-1
+; CHECK-NEXT:    // kill: def $w2 killed $w2 def $x2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    bic x0, x8, x9
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %notmask = shl i64 -1, %conv
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_b4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov x9, #-1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsl x9, x9, x2
+; CHECK-NEXT:    bic x0, x8, x9
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern c. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_c0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_c0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    neg w9, w2
+; CHECK-NEXT:    mov w10, #-1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_c1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x20
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    mov w10, #-1
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_c2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    neg w9, w2
+; CHECK-NEXT:    mov w10, #-1
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_c3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x20
+; CHECK-NEXT:    mov w10, #-1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    and w0, w9, w8
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_c4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    neg w9, w2
+; CHECK-NEXT:    mov w10, #-1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    lsr w9, w10, w9
+; CHECK-NEXT:    and w0, w8, w9
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_c0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_c0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    neg x9, x2
+; CHECK-NEXT:    mov x10, #-1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsr x9, x10, x9
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_c1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x40
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    mov x10, #-1
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsr x9, x10, x9
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_c2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    neg x9, x2
+; CHECK-NEXT:    mov x10, #-1
+; CHECK-NEXT:    lsr x9, x10, x9
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_c3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x40
+; CHECK-NEXT:    mov x10, #-1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    lsr x9, x10, x9
+; CHECK-NEXT:    and x0, x9, x8
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_c4_commutative:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    neg x9, x2
+; CHECK-NEXT:    mov x10, #-1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    lsr x9, x10, x9
+; CHECK-NEXT:    and x0, x8, x9
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern d. 32-bit.
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_d0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_d0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    neg w9, w2
+; CHECK-NEXT:    lsl w8, w8, w9
+; CHECK-NEXT:    lsr w0, w8, w9
+; CHECK-NEXT:    ret
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %highbitscleared = shl i32 %shifted, %numhighbits
+  %masked = lshr i32 %highbitscleared, %numhighbits
+  ret i32 %masked
+}
+
+define i32 @bextr32_d1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_d1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x20
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr w8, w0, w1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsl w8, w8, w9
+; CHECK-NEXT:    lsr w0, w8, w9
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %highbitscleared = shl i32 %shifted, %sh_prom
+  %masked = lshr i32 %highbitscleared, %sh_prom
+  ret i32 %masked
+}
+
+define i32 @bextr32_d2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_d2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    neg w9, w2
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    lsl w8, w8, w9
+; CHECK-NEXT:    lsr w0, w8, w9
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %highbitscleared = shl i32 %shifted, %numhighbits
+  %masked = lshr i32 %highbitscleared, %numhighbits
+  ret i32 %masked
+}
+
+define i32 @bextr32_d3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr32_d3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr w8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x20
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsr w8, w8, w1
+; CHECK-NEXT:    lsl w8, w8, w9
+; CHECK-NEXT:    lsr w0, w8, w9
+; CHECK-NEXT:    ret
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %highbitscleared = shl i32 %shifted, %sh_prom
+  %masked = lshr i32 %highbitscleared, %sh_prom
+  ret i32 %masked
+}
+
+; 64-bit.
+
+define i64 @bextr64_d0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_d0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    neg x9, x2
+; CHECK-NEXT:    lsl x8, x8, x9
+; CHECK-NEXT:    lsr x0, x8, x9
+; CHECK-NEXT:    ret
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %highbitscleared = shl i64 %shifted, %numhighbits
+  %masked = lshr i64 %highbitscleared, %numhighbits
+  ret i64 %masked
+}
+
+define i64 @bextr64_d1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_d1_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    orr w9, wzr, #0x40
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    lsr x8, x0, x1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsl x8, x8, x9
+; CHECK-NEXT:    lsr x0, x8, x9
+; CHECK-NEXT:    ret
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %highbitscleared = shl i64 %shifted, %sh_prom
+  %masked = lshr i64 %highbitscleared, %sh_prom
+  ret i64 %masked
+}
+
+define i64 @bextr64_d2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_d2_load:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    neg x9, x2
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    lsl x8, x8, x9
+; CHECK-NEXT:    lsr x0, x8, x9
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %highbitscleared = shl i64 %shifted, %numhighbits
+  %masked = lshr i64 %highbitscleared, %numhighbits
+  ret i64 %masked
+}
+
+define i64 @bextr64_d3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; CHECK-LABEL: bextr64_d3_load_indexzext:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr x8, [x0]
+; CHECK-NEXT:    orr w9, wzr, #0x40
+; CHECK-NEXT:    // kill: def $w1 killed $w1 def $x1
+; CHECK-NEXT:    sub w9, w9, w2
+; CHECK-NEXT:    lsr x8, x8, x1
+; CHECK-NEXT:    lsl x8, x8, x9
+; CHECK-NEXT:    lsr x0, x8, x9
+; CHECK-NEXT:    ret
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %highbitscleared = shl i64 %shifted, %sh_prom
+  %masked = lshr i64 %highbitscleared, %sh_prom
+  ret i64 %masked
+}

Added: llvm/trunk/test/CodeGen/X86/extract-bits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/extract-bits.ll?rev=342623&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/extract-bits.ll (added)
+++ llvm/trunk/test/CodeGen/X86/extract-bits.ll Thu Sep 20 00:54:49 2018
@@ -0,0 +1,5642 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI,X86-NOBMI
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1NOTBM,X86-BMI1NOTBM
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1TBM,X86-BMI1TBM
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1TBM,X86-BMI1TBM,BMI1TBMBMI2,X86-BMI1TBMBMI2
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1NOTBMBMI2,X86-BMI1NOTBMBMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI,X64-NOBMI
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1NOTBM,X64-BMI1NOTBM
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1TBM,X64-BMI1TBM
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1TBM,X64-BMI1TBM,BMI1TBMBMI2,X64-BMI1TBMBMI2
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1NOTBMBMI2,X64-BMI1NOTBMBMI2
+
+; *Please* keep in sync with test/CodeGen/AArch64/extract-bits.ll
+
+; https://bugs.llvm.org/show_bug.cgi?id=36419
+; https://bugs.llvm.org/show_bug.cgi?id=37603
+; https://bugs.llvm.org/show_bug.cgi?id=37610
+
+; Patterns:
+;   a) (x >> start) &  (1 << nbits) - 1
+;   b) (x >> start) & ~(-1 << nbits)
+;   c) (x >> start) &  (-1 >> (32 - y))
+;   d) (x >> start) << (32 - y) >> (32 - y)
+; are equivalent.
+
+declare void @use32(i32)
+declare void @use64(i64)
+
+; ---------------------------------------------------------------------------- ;
+; Pattern a. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_a0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    decl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; 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: 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:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    decl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; 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: 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:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %onebit = shl i32 1, %conv
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    decl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %esi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %esi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    decl %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_a2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %esi
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    decl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %esi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %esi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    decl %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_a3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %esi
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %onebit = shl i32 1, %conv
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_a4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    decl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; 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: 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:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+define i32 @bextr32_a5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_a5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    movl $1, %esi
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    decl %esi
+; X86-NOBMI-NEXT:    andl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, (%esp)
+; X86-NOBMI-NEXT:    calll use32
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    addl $4, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_a5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    movl $1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    decl %esi
+; X86-BMI1NOTBM-NEXT:    andl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
+; X86-BMI1NOTBM-NEXT:    calll use32
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    addl $4, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_a5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %edx, %esi
+; X86-BMI1BMI2-NEXT:    movl %ecx, (%esp)
+; X86-BMI1BMI2-NEXT:    calll use32
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    addl $8, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_a5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $1, %ebx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %ebx
+; X64-NOBMI-NEXT:    decl %ebx
+; X64-NOBMI-NEXT:    andl %edi, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %edi
+; X64-NOBMI-NEXT:    callq use32
+; X64-NOBMI-NEXT:    movl %ebx, %eax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $1, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    decl %ebx
+; X64-BMI1NOTBM-NEXT:    andl %edi, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X64-BMI1NOTBM-NEXT:    callq use32
+; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_a5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %ebx
+; X64-BMI1BMI2-NEXT:    movl %esi, %edi
+; X64-BMI1BMI2-NEXT:    callq use32
+; X64-BMI1BMI2-NEXT:    movl %ebx, %eax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %onebit = shl i32 1, %numlowbits
+  %mask = add nsw i32 %onebit, -1
+  %masked = and i32 %mask, %shifted
+  call void @use32(i32 %numskipbits)
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_a0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB6_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB6_2:
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB6_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB6_4:
+; X86-NOBMI-NEXT:    addl $-1, %eax
+; X86-NOBMI-NEXT:    adcl $-1, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB6_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB6_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    movb %ch, %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %ch
+; X86-BMI1NOTBM-NEXT:    je .LBB6_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB6_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB6_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB6_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shlxl %ebx, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %bl
+; X86-BMI1BMI2-NEXT:    je .LBB6_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI1BMI2-NEXT:  .LBB6_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %eax
+; X86-BMI1BMI2-NEXT:    adcl $-1, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    decq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB7_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB7_2:
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB7_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB7_4:
+; X86-NOBMI-NEXT:    addl $-1, %eax
+; X86-NOBMI-NEXT:    adcl $-1, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB7_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB7_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    movb %ch, %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %ch
+; X86-BMI1NOTBM-NEXT:    je .LBB7_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB7_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB7_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB7_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shlxl %ebx, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %bl
+; X86-BMI1BMI2-NEXT:    je .LBB7_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI1BMI2-NEXT:  .LBB7_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %eax
+; X86-BMI1BMI2-NEXT:    adcl $-1, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    decq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_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:    shrq %cl, %rdi
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %onebit = shl i64 1, %conv
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB8_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB8_2:
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB8_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB8_4:
+; X86-NOBMI-NEXT:    addl $-1, %eax
+; X86-NOBMI-NEXT:    adcl $-1, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB8_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB8_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    movb %ch, %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %ch
+; X86-BMI1NOTBM-NEXT:    je .LBB8_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB8_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl (%eax), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%eax), %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB8_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB8_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shlxl %ebx, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %bl
+; X86-BMI1BMI2-NEXT:    je .LBB8_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI1BMI2-NEXT:  .LBB8_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %eax
+; X86-BMI1BMI2-NEXT:    adcl $-1, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq (%rdi), %rsi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rsi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    decq %rax
+; X64-NOBMI-NEXT:    andq %rsi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_a2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rsi
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rsi
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB9_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB9_2:
+; X86-NOBMI-NEXT:    movl $1, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shldl %cl, %eax, %edx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB9_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB9_4:
+; X86-NOBMI-NEXT:    addl $-1, %eax
+; X86-NOBMI-NEXT:    adcl $-1, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB9_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB9_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    movb %ch, %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %ch
+; X86-BMI1NOTBM-NEXT:    je .LBB9_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB9_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl (%eax), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%eax), %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB9_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB9_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shlxl %ebx, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %bl
+; X86-BMI1BMI2-NEXT:    je .LBB9_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:    xorl %eax, %eax
+; X86-BMI1BMI2-NEXT:  .LBB9_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %eax
+; X86-BMI1BMI2-NEXT:    adcl $-1, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq (%rdi), %rsi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rsi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    decq %rax
+; X64-NOBMI-NEXT:    andq %rsi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_a3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rsi
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rsi
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %onebit = shl i64 1, %conv
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_a4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl %esi, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB10_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB10_2:
+; X86-NOBMI-NEXT:    movl $1, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shldl %cl, %esi, %edi
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB10_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %esi, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:  .LBB10_4:
+; X86-NOBMI-NEXT:    addl $-1, %esi
+; X86-NOBMI-NEXT:    adcl $-1, %edi
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB10_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB10_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:    movb %ch, %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %esi, %edi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %ch
+; X86-BMI1NOTBM-NEXT:    je .LBB10_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB10_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %bl
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB10_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB10_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %edi
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %ebx, %edi, %ecx
+; X86-BMI1BMI2-NEXT:    testb $32, %bl
+; X86-BMI1BMI2-NEXT:    je .LBB10_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ecx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %ecx, %ecx
+; X86-BMI1BMI2-NEXT:  .LBB10_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %ecx
+; X86-BMI1BMI2-NEXT:    adcl $-1, %esi
+; X86-BMI1BMI2-NEXT:    andl %ecx, %eax
+; X86-BMI1BMI2-NEXT:    andl %esi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movl $1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    decq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+define i64 @bextr64_a5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_a5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebp
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $12, %esp
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %esi, %ebp
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %ebp
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %ebx
+; X86-NOBMI-NEXT:    testb $32, %al
+; X86-NOBMI-NEXT:    je .LBB11_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %ebp, %ebx
+; X86-NOBMI-NEXT:    xorl %ebp, %ebp
+; X86-NOBMI-NEXT:  .LBB11_2:
+; X86-NOBMI-NEXT:    movl $1, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shldl %cl, %esi, %edi
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    testb $32, %dl
+; X86-NOBMI-NEXT:    je .LBB11_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %esi, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:  .LBB11_4:
+; X86-NOBMI-NEXT:    addl $-1, %esi
+; X86-NOBMI-NEXT:    adcl $-1, %edi
+; X86-NOBMI-NEXT:    andl %ebx, %esi
+; X86-NOBMI-NEXT:    andl %ebp, %edi
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    calll use64
+; X86-NOBMI-NEXT:    addl $16, %esp
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    movl %edi, %edx
+; X86-NOBMI-NEXT:    addl $12, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    popl %ebp
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebp
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $12, %esp
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebp
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %ebp
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB11_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %ebp, %ebx
+; X86-BMI1NOTBM-NEXT:    xorl %ebp, %ebp
+; X86-BMI1NOTBM-NEXT:  .LBB11_2:
+; X86-BMI1NOTBM-NEXT:    movl $1, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %esi, %edi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %dl
+; X86-BMI1NOTBM-NEXT:    je .LBB11_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB11_4:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    andl %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    andl %ebp, %edi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    calll use64
+; X86-BMI1NOTBM-NEXT:    addl $16, %esp
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    addl $12, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    popl %ebp
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_a5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebp
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $12, %esp
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %ebx
+; X86-BMI1BMI2-NEXT:    shrxl %eax, %esi, %ebp
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB11_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %ebp, %ebx
+; X86-BMI1BMI2-NEXT:    xorl %ebp, %ebp
+; X86-BMI1BMI2-NEXT:  .LBB11_2:
+; X86-BMI1BMI2-NEXT:    movl $1, %edi
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:    movl %edx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %edx, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %dl
+; X86-BMI1BMI2-NEXT:    je .LBB11_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB11_4:
+; X86-BMI1BMI2-NEXT:    addl $-1, %edi
+; X86-BMI1BMI2-NEXT:    adcl $-1, %esi
+; X86-BMI1BMI2-NEXT:    andl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    andl %ebp, %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-BMI1BMI2-NEXT:    pushl %eax
+; X86-BMI1BMI2-NEXT:    calll use64
+; X86-BMI1BMI2-NEXT:    addl $16, %esp
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:    movl %esi, %edx
+; X86-BMI1BMI2-NEXT:    addl $12, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    popl %ebp
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_a5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movl $1, %ebx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rbx
+; X64-NOBMI-NEXT:    decq %rbx
+; X64-NOBMI-NEXT:    andq %rdi, %rbx
+; X64-NOBMI-NEXT:    movq %rsi, %rdi
+; X64-NOBMI-NEXT:    callq use64
+; X64-NOBMI-NEXT:    movq %rbx, %rax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_a5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
+; X64-BMI1NOTBM-NEXT:    movl $1, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    decq %rbx
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
+; X64-BMI1NOTBM-NEXT:    callq use64
+; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_a5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
+; X64-BMI1BMI2-NEXT:    movq %rsi, %rdi
+; X64-BMI1BMI2-NEXT:    callq use64
+; X64-BMI1BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %onebit = shl i64 1, %numlowbits
+  %mask = add nsw i64 %onebit, -1
+  %masked = and i64 %mask, %shifted
+  call void @use64(i64 %numskipbits)
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern b. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_b0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $-1, %eax
+; 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: 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:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b1_indexzext(i32 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $-1, %eax
+; 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: 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:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %notmask = shl i32 -1, %conv
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl (%edx), %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %esi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %esi
+; X64-NOBMI-NEXT:    movl $-1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    notl %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $-1, %esi
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X64-BMI1NOTBM-NEXT:    andnl %eax, %esi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b3_load_indexzext(i32* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl (%edx), %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %esi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %esi
+; X64-NOBMI-NEXT:    movl $-1, %eax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    notl %eax
+; X64-NOBMI-NEXT:    andl %esi, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $-1, %esi
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X64-BMI1NOTBM-NEXT:    andnl %eax, %esi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %conv = zext i8 %numlowbits to i32
+  %notmask = shl i32 -1, %conv
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_b4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $-1, %eax
+; 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: 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:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+define i32 @bextr32_b5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_b5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    movl $-1, %esi
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    notl %esi
+; X86-NOBMI-NEXT:    andl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, (%esp)
+; X86-NOBMI-NEXT:    calll use32
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    addl $4, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_b5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
+; X86-BMI1NOTBM-NEXT:    calll use32
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    addl $4, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_b5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %edx, %esi
+; X86-BMI1BMI2-NEXT:    movl %ecx, (%esp)
+; X86-BMI1BMI2-NEXT:    calll use32
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    addl $8, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_b5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrl %cl, %edi
+; X64-NOBMI-NEXT:    movl $-1, %ebx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %ebx
+; X64-NOBMI-NEXT:    notl %ebx
+; X64-NOBMI-NEXT:    andl %edi, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %edi
+; X64-NOBMI-NEXT:    callq use32
+; X64-NOBMI-NEXT:    movl %ebx, %eax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X64-BMI1NOTBM-NEXT:    callq use32
+; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_b5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %ebx
+; X64-BMI1BMI2-NEXT:    movl %esi, %edi
+; X64-BMI1BMI2-NEXT:    callq use32
+; X64-BMI1BMI2-NEXT:    movl %ebx, %eax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %notmask = shl i32 -1, %numlowbits
+  %mask = xor i32 %notmask, -1
+  %masked = and i32 %mask, %shifted
+  call void @use32(i32 %numskipbits)
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_b0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB18_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB18_2:
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    shldl %cl, %edx, %edx
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB18_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB18_4:
+; X86-NOBMI-NEXT:    notl %edx
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB18_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB18_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB18_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB18_4:
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB18_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB18_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %edi
+; X86-BMI1BMI2-NEXT:    shlxl %eax, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB18_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB18_4:
+; X86-BMI1BMI2-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movq $-1, %rax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    notq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b1_indexzext(i64 %val, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB19_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB19_2:
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    shldl %cl, %edx, %edx
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB19_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB19_4:
+; X86-NOBMI-NEXT:    notl %edx
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB19_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB19_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB19_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB19_4:
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB19_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB19_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %edi
+; X86-BMI1BMI2-NEXT:    shlxl %eax, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB19_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB19_4:
+; X86-BMI1BMI2-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movq $-1, %rax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    notq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_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:    shrq %cl, %rdi
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %notmask = shl i64 -1, %conv
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB20_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB20_2:
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    shldl %cl, %edx, %edx
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB20_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB20_4:
+; X86-NOBMI-NEXT:    notl %edx
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl (%edx), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%edx), %edi
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB20_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB20_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB20_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB20_4:
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl (%edx), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%edx), %edi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB20_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB20_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %edi
+; X86-BMI1BMI2-NEXT:    shlxl %eax, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB20_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB20_4:
+; X86-BMI1BMI2-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq (%rdi), %rsi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rsi
+; X64-NOBMI-NEXT:    movq $-1, %rax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    notq %rax
+; X64-NOBMI-NEXT:    andq %rsi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movq $-1, %rsi
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rsi
+; X64-BMI1NOTBM-NEXT:    andnq %rax, %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b3_load_indexzext(i64* %w, i8 zeroext %numskipbits, i8 zeroext %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB21_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB21_2:
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    shldl %cl, %edx, %edx
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB21_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB21_4:
+; X86-NOBMI-NEXT:    notl %edx
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    notl %eax
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; 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:    movl (%edx), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%edx), %edi
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB21_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB21_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB21_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB21_4:
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl (%edx), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%edx), %edi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB21_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB21_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %edi
+; X86-BMI1BMI2-NEXT:    shlxl %eax, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB21_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB21_4:
+; X86-BMI1BMI2-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq (%rdi), %rsi
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rsi
+; X64-NOBMI-NEXT:    movq $-1, %rax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    notq %rax
+; X64-NOBMI-NEXT:    andq %rsi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_b3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; 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:    movq $-1, %rsi
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rsi
+; X64-BMI1NOTBM-NEXT:    andnq %rax, %rsi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %conv = zext i8 %numlowbits to i64
+  %notmask = shl i64 -1, %conv
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_b4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %ch
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl %esi, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB22_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB22_2:
+; X86-NOBMI-NEXT:    movl $-1, %edi
+; X86-NOBMI-NEXT:    movl $-1, %esi
+; X86-NOBMI-NEXT:    movb %ch, %cl
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    shldl %cl, %edi, %edi
+; X86-NOBMI-NEXT:    testb $32, %ch
+; X86-NOBMI-NEXT:    je .LBB22_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %esi, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:  .LBB22_4:
+; X86-NOBMI-NEXT:    notl %edi
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    notl %esi
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB22_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB22_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB22_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB22_4:
+; X86-BMI1NOTBM-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB22_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB22_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %edi
+; X86-BMI1BMI2-NEXT:    shlxl %eax, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB22_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB22_4:
+; X86-BMI1BMI2-NEXT:    andnl %edx, %edi, %edx
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebx, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movq $-1, %rax
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    notq %rax
+; X64-NOBMI-NEXT:    andq %rdi, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+define i64 @bextr64_b5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_b5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebp
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $12, %esp
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %esi, %ebp
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %ebp
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %ebx
+; X86-NOBMI-NEXT:    testb $32, %al
+; X86-NOBMI-NEXT:    je .LBB23_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %ebp, %ebx
+; X86-NOBMI-NEXT:    xorl %ebp, %ebp
+; X86-NOBMI-NEXT:  .LBB23_2:
+; X86-NOBMI-NEXT:    movl $-1, %esi
+; X86-NOBMI-NEXT:    movl $-1, %edi
+; X86-NOBMI-NEXT:    movl %edx, %ecx
+; X86-NOBMI-NEXT:    shll %cl, %edi
+; X86-NOBMI-NEXT:    shldl %cl, %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %dl
+; X86-NOBMI-NEXT:    je .LBB23_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB23_4:
+; X86-NOBMI-NEXT:    notl %esi
+; X86-NOBMI-NEXT:    andl %ebp, %esi
+; X86-NOBMI-NEXT:    notl %edi
+; X86-NOBMI-NEXT:    andl %ebx, %edi
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    calll use64
+; X86-NOBMI-NEXT:    addl $16, %esp
+; X86-NOBMI-NEXT:    movl %edi, %eax
+; X86-NOBMI-NEXT:    movl %esi, %edx
+; X86-NOBMI-NEXT:    addl $12, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    popl %ebp
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebp
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $12, %esp
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB23_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB23_2:
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %ebp
+; X86-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebp
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %dl
+; X86-BMI1NOTBM-NEXT:    je .LBB23_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %ebp, %ebx
+; X86-BMI1NOTBM-NEXT:    xorl %ebp, %ebp
+; X86-BMI1NOTBM-NEXT:  .LBB23_4:
+; X86-BMI1NOTBM-NEXT:    andnl %esi, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    andnl %edi, %ebp, %edi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    calll use64
+; X86-BMI1NOTBM-NEXT:    addl $16, %esp
+; X86-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edx
+; X86-BMI1NOTBM-NEXT:    addl $12, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    popl %ebp
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_b5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebp
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $12, %esp
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edi
+; X86-BMI1BMI2-NEXT:    shrxl %eax, %esi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB23_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %esi, %edi
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB23_2:
+; X86-BMI1BMI2-NEXT:    movl $-1, %ebp
+; X86-BMI1BMI2-NEXT:    shlxl %edx, %ebp, %ebx
+; X86-BMI1BMI2-NEXT:    movl %edx, %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %ebp, %ebp
+; X86-BMI1BMI2-NEXT:    testb $32, %dl
+; X86-BMI1BMI2-NEXT:    je .LBB23_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %ebp
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB23_4:
+; X86-BMI1BMI2-NEXT:    andnl %esi, %ebp, %esi
+; X86-BMI1BMI2-NEXT:    andnl %edi, %ebx, %edi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    pushl {{[0-9]+}}(%esp)
+; X86-BMI1BMI2-NEXT:    pushl %eax
+; X86-BMI1BMI2-NEXT:    calll use64
+; X86-BMI1BMI2-NEXT:    addl $16, %esp
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:    movl %esi, %edx
+; X86-BMI1BMI2-NEXT:    addl $12, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    popl %ebp
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_b5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rdi
+; X64-NOBMI-NEXT:    movq $-1, %rbx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rbx
+; X64-NOBMI-NEXT:    notq %rbx
+; X64-NOBMI-NEXT:    andq %rdi, %rbx
+; X64-NOBMI-NEXT:    movq %rsi, %rdi
+; X64-NOBMI-NEXT:    callq use64
+; X64-NOBMI-NEXT:    movq %rbx, %rax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_b5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
+; X64-BMI1NOTBM-NEXT:    callq use64
+; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_b5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
+; X64-BMI1BMI2-NEXT:    movq %rsi, %rdi
+; X64-BMI1BMI2-NEXT:    callq use64
+; X64-BMI1BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %notmask = shl i64 -1, %numlowbits
+  %mask = xor i64 %notmask, -1
+  %masked = and i64 %mask, %shifted
+  call void @use64(i64 %numskipbits)
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern c. 32-bit
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_c0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c2_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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c3_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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %mask = lshr i32 -1, %sh_prom
+  %masked = and i32 %mask, %shifted
+  ret i32 %masked
+}
+
+define i32 @bextr32_c4_commutative(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %shifted, %mask ; swapped order
+  ret i32 %masked
+}
+
+define i32 @bextr32_c5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_c5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl %eax, (%esp)
+; X86-NOBMI-NEXT:    calll use32
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    addl $8, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_c5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
+; X86-BMI1NOTBM-NEXT:    calll use32
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    addl $8, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_c5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %edx, %esi
+; X86-BMI1BMI2-NEXT:    movl %ecx, (%esp)
+; X86-BMI1BMI2-NEXT:    calll use32
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    addl $8, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_c5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %edi, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrl %cl, %ebx
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %ebx
+; X64-NOBMI-NEXT:    shrl %cl, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %edi
+; X64-NOBMI-NEXT:    callq use32
+; X64-NOBMI-NEXT:    movl %ebx, %eax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_c5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X64-BMI1NOTBM-NEXT:    callq use32
+; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_c5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %ebx
+; X64-BMI1BMI2-NEXT:    movl %esi, %edi
+; X64-BMI1BMI2-NEXT:    callq use32
+; X64-BMI1BMI2-NEXT:    movl %ebx, %eax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %mask = lshr i32 -1, %numhighbits
+  %masked = and i32 %mask, %shifted
+  call void @use32(i32 %numskipbits)
+  ret i32 %masked
+}
+
+; 64-bit
+
+define i64 @bextr64_c0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB30_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB30_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB30_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB30_4:
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB30_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB30_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB30_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB30_4:
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB30_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB30_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movl $-1, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB30_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB30_4:
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq %rdi, %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c0:
+; 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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB31_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB31_2:
+; X86-NOBMI-NEXT:    movb $64, %cl
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB31_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB31_4:
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB31_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB31_2:
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB31_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB31_4:
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB31_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB31_2:
+; X86-BMI1BMI2-NEXT:    movb $64, %cl
+; X86-BMI1BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl $-1, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB31_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB31_4:
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq %rdi, %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; 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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB32_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB32_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB32_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB32_4:
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB32_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB32_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB32_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB32_4:
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl (%eax), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%eax), %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB32_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB32_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movl $-1, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB32_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB32_4:
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq (%rdi), %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c2_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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %esi
+; X86-NOBMI-NEXT:    movl 4(%eax), %eax
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB33_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB33_2:
+; X86-NOBMI-NEXT:    movb $64, %cl
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl $-1, %eax
+; X86-NOBMI-NEXT:    movl $-1, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %eax, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB33_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB33_4:
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %esi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB33_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB33_2:
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB33_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB33_4:
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl (%eax), %esi
+; X86-BMI1BMI2-NEXT:    movl 4(%eax), %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB33_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB33_2:
+; X86-BMI1BMI2-NEXT:    movb $64, %cl
+; X86-BMI1BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl $-1, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %eax, %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB33_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB33_4:
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq (%rdi), %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; 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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %mask = lshr i64 -1, %sh_prom
+  %masked = and i64 %mask, %shifted
+  ret i64 %masked
+}
+
+define i64 @bextr64_c4_commutative(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c4_commutative:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl %esi, %edx
+; X86-NOBMI-NEXT:    shrl %cl, %edx
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB34_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:  .LBB34_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl $-1, %esi
+; X86-NOBMI-NEXT:    movl $-1, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB34_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB34_4:
+; X86-NOBMI-NEXT:    andl %edi, %edx
+; X86-NOBMI-NEXT:    andl %esi, %eax
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB34_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB34_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB34_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB34_4:
+; X86-BMI1NOTBM-NEXT:    andl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    andl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c4_commutative:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB34_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB34_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movl $-1, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB34_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB34_4:
+; X86-BMI1BMI2-NEXT:    andl %edi, %edx
+; X86-BMI1BMI2-NEXT:    andl %esi, %eax
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c4_commutative:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq %rdi, %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c4_commutative:
+; 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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c4_commutative:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %shifted, %mask ; swapped order
+  ret i64 %masked
+}
+
+define i64 @bextr64_c5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_c5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebp
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $12, %esp
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %esi, %ebx
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %ebx
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %edx
+; X86-NOBMI-NEXT:    testb $32, %al
+; X86-NOBMI-NEXT:    je .LBB35_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %ebx, %edx
+; X86-NOBMI-NEXT:    xorl %ebx, %ebx
+; X86-NOBMI-NEXT:  .LBB35_2:
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    movl $-1, %esi
+; X86-NOBMI-NEXT:    movl $-1, %edi
+; X86-NOBMI-NEXT:    shrl %cl, %edi
+; X86-NOBMI-NEXT:    shrdl %cl, %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB35_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    xorl %edi, %edi
+; X86-NOBMI-NEXT:  .LBB35_4:
+; X86-NOBMI-NEXT:    andl %ebx, %edi
+; X86-NOBMI-NEXT:    andl %edx, %esi
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    pushl %ebp
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    calll use64
+; X86-NOBMI-NEXT:    addl $16, %esp
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    movl %edi, %edx
+; X86-NOBMI-NEXT:    addl $12, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    popl %ebp
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_c5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebp
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $12, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %edx
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB35_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB35_2:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB35_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %edi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB35_4:
+; X86-BMI1NOTBM-NEXT:    andl %ebx, %edi
+; X86-BMI1NOTBM-NEXT:    andl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    pushl %ebp
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    calll use64
+; X86-BMI1NOTBM-NEXT:    addl $16, %esp
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    addl $12, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    popl %ebp
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_c5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebp
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $12, %esp
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edx
+; X86-BMI1BMI2-NEXT:    shrxl %eax, %esi, %ebx
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB35_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edx
+; X86-BMI1BMI2-NEXT:    xorl %ebx, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB35_2:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movl $-1, %esi
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %edi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB35_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edi, %edi
+; X86-BMI1BMI2-NEXT:  .LBB35_4:
+; X86-BMI1BMI2-NEXT:    andl %ebx, %edi
+; X86-BMI1BMI2-NEXT:    andl %edx, %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    pushl %ebp
+; X86-BMI1BMI2-NEXT:    pushl %eax
+; X86-BMI1BMI2-NEXT:    calll use64
+; X86-BMI1BMI2-NEXT:    addl $16, %esp
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    movl %edi, %edx
+; X86-BMI1BMI2-NEXT:    addl $12, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    popl %ebp
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_c5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movq %rdi, %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rbx
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rbx
+; X64-NOBMI-NEXT:    shrq %cl, %rbx
+; X64-NOBMI-NEXT:    movq %rsi, %rdi
+; X64-NOBMI-NEXT:    callq use64
+; X64-NOBMI-NEXT:    movq %rbx, %rax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_c5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
+; X64-BMI1NOTBM-NEXT:    callq use64
+; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_c5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
+; X64-BMI1BMI2-NEXT:    movq %rsi, %rdi
+; X64-BMI1BMI2-NEXT:    callq use64
+; X64-BMI1BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %mask = lshr i64 -1, %numhighbits
+  %masked = and i64 %mask, %shifted
+  call void @use64(i64 %numskipbits)
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Pattern d. 32-bit.
+; ---------------------------------------------------------------------------- ;
+
+define i32 @bextr32_d0(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_d0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_d0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_d0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_d0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_d0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_d0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %highbitscleared = shl i32 %shifted, %numhighbits
+  %masked = lshr i32 %highbitscleared, %numhighbits
+  ret i32 %masked
+}
+
+define i32 @bextr32_d1_indexzext(i32 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_d1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_d1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_d1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl %edi, %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_d1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %highbitscleared = shl i32 %shifted, %sh_prom
+  %masked = lshr i32 %highbitscleared, %sh_prom
+  ret i32 %masked
+}
+
+define i32 @bextr32_d2_load(i32* %w, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_d2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_d2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_d2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_d2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_d2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %highbitscleared = shl i32 %shifted, %numhighbits
+  %masked = lshr i32 %highbitscleared, %numhighbits
+  ret i32 %masked
+}
+
+define i32 @bextr32_d3_load_indexzext(i32* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_d3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shll %cl, %eax
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %dl
+; X86-BMI1BMI2-NEXT:    shrxl %edx, (%ecx), %ecx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %ecx, %eax
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_d3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movl (%rdi), %eax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %eax
+; X64-NOBMI-NEXT:    shrl %cl, %eax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxl %esi, (%rdi), %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %eax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i32, i32* %w
+  %skip = zext i8 %numskipbits to i32
+  %shifted = lshr i32 %val, %skip
+  %numhighbits = sub i8 32, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i32
+  %highbitscleared = shl i32 %shifted, %sh_prom
+  %masked = lshr i32 %highbitscleared, %sh_prom
+  ret i32 %masked
+}
+
+define i32 @bextr32_d5_skipextrauses(i32 %val, i32 %numskipbits, i32 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr32_d5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    xorl %ecx, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shll %cl, %esi
+; X86-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    movl %eax, (%esp)
+; X86-NOBMI-NEXT:    calll use32
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    addl $8, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr32_d5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
+; X86-BMI1NOTBM-NEXT:    calll use32
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    addl $8, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr32_d5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    bzhil %eax, %edx, %esi
+; X86-BMI1BMI2-NEXT:    movl %ecx, (%esp)
+; X86-BMI1BMI2-NEXT:    calll use32
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    addl $8, %esp
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr32_d5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movl %edi, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrl %cl, %ebx
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shll %cl, %ebx
+; X64-NOBMI-NEXT:    shrl %cl, %ebx
+; X64-NOBMI-NEXT:    movl %esi, %edi
+; X64-NOBMI-NEXT:    callq use32
+; X64-NOBMI-NEXT:    movl %ebx, %eax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr32_d5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X64-BMI1NOTBM-NEXT:    callq use32
+; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr32_d5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxl %esi, %edi, %eax
+; X64-BMI1BMI2-NEXT:    bzhil %edx, %eax, %ebx
+; X64-BMI1BMI2-NEXT:    movl %esi, %edi
+; X64-BMI1BMI2-NEXT:    callq use32
+; X64-BMI1BMI2-NEXT:    movl %ebx, %eax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i32 %val, %numskipbits
+  %numhighbits = sub i32 32, %numlowbits
+  %highbitscleared = shl i32 %shifted, %numhighbits
+  %masked = lshr i32 %highbitscleared, %numhighbits
+  call void @use32(i32 %numskipbits)
+  ret i32 %masked
+}
+
+; 64-bit.
+
+define i64 @bextr64_d0(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_d0:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB41_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB41_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
+; X86-NOBMI-NEXT:    shll %cl, %edi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl %edi, %ebx
+; X86-NOBMI-NEXT:    jne .LBB41_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %ebx
+; X86-NOBMI-NEXT:  .LBB41_4:
+; X86-NOBMI-NEXT:    movl %ebx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl $0, %edx
+; X86-NOBMI-NEXT:    jne .LBB41_6
+; X86-NOBMI-NEXT:  # %bb.5:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:  .LBB41_6:
+; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    jne .LBB41_8
+; X86-NOBMI-NEXT:  # %bb.7:
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:  .LBB41_8:
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_d0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB41_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB41_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    jne .LBB41_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB41_4:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB41_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB41_6:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB41_8
+; X86-BMI1NOTBM-NEXT:  # %bb.7:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB41_8:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_d0:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB41_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB41_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB41_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    movl $0, %edi
+; X86-BMI1BMI2-NEXT:  .LBB41_4:
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %eax
+; X86-BMI1BMI2-NEXT:    jne .LBB41_6
+; X86-BMI1BMI2-NEXT:  # %bb.5:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:  .LBB41_6:
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    jne .LBB41_8
+; X86-BMI1BMI2-NEXT:  # %bb.7:
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:  .LBB41_8:
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_d0:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq %rdi, %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_d0:
+; 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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_d0:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %highbitscleared = shl i64 %shifted, %numhighbits
+  %masked = lshr i64 %highbitscleared, %numhighbits
+  ret i64 %masked
+}
+
+define i64 @bextr64_d1_indexzext(i64 %val, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_d1_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB42_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB42_2:
+; X86-NOBMI-NEXT:    movb $64, %cl
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
+; X86-NOBMI-NEXT:    shll %cl, %edi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl %edi, %ebx
+; X86-NOBMI-NEXT:    jne .LBB42_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %ebx
+; X86-NOBMI-NEXT:  .LBB42_4:
+; X86-NOBMI-NEXT:    movl %ebx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl $0, %edx
+; X86-NOBMI-NEXT:    jne .LBB42_6
+; X86-NOBMI-NEXT:  # %bb.5:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:  .LBB42_6:
+; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    jne .LBB42_8
+; X86-NOBMI-NEXT:  # %bb.7:
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:  .LBB42_8:
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_d1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB42_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB42_2:
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    jne .LBB42_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB42_4:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB42_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB42_6:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB42_8
+; X86-BMI1NOTBM-NEXT:  # %bb.7:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB42_8:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_d1_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %esi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB42_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB42_2:
+; X86-BMI1BMI2-NEXT:    movb $64, %cl
+; X86-BMI1BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB42_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    movl $0, %edi
+; X86-BMI1BMI2-NEXT:  .LBB42_4:
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %eax
+; X86-BMI1BMI2-NEXT:    jne .LBB42_6
+; X86-BMI1BMI2-NEXT:  # %bb.5:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:  .LBB42_6:
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    jne .LBB42_8
+; X86-BMI1BMI2-NEXT:  # %bb.7:
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:  .LBB42_8:
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_d1_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq %rdi, %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_d1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; 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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_d1_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %highbitscleared = shl i64 %shifted, %sh_prom
+  %masked = lshr i64 %highbitscleared, %sh_prom
+  ret i64 %masked
+}
+
+define i64 @bextr64_d2_load(i64* %w, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_d2_load:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %edi
+; X86-NOBMI-NEXT:    movl 4(%eax), %edx
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB43_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB43_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
+; X86-NOBMI-NEXT:    shll %cl, %edi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl %edi, %ebx
+; X86-NOBMI-NEXT:    jne .LBB43_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %ebx
+; X86-NOBMI-NEXT:  .LBB43_4:
+; X86-NOBMI-NEXT:    movl %ebx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl $0, %edx
+; X86-NOBMI-NEXT:    jne .LBB43_6
+; X86-NOBMI-NEXT:  # %bb.5:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:  .LBB43_6:
+; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    jne .LBB43_8
+; X86-NOBMI-NEXT:  # %bb.7:
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:  .LBB43_8:
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_d2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %edi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %edx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB43_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB43_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    jne .LBB43_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB43_4:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB43_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB43_6:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB43_8
+; X86-BMI1NOTBM-NEXT:  # %bb.7:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB43_8:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_d2_load:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl (%edx), %eax
+; X86-BMI1BMI2-NEXT:    movl 4(%edx), %edx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %esi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB43_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB43_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB43_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    movl $0, %edi
+; X86-BMI1BMI2-NEXT:  .LBB43_4:
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %eax
+; X86-BMI1BMI2-NEXT:    jne .LBB43_6
+; X86-BMI1BMI2-NEXT:  # %bb.5:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:  .LBB43_6:
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    jne .LBB43_8
+; X86-BMI1BMI2-NEXT:  # %bb.7:
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:  .LBB43_8:
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_d2_load:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movq %rsi, %rcx
+; X64-NOBMI-NEXT:    movq (%rdi), %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; 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:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_d2_load:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %highbitscleared = shl i64 %shifted, %numhighbits
+  %masked = lshr i64 %highbitscleared, %numhighbits
+  ret i64 %masked
+}
+
+define i64 @bextr64_d3_load_indexzext(i64* %w, i8 %numskipbits, i8 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_d3_load_indexzext:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl (%eax), %edi
+; X86-NOBMI-NEXT:    movl 4(%eax), %edx
+; X86-NOBMI-NEXT:    movl %edx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    shrdl %cl, %edx, %edi
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    je .LBB44_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %eax, %edi
+; X86-NOBMI-NEXT:    xorl %eax, %eax
+; X86-NOBMI-NEXT:  .LBB44_2:
+; X86-NOBMI-NEXT:    movb $64, %cl
+; X86-NOBMI-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-NOBMI-NEXT:    shldl %cl, %edi, %eax
+; X86-NOBMI-NEXT:    shll %cl, %edi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl %edi, %ebx
+; X86-NOBMI-NEXT:    jne .LBB44_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %eax, %ebx
+; X86-NOBMI-NEXT:  .LBB44_4:
+; X86-NOBMI-NEXT:    movl %ebx, %eax
+; X86-NOBMI-NEXT:    shrl %cl, %eax
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl $0, %edx
+; X86-NOBMI-NEXT:    jne .LBB44_6
+; X86-NOBMI-NEXT:  # %bb.5:
+; X86-NOBMI-NEXT:    movl %edi, %esi
+; X86-NOBMI-NEXT:    movl %eax, %edx
+; X86-NOBMI-NEXT:  .LBB44_6:
+; X86-NOBMI-NEXT:    shrdl %cl, %ebx, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    jne .LBB44_8
+; X86-NOBMI-NEXT:  # %bb.7:
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:  .LBB44_8:
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_d3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %edi
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %edx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edx, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB44_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB44_2:
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    jne .LBB44_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ebx
+; X86-BMI1NOTBM-NEXT:  .LBB44_4:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB44_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB44_6:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB44_8
+; X86-BMI1NOTBM-NEXT:  # %bb.7:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB44_8:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_d3_load_indexzext:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl (%edx), %eax
+; X86-BMI1BMI2-NEXT:    movl 4(%edx), %edx
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %esi
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %eax
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB44_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %esi, %eax
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB44_2:
+; X86-BMI1BMI2-NEXT:    movb $64, %cl
+; X86-BMI1BMI2-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1BMI2-NEXT:    shldl %cl, %eax, %esi
+; X86-BMI1BMI2-NEXT:    shlxl %ecx, %eax, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB44_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:    movl $0, %edi
+; X86-BMI1BMI2-NEXT:  .LBB44_4:
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %esi, %eax
+; X86-BMI1BMI2-NEXT:    jne .LBB44_6
+; X86-BMI1BMI2-NEXT:  # %bb.5:
+; X86-BMI1BMI2-NEXT:    movl %eax, %edx
+; X86-BMI1BMI2-NEXT:  .LBB44_6:
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %esi, %edi
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    jne .LBB44_8
+; X86-BMI1BMI2-NEXT:  # %bb.7:
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:  .LBB44_8:
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_d3_load_indexzext:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    movq (%rdi), %rax
+; X64-NOBMI-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    negb %dl
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rax
+; X64-NOBMI-NEXT:    shrq %cl, %rax
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_d3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; 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:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_d3_load_indexzext:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    # kill: def $edx killed $edx def $rdx
+; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, (%rdi), %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rax
+; X64-BMI1BMI2-NEXT:    retq
+  %val = load i64, i64* %w
+  %skip = zext i8 %numskipbits to i64
+  %shifted = lshr i64 %val, %skip
+  %numhighbits = sub i8 64, %numlowbits
+  %sh_prom = zext i8 %numhighbits to i64
+  %highbitscleared = shl i64 %shifted, %sh_prom
+  %masked = lshr i64 %highbitscleared, %sh_prom
+  ret i64 %masked
+}
+
+define i64 @bextr64_d5_skipextrauses(i64 %val, i64 %numskipbits, i64 %numlowbits) nounwind {
+; X86-NOBMI-LABEL: bextr64_d5_skipextrauses:
+; X86-NOBMI:       # %bb.0:
+; X86-NOBMI-NEXT:    pushl %ebp
+; X86-NOBMI-NEXT:    pushl %ebx
+; X86-NOBMI-NEXT:    pushl %edi
+; X86-NOBMI-NEXT:    pushl %esi
+; X86-NOBMI-NEXT:    subl $12, %esp
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NOBMI-NEXT:    movl %edx, %esi
+; X86-NOBMI-NEXT:    movl %eax, %ecx
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    shrdl %cl, %edx, %ebx
+; X86-NOBMI-NEXT:    xorl %edx, %edx
+; X86-NOBMI-NEXT:    testb $32, %al
+; X86-NOBMI-NEXT:    je .LBB45_2
+; X86-NOBMI-NEXT:  # %bb.1:
+; X86-NOBMI-NEXT:    movl %esi, %ebx
+; X86-NOBMI-NEXT:    xorl %esi, %esi
+; X86-NOBMI-NEXT:  .LBB45_2:
+; X86-NOBMI-NEXT:    movl $64, %ecx
+; X86-NOBMI-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    shldl %cl, %ebx, %esi
+; X86-NOBMI-NEXT:    shll %cl, %ebx
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl %ebx, %ebp
+; X86-NOBMI-NEXT:    jne .LBB45_4
+; X86-NOBMI-NEXT:  # %bb.3:
+; X86-NOBMI-NEXT:    movl %esi, %ebp
+; X86-NOBMI-NEXT:  .LBB45_4:
+; X86-NOBMI-NEXT:    movl %ebp, %esi
+; X86-NOBMI-NEXT:    shrl %cl, %esi
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl $0, %edi
+; X86-NOBMI-NEXT:    jne .LBB45_6
+; X86-NOBMI-NEXT:  # %bb.5:
+; X86-NOBMI-NEXT:    movl %ebx, %edx
+; X86-NOBMI-NEXT:    movl %esi, %edi
+; X86-NOBMI-NEXT:  .LBB45_6:
+; X86-NOBMI-NEXT:    shrdl %cl, %ebp, %edx
+; X86-NOBMI-NEXT:    testb $32, %cl
+; X86-NOBMI-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NOBMI-NEXT:    jne .LBB45_8
+; X86-NOBMI-NEXT:  # %bb.7:
+; X86-NOBMI-NEXT:    movl %edx, %esi
+; X86-NOBMI-NEXT:  .LBB45_8:
+; X86-NOBMI-NEXT:    subl $8, %esp
+; X86-NOBMI-NEXT:    pushl %ecx
+; X86-NOBMI-NEXT:    pushl %eax
+; X86-NOBMI-NEXT:    calll use64
+; X86-NOBMI-NEXT:    addl $16, %esp
+; X86-NOBMI-NEXT:    movl %esi, %eax
+; X86-NOBMI-NEXT:    movl %edi, %edx
+; X86-NOBMI-NEXT:    addl $12, %esp
+; X86-NOBMI-NEXT:    popl %esi
+; X86-NOBMI-NEXT:    popl %edi
+; X86-NOBMI-NEXT:    popl %ebx
+; X86-NOBMI-NEXT:    popl %ebp
+; X86-NOBMI-NEXT:    retl
+;
+; X86-BMI1NOTBM-LABEL: bextr64_d5_skipextrauses:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebp
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    subl $12, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edx, %ebx
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    testb $32, %al
+; X86-BMI1NOTBM-NEXT:    je .LBB45_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB45_2:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %ebx, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %ebp
+; X86-BMI1NOTBM-NEXT:    jne .LBB45_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebp
+; X86-BMI1NOTBM-NEXT:  .LBB45_4:
+; X86-BMI1NOTBM-NEXT:    movl %ebp, %esi
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edi
+; X86-BMI1NOTBM-NEXT:    jne .LBB45_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %edx
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB45_6:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %ebp, %edx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    jne .LBB45_8
+; X86-BMI1NOTBM-NEXT:  # %bb.7:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB45_8:
+; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    pushl %ecx
+; X86-BMI1NOTBM-NEXT:    pushl %eax
+; X86-BMI1NOTBM-NEXT:    calll use64
+; X86-BMI1NOTBM-NEXT:    addl $16, %esp
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    movl %edi, %edx
+; X86-BMI1NOTBM-NEXT:    addl $12, %esp
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    popl %ebp
+; X86-BMI1NOTBM-NEXT:    retl
+;
+; X86-BMI1BMI2-LABEL: bextr64_d5_skipextrauses:
+; X86-BMI1BMI2:       # %bb.0:
+; X86-BMI1BMI2-NEXT:    pushl %ebx
+; X86-BMI1BMI2-NEXT:    pushl %edi
+; X86-BMI1BMI2-NEXT:    pushl %esi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1BMI2-NEXT:    movl %eax, %ecx
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %edi
+; X86-BMI1BMI2-NEXT:    shrxl %eax, %edx, %edx
+; X86-BMI1BMI2-NEXT:    xorl %esi, %esi
+; X86-BMI1BMI2-NEXT:    testb $32, %al
+; X86-BMI1BMI2-NEXT:    je .LBB45_2
+; X86-BMI1BMI2-NEXT:  # %bb.1:
+; X86-BMI1BMI2-NEXT:    movl %edx, %edi
+; X86-BMI1BMI2-NEXT:    xorl %edx, %edx
+; X86-BMI1BMI2-NEXT:  .LBB45_2:
+; X86-BMI1BMI2-NEXT:    movl $64, %ecx
+; X86-BMI1BMI2-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    shldl %cl, %edi, %edx
+; X86-BMI1BMI2-NEXT:    shlxl %ecx, %edi, %ebx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    je .LBB45_4
+; X86-BMI1BMI2-NEXT:  # %bb.3:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edx
+; X86-BMI1BMI2-NEXT:    movl $0, %ebx
+; X86-BMI1BMI2-NEXT:  .LBB45_4:
+; X86-BMI1BMI2-NEXT:    shrxl %ecx, %edx, %edi
+; X86-BMI1BMI2-NEXT:    jne .LBB45_6
+; X86-BMI1BMI2-NEXT:  # %bb.5:
+; X86-BMI1BMI2-NEXT:    movl %edi, %esi
+; X86-BMI1BMI2-NEXT:  .LBB45_6:
+; X86-BMI1BMI2-NEXT:    shrdl %cl, %edx, %ebx
+; X86-BMI1BMI2-NEXT:    testb $32, %cl
+; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1BMI2-NEXT:    jne .LBB45_8
+; X86-BMI1BMI2-NEXT:  # %bb.7:
+; X86-BMI1BMI2-NEXT:    movl %ebx, %edi
+; X86-BMI1BMI2-NEXT:  .LBB45_8:
+; X86-BMI1BMI2-NEXT:    subl $8, %esp
+; X86-BMI1BMI2-NEXT:    pushl %ecx
+; X86-BMI1BMI2-NEXT:    pushl %eax
+; X86-BMI1BMI2-NEXT:    calll use64
+; X86-BMI1BMI2-NEXT:    addl $16, %esp
+; X86-BMI1BMI2-NEXT:    movl %edi, %eax
+; X86-BMI1BMI2-NEXT:    movl %esi, %edx
+; X86-BMI1BMI2-NEXT:    popl %esi
+; X86-BMI1BMI2-NEXT:    popl %edi
+; X86-BMI1BMI2-NEXT:    popl %ebx
+; X86-BMI1BMI2-NEXT:    retl
+;
+; X64-NOBMI-LABEL: bextr64_d5_skipextrauses:
+; X64-NOBMI:       # %bb.0:
+; X64-NOBMI-NEXT:    pushq %rbx
+; X64-NOBMI-NEXT:    movq %rdi, %rbx
+; X64-NOBMI-NEXT:    movl %esi, %ecx
+; X64-NOBMI-NEXT:    shrq %cl, %rbx
+; X64-NOBMI-NEXT:    negl %edx
+; X64-NOBMI-NEXT:    movl %edx, %ecx
+; X64-NOBMI-NEXT:    shlq %cl, %rbx
+; X64-NOBMI-NEXT:    shrq %cl, %rbx
+; X64-NOBMI-NEXT:    movq %rsi, %rdi
+; X64-NOBMI-NEXT:    callq use64
+; X64-NOBMI-NEXT:    movq %rbx, %rax
+; X64-NOBMI-NEXT:    popq %rbx
+; X64-NOBMI-NEXT:    retq
+;
+; X64-BMI1NOTBM-LABEL: bextr64_d5_skipextrauses:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
+; X64-BMI1NOTBM-NEXT:    callq use64
+; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
+; X64-BMI1NOTBM-NEXT:    popq %rbx
+; X64-BMI1NOTBM-NEXT:    retq
+;
+; X64-BMI1BMI2-LABEL: bextr64_d5_skipextrauses:
+; X64-BMI1BMI2:       # %bb.0:
+; X64-BMI1BMI2-NEXT:    pushq %rbx
+; X64-BMI1BMI2-NEXT:    shrxq %rsi, %rdi, %rax
+; X64-BMI1BMI2-NEXT:    bzhiq %rdx, %rax, %rbx
+; X64-BMI1BMI2-NEXT:    movq %rsi, %rdi
+; X64-BMI1BMI2-NEXT:    callq use64
+; X64-BMI1BMI2-NEXT:    movq %rbx, %rax
+; X64-BMI1BMI2-NEXT:    popq %rbx
+; X64-BMI1BMI2-NEXT:    retq
+  %shifted = lshr i64 %val, %numskipbits
+  %numhighbits = sub i64 64, %numlowbits
+  %highbitscleared = shl i64 %shifted, %numhighbits
+  %masked = lshr i64 %highbitscleared, %numhighbits
+  call void @use64(i64 %numskipbits)
+  ret i64 %masked
+}
+
+; ---------------------------------------------------------------------------- ;
+; Constant
+; ---------------------------------------------------------------------------- ;
+
+define void @pr38938(i32* %a0, i64* %a1) {
+; X86-LABEL: pr38938:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X86-NEXT:    movl (%ecx), %ecx
+; X86-NEXT:    shrl $19, %ecx
+; X86-NEXT:    andl $4092, %ecx # imm = 0xFFC
+; X86-NEXT:    incl (%eax,%ecx)
+; X86-NEXT:    retl
+;
+; X64-LABEL: pr38938:
+; X64:       # %bb.0:
+; X64-NEXT:    movq (%rsi), %rax
+; X64-NEXT:    shrq $19, %rax
+; X64-NEXT:    andl $4092, %eax # imm = 0xFFC
+; X64-NEXT:    incl (%rdi,%rax)
+; X64-NEXT:    retq
+  %tmp = load i64, i64* %a1, align 8
+  %tmp1 = lshr i64 %tmp, 21
+  %tmp2 = and i64 %tmp1, 1023
+  %tmp3 = getelementptr inbounds i32, i32* %a0, i64 %tmp2
+  %tmp4 = load i32, i32* %tmp3, align 4
+  %tmp5 = add nsw i32 %tmp4, 1
+  store i32 %tmp5, i32* %tmp3, align 4
+  ret void
+}

Modified: llvm/trunk/test/CodeGen/X86/extract-lowbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/extract-lowbits.ll?rev=342623&r1=342622&r2=342623&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/extract-lowbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/extract-lowbits.ll Thu Sep 20 00:54:49 2018
@@ -1,11 +1,11 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,NOBMI,X86-NOBMI
-; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1
+; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1NOTBM,X86-BMI1NOTBM
 ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1TBM,X86-BMI1TBM
 ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1TBM,X86-BMI1TBM,BMI1TBMBMI2,X86-BMI1TBMBMI2
 ; RUN: llc -mtriple=i686-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X86,BMI1,X86-BMI1,BMI1BMI2,X86-BMI1BMI2,BMI1NOTBMBMI2,X86-BMI1NOTBMBMI2
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=-bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,NOBMI,X64-NOBMI
-; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1NOTBM,X64-BMI1NOTBM
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,-bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1TBM,X64-BMI1TBM
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,+tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1TBM,X64-BMI1TBM,BMI1TBMBMI2,X64-BMI1TBMBMI2
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+bmi,-tbm,+bmi2 < %s | FileCheck %s --check-prefixes=CHECK,X64,BMI1,X64-BMI1,BMI1BMI2,X64-BMI1BMI2,BMI1NOTBMBMI2,X64-BMI1NOTBMBMI2
@@ -37,6 +37,15 @@ define i32 @bzhi32_a0(i32 %val, i32 %num
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_a0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_a0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -53,6 +62,16 @@ define i32 @bzhi32_a0(i32 %val, i32 %num
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_a0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_a0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -73,6 +92,15 @@ define i32 @bzhi32_a1_indexzext(i32 %val
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_a1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_a1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -89,6 +117,16 @@ define i32 @bzhi32_a1_indexzext(i32 %val
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_a1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_a1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -111,6 +149,16 @@ define i32 @bzhi32_a2_load(i32* %w, i32
 ; X86-NOBMI-NEXT:    andl (%edx), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_a2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl (%edx), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_a2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -128,6 +176,16 @@ define i32 @bzhi32_a2_load(i32* %w, i32
 ; X64-NOBMI-NEXT:    andl (%rdi), %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_a2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_a2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -150,6 +208,16 @@ define i32 @bzhi32_a3_load_indexzext(i32
 ; X86-NOBMI-NEXT:    andl (%edx), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_a3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl (%edx), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_a3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -167,6 +235,16 @@ define i32 @bzhi32_a3_load_indexzext(i32
 ; X64-NOBMI-NEXT:    andl (%rdi), %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_a3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_a3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -189,6 +267,15 @@ define i32 @bzhi32_a4_commutative(i32 %v
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_a4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    decl %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_a4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -205,6 +292,16 @@ define i32 @bzhi32_a4_commutative(i32 %v
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_a4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    decl %eax
+; X64-BMI1NOTBM-NEXT:    andl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_a4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -237,6 +334,25 @@ define i64 @bzhi64_a0(i64 %val, i64 %num
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_a0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB5_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB5_2:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_a0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
@@ -266,6 +382,16 @@ define i64 @bzhi64_a0(i64 %val, i64 %num
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_a0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_a0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -296,6 +422,25 @@ define i64 @bzhi64_a1_indexzext(i64 %val
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_a1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB6_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB6_2:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_a1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
@@ -325,6 +470,16 @@ define i64 @bzhi64_a1_indexzext(i64 %val
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_a1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_a1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -360,6 +515,28 @@ define i64 @bzhi64_a2_load(i64* %w, i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_a2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB7_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB7_2:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl 4(%esi), %edx
+; X86-BMI1NOTBM-NEXT:    andl (%esi), %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_a2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -392,6 +569,16 @@ define i64 @bzhi64_a2_load(i64* %w, i64
 ; X64-NOBMI-NEXT:    andq (%rdi), %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_a2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_a2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, (%rdi), %rax
@@ -426,6 +613,28 @@ define i64 @bzhi64_a3_load_indexzext(i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_a3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB8_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB8_2:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl 4(%esi), %edx
+; X86-BMI1NOTBM-NEXT:    andl (%esi), %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_a3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -458,6 +667,16 @@ define i64 @bzhi64_a3_load_indexzext(i64
 ; X64-NOBMI-NEXT:    andq (%rdi), %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_a3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_a3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -491,6 +710,25 @@ define i64 @bzhi64_a4_commutative(i64 %v
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_a4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $1, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB9_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %eax, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB9_2:
+; X86-BMI1NOTBM-NEXT:    addl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    adcl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_a4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %cl
@@ -520,6 +758,16 @@ define i64 @bzhi64_a4_commutative(i64 %v
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_a4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movl $1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    decq %rax
+; X64-BMI1NOTBM-NEXT:    andq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_a4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -544,6 +792,14 @@ define i32 @bzhi32_b0(i32 %val, i32 %num
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_b0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_b0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -560,6 +816,15 @@ define i32 @bzhi32_b0(i32 %val, i32 %num
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_b0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_b0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -580,6 +845,14 @@ define i32 @bzhi32_b1_indexzext(i32 %val
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_b1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_b1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -596,6 +869,15 @@ define i32 @bzhi32_b1_indexzext(i32 %val
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_b1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_b1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -618,6 +900,15 @@ define i32 @bzhi32_b2_load(i32* %w, i32
 ; X86-NOBMI-NEXT:    andl (%edx), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_b2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edx
+; X86-BMI1NOTBM-NEXT:    andnl (%eax), %edx, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_b2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -635,6 +926,15 @@ define i32 @bzhi32_b2_load(i32* %w, i32
 ; X64-NOBMI-NEXT:    andl (%rdi), %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_b2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl (%rdi), %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_b2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -657,6 +957,15 @@ define i32 @bzhi32_b3_load_indexzext(i32
 ; X86-NOBMI-NEXT:    andl (%edx), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %edx
+; X86-BMI1NOTBM-NEXT:    andnl (%eax), %edx, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -674,6 +983,15 @@ define i32 @bzhi32_b3_load_indexzext(i32
 ; X64-NOBMI-NEXT:    andl (%rdi), %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_b3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl (%rdi), %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_b3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -696,6 +1014,14 @@ define i32 @bzhi32_b4_commutative(i32 %v
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_b4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_b4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -712,6 +1038,15 @@ define i32 @bzhi32_b4_commutative(i32 %v
 ; X64-NOBMI-NEXT:    andl %edi, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_b4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    andnl %edi, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_b4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -744,6 +1079,25 @@ define i64 @bzhi64_b0(i64 %val, i64 %num
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_b0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB15_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB15_2:
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %edx
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_b0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -772,6 +1126,15 @@ define i64 @bzhi64_b0(i64 %val, i64 %num
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_b0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_b0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -802,6 +1165,25 @@ define i64 @bzhi64_b1_indexzext(i64 %val
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_b1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB16_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB16_2:
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %edx
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_b1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -830,6 +1212,15 @@ define i64 @bzhi64_b1_indexzext(i64 %val
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_b1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_b1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -865,6 +1256,26 @@ define i64 @bzhi64_b2_load(i64* %w, i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_b2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %edx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB17_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB17_2:
+; X86-BMI1NOTBM-NEXT:    andnl 4(%eax), %edx, %edx
+; X86-BMI1NOTBM-NEXT:    andnl (%eax), %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_b2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -894,6 +1305,15 @@ define i64 @bzhi64_b2_load(i64* %w, i64
 ; X64-NOBMI-NEXT:    andq (%rdi), %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_b2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq (%rdi), %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_b2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, (%rdi), %rax
@@ -928,6 +1348,26 @@ define i64 @bzhi64_b3_load_indexzext(i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_b3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %edx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB18_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edx
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB18_2:
+; X86-BMI1NOTBM-NEXT:    andnl 4(%eax), %edx, %edx
+; X86-BMI1NOTBM-NEXT:    andnl (%eax), %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -957,6 +1397,15 @@ define i64 @bzhi64_b3_load_indexzext(i64
 ; X64-NOBMI-NEXT:    andq (%rdi), %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_b3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq (%rdi), %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_b3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -990,6 +1439,25 @@ define i64 @bzhi64_b4_commutative(i64 %v
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_b4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB19_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %esi, %esi
+; X86-BMI1NOTBM-NEXT:  .LBB19_2:
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %eax, %edx
+; X86-BMI1NOTBM-NEXT:    andnl {{[0-9]+}}(%esp), %esi, %eax
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_b4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -1018,6 +1486,15 @@ define i64 @bzhi64_b4_commutative(i64 %v
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_b4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq $-1, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    andnq %rdi, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_b4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -1043,6 +1520,16 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_c0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_c0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1059,6 +1546,16 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_c0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_c0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -1080,6 +1577,16 @@ define i32 @bzhi32_c1_indexzext(i32 %val
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_c1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -1096,6 +1603,16 @@ define i32 @bzhi32_c1_indexzext(i32 %val
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_c1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -1119,6 +1636,17 @@ define i32 @bzhi32_c2_load(i32* %w, i32
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_c2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_c2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1136,6 +1664,16 @@ define i32 @bzhi32_c2_load(i32* %w, i32
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_c2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_c2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -1159,6 +1697,17 @@ define i32 @bzhi32_c3_load_indexzext(i32
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1176,6 +1725,16 @@ define i32 @bzhi32_c3_load_indexzext(i32
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -1199,6 +1758,16 @@ define i32 @bzhi32_c4_commutative(i32 %v
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_c4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_c4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1215,6 +1784,16 @@ define i32 @bzhi32_c4_commutative(i32 %v
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_c4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_c4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -1246,6 +1825,24 @@ define i64 @bzhi64_c0(i64 %val, i64 %num
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %edx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_c0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB25_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB25_2:
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_c0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl $64, %ecx
@@ -1273,6 +1870,16 @@ define i64 @bzhi64_c0(i64 %val, i64 %num
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_c0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_c0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -1302,6 +1909,24 @@ define i64 @bzhi64_c1_indexzext(i64 %val
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %edx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_c1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB26_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB26_2:
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_c1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb $64, %cl
@@ -1329,6 +1954,16 @@ define i64 @bzhi64_c1_indexzext(i64 %val
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_c1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_c1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -1363,6 +1998,27 @@ define i64 @bzhi64_c2_load(i64* %w, i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_c2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB27_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB27_2:
+; X86-BMI1NOTBM-NEXT:    andl (%esi), %eax
+; X86-BMI1NOTBM-NEXT:    andl 4(%esi), %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_c2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -1393,6 +2049,16 @@ define i64 @bzhi64_c2_load(i64* %w, i64
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_c2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_c2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, (%rdi), %rax
@@ -1426,6 +2092,27 @@ define i64 @bzhi64_c3_load_indexzext(i64
 ; X86-NOBMI-NEXT:    popl %esi
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_c3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB28_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB28_2:
+; X86-BMI1NOTBM-NEXT:    andl (%esi), %eax
+; X86-BMI1NOTBM-NEXT:    andl 4(%esi), %edx
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %esi
@@ -1456,6 +2143,16 @@ define i64 @bzhi64_c3_load_indexzext(i64
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_c3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -1488,6 +2185,24 @@ define i64 @bzhi64_c4_commutative(i64 %v
 ; X86-NOBMI-NEXT:    andl {{[0-9]+}}(%esp), %edx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_c4_commutative:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl $-1, %eax
+; X86-BMI1NOTBM-NEXT:    movl $-1, %edx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %eax, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    je .LBB29_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %edx, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %edx, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB29_2:
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    andl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_c4_commutative:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl $64, %ecx
@@ -1515,6 +2230,16 @@ define i64 @bzhi64_c4_commutative(i64 %v
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_c4_commutative:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_c4_commutative:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -1540,6 +2265,16 @@ define i32 @bzhi32_d0(i32 %val, i32 %num
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_d0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_d0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1556,6 +2291,16 @@ define i32 @bzhi32_d0(i32 %val, i32 %num
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_d0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_d0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -1577,6 +2322,16 @@ define i32 @bzhi32_d1_indexzext(i32 %val
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_d1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_d1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movb {{[0-9]+}}(%esp), %al
@@ -1593,6 +2348,16 @@ define i32 @bzhi32_d1_indexzext(i32 %val
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_d1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_d1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, %edi, %eax
@@ -1616,6 +2381,17 @@ define i32 @bzhi32_d2_load(i32* %w, i32
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_d2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_d2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1633,6 +2409,16 @@ define i32 @bzhi32_d2_load(i32* %w, i32
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_d2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_d2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -1656,6 +2442,17 @@ define i32 @bzhi32_d3_load_indexzext(i32
 ; X86-NOBMI-NEXT:    shrl %cl, %eax
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -1673,6 +2470,16 @@ define i32 @bzhi32_d3_load_indexzext(i32
 ; X64-NOBMI-NEXT:    shrl %cl, %eax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhil %esi, (%rdi), %eax
@@ -1727,6 +2534,45 @@ define i64 @bzhi64_d0(i64 %val, i64 %num
 ; X86-NOBMI-NEXT:    popl %ebx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_d0:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    jne .LBB34_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB34_2:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB34_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB34_4:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB34_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB34_6:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_d0:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %edi
@@ -1769,6 +2615,16 @@ define i64 @bzhi64_d0(i64 %val, i64 %num
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_d0:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_d0:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, %rdi, %rax
@@ -1819,6 +2675,45 @@ define i64 @bzhi64_d1_indexzext(i64 %val
 ; X86-NOBMI-NEXT:    popl %ebx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_d1_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    jne .LBB35_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB35_2:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB35_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB35_4:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB35_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB35_6:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_d1_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %edi
@@ -1861,6 +2756,16 @@ define i64 @bzhi64_d1_indexzext(i64 %val
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_d1_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_d1_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -1914,6 +2819,46 @@ define i64 @bzhi64_d2_load(i64* %w, i64
 ; X86-NOBMI-NEXT:    popl %ebx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_d2_load:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %edx
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl $64, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    jne .LBB36_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB36_2:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB36_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB36_4:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB36_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB36_6:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_d2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %edi
@@ -1957,6 +2902,16 @@ define i64 @bzhi64_d2_load(i64* %w, i64
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_d2_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_d2_load:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    bzhiq %rsi, (%rdi), %rax
@@ -2009,6 +2964,46 @@ define i64 @bzhi64_d3_load_indexzext(i64
 ; X86-NOBMI-NEXT:    popl %ebx
 ; X86-NOBMI-NEXT:    retl
 ;
+; X86-BMI1NOTBM-LABEL: bzhi64_d3_load_indexzext:
+; X86-BMI1NOTBM:       # %bb.0:
+; X86-BMI1NOTBM-NEXT:    pushl %ebx
+; X86-BMI1NOTBM-NEXT:    pushl %edi
+; X86-BMI1NOTBM-NEXT:    pushl %esi
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %edx
+; X86-BMI1NOTBM-NEXT:    movl 4(%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movb $64, %cl
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    movl %edx, %esi
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    shldl %cl, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl %esi, %edi
+; X86-BMI1NOTBM-NEXT:    jne .LBB37_2
+; X86-BMI1NOTBM-NEXT:  # %bb.1:
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edi
+; X86-BMI1NOTBM-NEXT:  .LBB37_2:
+; X86-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ebx, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    movl $0, %edx
+; X86-BMI1NOTBM-NEXT:    jne .LBB37_4
+; X86-BMI1NOTBM-NEXT:  # %bb.3:
+; X86-BMI1NOTBM-NEXT:    movl %esi, %ebx
+; X86-BMI1NOTBM-NEXT:    movl %eax, %edx
+; X86-BMI1NOTBM-NEXT:  .LBB37_4:
+; X86-BMI1NOTBM-NEXT:    shrdl %cl, %edi, %ebx
+; X86-BMI1NOTBM-NEXT:    testb $32, %cl
+; X86-BMI1NOTBM-NEXT:    jne .LBB37_6
+; X86-BMI1NOTBM-NEXT:  # %bb.5:
+; X86-BMI1NOTBM-NEXT:    movl %ebx, %eax
+; X86-BMI1NOTBM-NEXT:  .LBB37_6:
+; X86-BMI1NOTBM-NEXT:    popl %esi
+; X86-BMI1NOTBM-NEXT:    popl %edi
+; X86-BMI1NOTBM-NEXT:    popl %ebx
+; X86-BMI1NOTBM-NEXT:    retl
+;
 ; X86-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    pushl %edi
@@ -2052,6 +3047,16 @@ define i64 @bzhi64_d3_load_indexzext(i64
 ; X64-NOBMI-NEXT:    shrq %cl, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_d3_load_indexzext:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext:
 ; X64-BMI1BMI2:       # %bb.0:
 ; X64-BMI1BMI2-NEXT:    # kill: def $esi killed $esi def $rsi
@@ -2189,6 +3194,12 @@ define i64 @bzhi64_constant_mask64(i64 %
 ; X64-NOBMI-NEXT:    andq %rdi, %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_constant_mask64:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl $15872, %eax # imm = 0x3E00
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1TBM-LABEL: bzhi64_constant_mask64:
 ; X64-BMI1TBM:       # %bb.0:
 ; X64-BMI1TBM-NEXT:    bextrq $15872, %rdi, %rax # imm = 0x3E00
@@ -2218,6 +3229,12 @@ define i64 @bzhi64_constant_mask64_load(
 ; X64-NOBMI-NEXT:    andq (%rdi), %rax
 ; X64-NOBMI-NEXT:    retq
 ;
+; X64-BMI1NOTBM-LABEL: bzhi64_constant_mask64_load:
+; X64-BMI1NOTBM:       # %bb.0:
+; X64-BMI1NOTBM-NEXT:    movl $15872, %eax # imm = 0x3E00
+; X64-BMI1NOTBM-NEXT:    bextrq %rax, (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    retq
+;
 ; X64-BMI1TBM-LABEL: bzhi64_constant_mask64_load:
 ; X64-BMI1TBM:       # %bb.0:
 ; X64-BMI1TBM-NEXT:    bextrq $15872, (%rdi), %rax # imm = 0x3E00




More information about the llvm-commits mailing list