[InstCombine][NFC] Tests for 'signed truncation check' optimization

[spatel]

Tests LGTM. It might help me or other reviewers see how these patterns escape the current set of transforms in foldAndOfICmps(), so please commit.

[lebedev.ri]

s/%tmp/%t/

test/Transforms/InstCombine/signed-truncation-check.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+; General pattern:
+;   X & Y
+;
+; Where Y is checking that all the high bits (covered by a mask 4294967168)
+; are uniform, i.e.  %arg & 4294967168  can be either  4294967168  or  0
+; Pattern can be one of:
+;   %t = add        i32 %arg,    128
+;   %r = icmp   ult i32 %t,      256
+; Or
+;   %t0 = shl       i32 %arg,    24
+;   %t1 = ashr      i32 %t0,     24
+;   %r  = icmp  eq  i32 %t1,     %arg
+; Or
+;   %t0 = trunc     i32 %arg  to i8
+;   %t1 = sext      i8  %t0   to i32
+;   %r  = icmp  eq  i32 %t1,     %arg
+; This pattern is a signed truncation check.
+;
+; And X is checking that some bit in that same mask is zero.
+; I.e. can be one of:
+;   %r = icmp sgt i32   %arg,    -1
+; Or
+;   %t = and      i32   %arg,    2147483648
+;   %r = icmp eq  i32   %t,      0
+;
+; Since we are checking that all the bits in that mask are the same,
+; and a particular bit is zero, what we are really checking is that all the
+; masked bits are zero.
+; So this should be transformed to:
+;   %r = icmp ult i32 %arg, 128
+
+; ============================================================================ ;
+; Basic positive test
+; ============================================================================ ;
+
+define i1 @positive_with_signbit(i32 %arg) {
+; CHECK-LABEL: @positive_with_signbit(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg, 128
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+define i1 @positive_with_mask(i32 %arg) {
+; CHECK-LABEL: @positive_with_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], 1107296256
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 1107296256
+  %tmp2 = icmp eq i32 %tmp1, 0
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @positive_with_icmp(i32 %arg) {
+; CHECK-LABEL: @positive_with_icmp(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult i32 [[ARG:%.*]], 512
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp ult i32 %arg, 512
+  %tmp2 = add i32 %arg, 128
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+; Still the same
+define i1 @positive_with_aggressive_icmp(i32 %arg) {
+; CHECK-LABEL: @positive_with_aggressive_icmp(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult i32 [[ARG:%.*]], 128
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 256
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 512
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp ult i32 %arg, 128
+  %tmp2 = add i32 %arg, 256
+  %tmp3 = icmp ult i32 %tmp2, 512
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+; I'm sure there is a bunch more patterns possible :/
+
+; ============================================================================ ;
+; Vector tests
+; ============================================================================ ;
+
+define <2 x i1> @positive_vec_splat(<2 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_splat(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <2 x i32> [[ARG:%.*]], <i32 -1, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[ARG]], <i32 128, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <2 x i32> [[TMP2]], <i32 256, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <2 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <2 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <2 x i32> %arg, <i32 -1, i32 -1>
+  %tmp2 = add <2 x i32> %arg, <i32 128, i32 128>
+  %tmp3 = icmp ult <2 x i32> %tmp2, <i32 256, i32 256>
+  %tmp4 = and <2 x i1> %tmp1, %tmp3
+  ret <2 x i1> %tmp4
+}
+
+define <2 x i1> @positive_vec_nonsplat(<2 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_nonsplat(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <2 x i32> [[ARG:%.*]], <i32 -1, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <2 x i32> [[ARG]], <i32 128, i32 256>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <2 x i32> [[TMP2]], <i32 256, i32 512>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <2 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <2 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <2 x i32> %arg, <i32 -1, i32 -1>
+  %tmp2 = add <2 x i32> %arg, <i32 128, i32 256>
+  %tmp3 = icmp ult <2 x i32> %tmp2, <i32 256, i32 512>
+  %tmp4 = and <2 x i1> %tmp1, %tmp3
+  ret <2 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef0(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef0(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 undef, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 128, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 256, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 undef, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 128, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 256, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef1(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef1(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 -1, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 undef, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 256, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 -1, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 undef, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 256, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef2(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef2(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 -1, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 128, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 undef, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 -1, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 128, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 undef, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef3(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef3(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 undef, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 undef, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 256, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 undef, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 undef, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 256, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef4(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef4(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 undef, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 128, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 undef, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 undef, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 128, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 undef, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef5(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef5(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 -1, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 undef, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 undef, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 -1, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 undef, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 undef, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+define <3 x i1> @positive_vec_undef6(<3 x i32> %arg) {
+; CHECK-LABEL: @positive_vec_undef6(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i32> [[ARG:%.*]], <i32 -1, i32 undef, i32 -1>
+; CHECK-NEXT:    [[TMP2:%.*]] = add <3 x i32> [[ARG]], <i32 128, i32 undef, i32 128>
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult <3 x i32> [[TMP2]], <i32 256, i32 undef, i32 256>
+; CHECK-NEXT:    [[TMP4:%.*]] = and <3 x i1> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret <3 x i1> [[TMP4]]
+;
+  %tmp1 = icmp sgt <3 x i32> %arg, <i32 -1, i32 undef, i32 -1>
+  %tmp2 = add <3 x i32> %arg, <i32 128, i32 undef, i32 128>
+  %tmp3 = icmp ult <3 x i32> %tmp2, <i32 256, i32 undef, i32 256>
+  %tmp4 = and <3 x i1> %tmp1, %tmp3
+  ret <3 x i1> %tmp4
+}
+
+; ============================================================================ ;
+; Commutativity tests.
+; ============================================================================ ;
+
+declare i32 @gen32()
+
+define i1 @commutative() {
+; CHECK-LABEL: @commutative(
+; CHECK-NEXT:    [[ARG:%.*]] = call i32 @gen32()
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP3]], [[TMP1]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %arg = call i32 @gen32()
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg, 128
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp3, %tmp1 ; swapped order
+  ret i1 %tmp4
+}
+
+define i1 @commutative_with_icmp() {
+; CHECK-LABEL: @commutative_with_icmp(
+; CHECK-NEXT:    [[ARG:%.*]] = call i32 @gen32()
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp ult i32 [[ARG]], 512
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP3]], [[TMP1]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %arg = call i32 @gen32()
+  %tmp1 = icmp ult i32 %arg, 512
+  %tmp2 = add i32 %arg, 128
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp3, %tmp1 ; swapped order
+  ret i1 %tmp4
+}
+
+; ============================================================================ ;
+; Truncations.
+; ============================================================================ ;
+
+define i1 @positive_trunc_signbit(i32 %arg) {
+; CHECK-LABEL: @positive_trunc_signbit(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i8
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i8 [[TMP1]], -1
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = trunc i32 %arg to i8
+  %tmp2 = icmp sgt i8 %tmp1, -1
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @positive_trunc_base(i32 %arg) {
+; CHECK-LABEL: @positive_trunc_base(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i16
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i16 [[TMP1]], -1
+; CHECK-NEXT:    [[TMP3:%.*]] = add i16 [[TMP1]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i16 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = trunc i32 %arg to i16
+  %tmp2 = icmp sgt i16 %tmp1, -1
+  %tmp3 = add i16 %tmp1, 128
+  %tmp4 = icmp ult i16 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @positive_different_trunc_both(i32 %arg) {
+; CHECK-LABEL: @positive_different_trunc_both(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i15
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i15 [[TMP1]], -1
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 [[ARG]] to i16
+; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[TMP3]], 128
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i16 [[TMP4]], 256
+; CHECK-NEXT:    [[TMP6:%.*]] = and i1 [[TMP2]], [[TMP5]]
+; CHECK-NEXT:    ret i1 [[TMP6]]
+;
+  %tmp1 = trunc i32 %arg to i15
+  %tmp2 = icmp sgt i15 %tmp1, -1
+  %tmp3 = trunc i32 %arg to i16
+  %tmp4 = add i16 %tmp3, 128
+  %tmp5 = icmp ult i16 %tmp4, 256
+  %tmp6 = and i1 %tmp2, %tmp5
+  ret i1 %tmp6
+}
+
+; ============================================================================ ;
+; One-use tests.
+;
+; We will only produce one instruction, so we do not care about one-use.
+; But, we *could* handle more patterns that we weren't able to canonicalize
+; because of extra-uses.
+; ============================================================================ ;
+
+declare void @use32(i32)
+declare void @use8(i8)
+declare void @use1(i1)
+
+define i1 @oneuse_with_signbit(i32 %arg) {
+; CHECK-LABEL: @oneuse_with_signbit(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    call void @use1(i1 [[TMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    call void @use32(i32 [[TMP2]])
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    call void @use1(i1 [[TMP3]])
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  call void @use1(i1 %tmp1)
+  %tmp2 = add i32 %arg, 128
+  call void @use32(i32 %tmp2)
+  %tmp3 = icmp ult i32 %tmp2, 256
+  call void @use1(i1 %tmp3)
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+define i1 @oneuse_with_mask(i32 %arg) {
+; CHECK-LABEL: @oneuse_with_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], 603979776
+; CHECK-NEXT:    call void @use32(i32 [[TMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    call void @use1(i1 [[TMP2]])
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    call void @use32(i32 [[TMP3]])
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    call void @use1(i1 [[TMP4]])
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 603979776 ; some bit within the target 4294967168 mask.
+  call void @use32(i32 %tmp1)
+  %tmp2 = icmp eq i32 %tmp1, 0
+  call void @use1(i1 %tmp2)
+  %tmp3 = add i32 %arg, 128
+  call void @use32(i32 %tmp3)
+  %tmp4 = icmp ult i32 %tmp3, 256
+  call void @use1(i1 %tmp4)
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @oneuse_shl_ashr(i32 %arg) {
+; CHECK-LABEL: @oneuse_shl_ashr(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i8
+; CHECK-NEXT:    call void @use8(i8 [[TMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i8 [[TMP1]], -1
+; CHECK-NEXT:    call void @use1(i1 [[TMP2]])
+; CHECK-NEXT:    [[TMP3:%.*]] = shl i32 [[ARG]], 24
+; CHECK-NEXT:    call void @use32(i32 [[TMP3]])
+; CHECK-NEXT:    [[TMP4:%.*]] = ashr exact i32 [[TMP3]], 24
+; CHECK-NEXT:    call void @use32(i32 [[TMP4]])
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[TMP4]], [[ARG]]
+; CHECK-NEXT:    call void @use1(i1 [[TMP5]])
+; CHECK-NEXT:    [[TMP6:%.*]] = and i1 [[TMP2]], [[TMP5]]
+; CHECK-NEXT:    ret i1 [[TMP6]]
+;
+  %tmp1 = trunc i32 %arg to i8
+  call void @use8(i8 %tmp1)
+  %tmp2 = icmp sgt i8 %tmp1, -1
+  call void @use1(i1 %tmp2)
+  %tmp3 = shl i32 %arg, 24
+  call void @use32(i32 %tmp3)
+  %tmp4 = ashr i32 %tmp3, 24
+  call void @use32(i32 %tmp4)
+  %tmp5 = icmp eq i32 %tmp4, %arg
+  call void @use1(i1 %tmp5)
+  %tmp6 = and i1 %tmp2, %tmp5
+  ret i1 %tmp6
+}
+
+define zeroext i1 @oneuse_trunc_sext(i32 %arg) {
+; CHECK-LABEL: @oneuse_trunc_sext(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i8
+; CHECK-NEXT:    call void @use8(i8 [[TMP1]])
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i8 [[TMP1]], -1
+; CHECK-NEXT:    call void @use1(i1 [[TMP2]])
+; CHECK-NEXT:    [[TMP3:%.*]] = trunc i32 [[ARG]] to i8
+; CHECK-NEXT:    call void @use8(i8 [[TMP3]])
+; CHECK-NEXT:    [[TMP4:%.*]] = sext i8 [[TMP3]] to i32
+; CHECK-NEXT:    call void @use32(i32 [[TMP4]])
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[TMP4]], [[ARG]]
+; CHECK-NEXT:    call void @use1(i1 [[TMP5]])
+; CHECK-NEXT:    [[TMP6:%.*]] = and i1 [[TMP2]], [[TMP5]]
+; CHECK-NEXT:    ret i1 [[TMP6]]
+;
+  %tmp1 = trunc i32 %arg to i8
+  call void @use8(i8 %tmp1)
+  %tmp2 = icmp sgt i8 %tmp1, -1
+  call void @use1(i1 %tmp2)
+  %tmp3 = trunc i32 %arg to i8
+  call void @use8(i8 %tmp3)
+  %tmp4 = sext i8 %tmp3 to i32
+  call void @use32(i32 %tmp4)
+  %tmp5 = icmp eq i32 %tmp4, %arg
+  call void @use1(i1 %tmp5)
+  %tmp6 = and i1 %tmp2, %tmp5
+  ret i1 %tmp6
+}
+
+; ============================================================================ ;
+; Negative tests
+; ============================================================================ ;
+
+define i1 @negative_not_arg(i32 %arg, i32 %arg2) {
+; CHECK-LABEL: @negative_not_arg(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG2:%.*]], 128
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg2, 128 ; not %arg
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+define i1 @negative_trunc_not_arg(i32 %arg, i32 %arg2) {
+; CHECK-LABEL: @negative_trunc_not_arg(
+; CHECK-NEXT:    [[TMP1:%.*]] = trunc i32 [[ARG:%.*]] to i8
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp sgt i8 [[TMP1]], -1
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG2:%.*]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = trunc i32 %arg to i8
+  %tmp2 = icmp sgt i8 %tmp1, -1
+  %tmp3 = add i32 %arg2, 128 ; not %arg
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @positive_with_mask_not_arg(i32 %arg, i32 %arg2) {
+; CHECK-LABEL: @positive_with_mask_not_arg(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], 1140850688
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG2:%.*]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 1140850688
+  %tmp2 = icmp eq i32 %tmp1, 0
+  %tmp3 = add i32 %arg2, 128 ; not %arg
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @negative_with_nonuniform_bad_mask(i32 %arg) {
+; CHECK-LABEL: @negative_with_nonuniform_bad_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], 1711276033
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 1711276033 ; lowest bit is set
+  %tmp2 = icmp eq i32 %tmp1, 0
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @negative_with_uniform_bad_mask(i32 %arg) {
+; CHECK-LABEL: @negative_with_uniform_bad_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], -16777152
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 4278190144 ; 7'th bit is set
+  %tmp2 = icmp eq i32 %tmp1, 0
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @negative_with_wrong_mask(i32 %arg) {
+; CHECK-LABEL: @negative_with_wrong_mask(
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[ARG:%.*]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp eq i32 [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = and i32 %arg, 1 ; not even checking the right mask
+  %tmp2 = icmp eq i32 %tmp1, 0
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @negative_not_less_than(i32 %arg) {
+; CHECK-LABEL: @negative_not_less_than(
+; CHECK-NEXT:    ret i1 false
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg, 256 ; should be less than 256
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+define i1 @negative_not_power_of_two(i32 %arg) {
+; CHECK-LABEL: @negative_not_power_of_two(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 255
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg, 255 ; should be power of two
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+define i1 @negative_not_next_power_of_two(i32 %arg) {
+; CHECK-LABEL: @negative_not_next_power_of_two(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[ARG]], 64
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ult i32 [[TMP2]], 256
+; CHECK-NEXT:    [[TMP4:%.*]] = and i1 [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    ret i1 [[TMP4]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1
+  %tmp2 = add i32 %arg, 64 ; should be 256 >> 1
+  %tmp3 = icmp ult i32 %tmp2, 256
+  %tmp4 = and i1 %tmp1, %tmp3
+  ret i1 %tmp4
+}
+
+; I don't think this can be folded, at least not into single instruction.
+define i1 @two_signed_truncation_checks(i32 %arg) {
+; CHECK-LABEL: @two_signed_truncation_checks(
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[ARG:%.*]], 512
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult i32 [[TMP1]], 1024
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[ARG]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP2]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = add i32 %arg, 512
+  %tmp2 = icmp ult i32 %tmp1, 1024
+  %tmp3 = add i32 %arg, 128
+  %tmp4 = icmp ult i32 %tmp3, 256
+  %tmp5 = and i1 %tmp2, %tmp4
+  ret i1 %tmp5
+}
+
+define i1 @bad_trunc_stc(i32 %arg) {
+; CHECK-LABEL: @bad_trunc_stc(
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt i32 [[ARG:%.*]], -1
+; CHECK-NEXT:    [[TMP2:%.*]] = trunc i32 [[ARG]] to i16
+; CHECK-NEXT:    [[TMP3:%.*]] = add i16 [[TMP2]], 128
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i16 [[TMP3]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = and i1 [[TMP1]], [[TMP4]]
+; CHECK-NEXT:    ret i1 [[TMP5]]
+;
+  %tmp1 = icmp sgt i32 %arg, -1 ; checks a bit outside of the i16
+  %tmp2 = trunc i32 %arg to i16
+  %tmp3 = add i16 %tmp2, 128
+  %tmp4 = icmp ult i16 %tmp3, 256
+  %tmp5 = and i1 %tmp1, %tmp4
+  ret i1 %tmp5
+}