test/Transforms/PhaseOrdering/X86/sad_variations.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt < %s -passes="default<O2>" -S -mattr=+sse2 | FileCheck %s
; RUN: opt < %s -passes="default<O2>" -S -mattr=+avx  | FileCheck %s
; RUN: opt < %s -passes="default<O2>" -S -mattr=+avx2 | FileCheck %s
; RUN: opt < %s -passes="default<O2>" -S -mattr=+avx512f  | FileCheck %s
; RUN: opt < %s -passes="default<O2>" -S -mattr=+avx512bw | FileCheck %s

define i32 @sad8_32bit_icmp_sge(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i32 %stride) local_unnamed_addr #0 {
; CHECK-LABEL: define i32 @sad8_32bit_icmp_sge(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i32 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = tail call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[I7]])
; CHECK-NEXT:    ret i32 [[I8]]
;
entry:
  %idx.ext = zext i32 %stride to i64
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp sgt <8 x i32> %i4, splat (i32 -1)
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i4, <8 x i32> %i6
  %rdx.shuf = shufflevector <8 x i32> %i7, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i32> %i7, %rdx.shuf
  %rdx.shuf229 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx230 = add <8 x i32> %bin.rdx, %rdx.shuf229
  %rdx.shuf231 = shufflevector <8 x i32> %bin.rdx230, <8 x i32> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx232 = add <8 x i32> %bin.rdx230, %rdx.shuf231
  %i8 = extractelement <8 x i32> %bin.rdx232, i32 0
  ret i32 %i8
}

define i32 @sad8_32bit_icmp_sgt(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i32 %stride) local_unnamed_addr #1 {
; CHECK-LABEL: define i32 @sad8_32bit_icmp_sgt(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i32 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = tail call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[I7]])
; CHECK-NEXT:    ret i32 [[I8]]
;
entry:
  %idx.ext = zext i32 %stride to i64
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp sgt <8 x i32> %i4, zeroinitializer
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i4, <8 x i32> %i6
  %rdx.shuf = shufflevector <8 x i32> %i7, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i32> %i7, %rdx.shuf
  %rdx.shuf229 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx230 = add <8 x i32> %bin.rdx, %rdx.shuf229
  %rdx.shuf231 = shufflevector <8 x i32> %bin.rdx230, <8 x i32> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx232 = add <8 x i32> %bin.rdx230, %rdx.shuf231
  %i8 = extractelement <8 x i32> %bin.rdx232, i32 0
  ret i32 %i8
}

define i32 @sad8_32bit_icmp_sle(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i32 %stride) local_unnamed_addr #2 {
; CHECK-LABEL: define i32 @sad8_32bit_icmp_sle(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i32 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = tail call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[I7]])
; CHECK-NEXT:    ret i32 [[I8]]
;
entry:
  %idx.ext = zext i32 %stride to i64
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp slt <8 x i32> %i4, splat (i32 1)
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i6, <8 x i32> %i4
  %rdx.shuf = shufflevector <8 x i32> %i7, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i32> %i7, %rdx.shuf
  %rdx.shuf229 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx230 = add <8 x i32> %bin.rdx, %rdx.shuf229
  %rdx.shuf231 = shufflevector <8 x i32> %bin.rdx230, <8 x i32> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx232 = add <8 x i32> %bin.rdx230, %rdx.shuf231
  %i8 = extractelement <8 x i32> %bin.rdx232, i32 0
  ret i32 %i8
}

define i32 @sad8_32bit_icmp_slt(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i32 %stride) local_unnamed_addr #3 {
; CHECK-LABEL: define i32 @sad8_32bit_icmp_slt(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i32 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = tail call i32 @llvm.vector.reduce.add.v8i32(<8 x i32> [[I7]])
; CHECK-NEXT:    ret i32 [[I8]]
;
entry:
  %idx.ext = zext i32 %stride to i64
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp slt <8 x i32> %i4, zeroinitializer
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i6, <8 x i32> %i4
  %rdx.shuf = shufflevector <8 x i32> %i7, <8 x i32> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i32> %i7, %rdx.shuf
  %rdx.shuf229 = shufflevector <8 x i32> %bin.rdx, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx230 = add <8 x i32> %bin.rdx, %rdx.shuf229
  %rdx.shuf231 = shufflevector <8 x i32> %bin.rdx230, <8 x i32> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx232 = add <8 x i32> %bin.rdx230, %rdx.shuf231
  %i8 = extractelement <8 x i32> %bin.rdx232, i32 0
  ret i32 %i8
}

define i64 @sad8_64bit_icmp_sext_slt(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i64 %stride) local_unnamed_addr #4 {
; CHECK-LABEL: define i64 @sad8_64bit_icmp_sext_slt(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i64 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = zext nneg <8 x i32> [[I7]] to <8 x i64>
; CHECK-NEXT:    [[I9:%.*]] = tail call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[I8]])
; CHECK-NEXT:    ret i64 [[I9]]
;
entry:
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp slt <8 x i32> %i4, zeroinitializer
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i6, <8 x i32> %i4
  %i8 = sext <8 x i32> %i7 to <8 x i64>
  %rdx.shuf = shufflevector <8 x i64> %i8, <8 x i64> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i64> %rdx.shuf, %i8
  %rdx.shuf236 = shufflevector <8 x i64> %bin.rdx, <8 x i64> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx237 = add <8 x i64> %bin.rdx, %rdx.shuf236
  %rdx.shuf238 = shufflevector <8 x i64> %bin.rdx237, <8 x i64> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx239 = add <8 x i64> %bin.rdx237, %rdx.shuf238
  %i9 = extractelement <8 x i64> %bin.rdx239, i32 0
  ret i64 %i9
}

define i64 @sad8_64bit_icmp_zext_slt(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i64 %stride) local_unnamed_addr #4 {
; CHECK-LABEL: define i64 @sad8_64bit_icmp_zext_slt(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i64 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i32>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i32>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i32> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i32> @llvm.abs.v8i32(<8 x i32> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = zext nneg <8 x i32> [[I7]] to <8 x i64>
; CHECK-NEXT:    [[I9:%.*]] = tail call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[I8]])
; CHECK-NEXT:    ret i64 [[I9]]
;
entry:
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i32>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i32>
  %i4 = sub nsw <8 x i32> %i1, %i3
  %i5 = icmp slt <8 x i32> %i4, zeroinitializer
  %i6 = sub nsw <8 x i32> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i32> %i6, <8 x i32> %i4
  %i8 = zext <8 x i32> %i7 to <8 x i64>
  %rdx.shuf = shufflevector <8 x i64> %i8, <8 x i64> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i64> %rdx.shuf, %i8
  %rdx.shuf236 = shufflevector <8 x i64> %bin.rdx, <8 x i64> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx237 = add <8 x i64> %bin.rdx, %rdx.shuf236
  %rdx.shuf238 = shufflevector <8 x i64> %bin.rdx237, <8 x i64> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx239 = add <8 x i64> %bin.rdx237, %rdx.shuf238
  %i9 = extractelement <8 x i64> %bin.rdx239, i32 0
  ret i64 %i9
}

define i64 @sad8_early_64bit_icmp_zext_slt(ptr nocapture readonly %cur, ptr nocapture readonly %ref, i64 %stride) local_unnamed_addr #4 {
; CHECK-LABEL: define i64 @sad8_early_64bit_icmp_zext_slt(
; CHECK-SAME: ptr readonly captures(none) [[CUR:%.*]], ptr readonly captures(none) [[REF:%.*]], i64 [[STRIDE:%.*]]) local_unnamed_addr #[[ATTR0]] {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[I:%.*]] = load <8 x i8>, ptr [[CUR]], align 1
; CHECK-NEXT:    [[I1:%.*]] = zext <8 x i8> [[I]] to <8 x i64>
; CHECK-NEXT:    [[I2:%.*]] = load <8 x i8>, ptr [[REF]], align 1
; CHECK-NEXT:    [[I3:%.*]] = zext <8 x i8> [[I2]] to <8 x i64>
; CHECK-NEXT:    [[I4:%.*]] = sub nsw <8 x i64> [[I1]], [[I3]]
; CHECK-NEXT:    [[I7:%.*]] = tail call <8 x i64> @llvm.abs.v8i64(<8 x i64> [[I4]], i1 true)
; CHECK-NEXT:    [[I8:%.*]] = tail call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[I7]])
; CHECK-NEXT:    ret i64 [[I8]]
;
entry:
  br label %for.body

for.body:                                         ; preds = %entry
  %i = load <8 x i8>, ptr %cur, align 1
  %i1 = zext <8 x i8> %i to <8 x i64>
  %i2 = load <8 x i8>, ptr %ref, align 1
  %i3 = zext <8 x i8> %i2 to <8 x i64>
  %i4 = sub nsw <8 x i64> %i1, %i3
  %i5 = icmp slt <8 x i64> %i4, zeroinitializer
  %i6 = sub nsw <8 x i64> zeroinitializer, %i4
  %i7 = select <8 x i1> %i5, <8 x i64> %i6, <8 x i64> %i4
  %rdx.shuf = shufflevector <8 x i64> %i7, <8 x i64> undef, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx = add <8 x i64> %rdx.shuf, %i7
  %rdx.shuf236 = shufflevector <8 x i64> %bin.rdx, <8 x i64> undef, <8 x i32> <i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx237 = add <8 x i64> %bin.rdx, %rdx.shuf236
  %rdx.shuf238 = shufflevector <8 x i64> %bin.rdx237, <8 x i64> undef, <8 x i32> <i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
  %bin.rdx239 = add <8 x i64> %bin.rdx237, %rdx.shuf238
  %i8 = extractelement <8 x i64> %bin.rdx239, i32 0
  ret i64 %i8
}