[llvm] [PowerPC] Custom Lower ABDU using branchless carry (PR #182704)
Lei Huang via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 2 08:21:23 PDT 2026
================
@@ -0,0 +1,1020 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=powerpc-unknown-linux-gnu %s -o - | FileCheck %s --check-prefixes=CHECK,PPC32
+; RUN: llc -mtriple=powerpc64-unknown-linux-gnu %s -o - | FileCheck %s --check-prefixes=CHECK,PPC64
+
+;
+; trunc(abs(sub(zext(a),zext(b)))) -> abdu(a,b)
+;
+
+define i8 @abd_ext_i8(i8 %a, i8 %b) nounwind {
+; PPC32-LABEL: abd_ext_i8:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 24
+; PPC32-NEXT: clrlwi 4, 4, 24
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i8:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 56
+; PPC64-NEXT: clrldi 4, 4, 56
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i8 %a to i64
+ %bext = zext i8 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i8
+ ret i8 %trunc
+}
+
+define i8 @abd_ext_i8_undef(i8 %a, i8 %b) nounwind {
+; PPC32-LABEL: abd_ext_i8_undef:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 24
+; PPC32-NEXT: clrlwi 4, 4, 24
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i8_undef:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 56
+; PPC64-NEXT: clrldi 4, 4, 56
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i8 %a to i64
+ %bext = zext i8 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 true)
+ %trunc = trunc i64 %abs to i8
+ ret i8 %trunc
+}
+
+define i16 @abd_ext_i16(i16 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_ext_i16:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i16:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i16 %a to i64
+ %bext = zext i16 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i16
+ ret i16 %trunc
+}
+
+define i16 @abd_ext_i16_i32(i16 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_ext_i16_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i16_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i16 %a to i64
+ %bext = zext i32 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i16
+ ret i16 %trunc
+}
+
+define i16 @abd_ext_i16_undef(i16 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_ext_i16_undef:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i16_undef:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i16 %a to i64
+ %bext = zext i16 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 true)
+ %trunc = trunc i64 %abs to i16
+ ret i16 %trunc
+}
+
+define i32 @abd_ext_i32(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_ext_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i32 %a to i64
+ %bext = zext i32 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i32
+ ret i32 %trunc
+}
+
+define i32 @abd_ext_i32_i16(i32 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_ext_i32_i16:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i32_i16:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i32 %a to i64
+ %bext = zext i16 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i32
+ ret i32 %trunc
+}
+
+define i32 @abd_ext_i32_undef(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_ext_i32_undef:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i32_undef:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %aext = zext i32 %a to i64
+ %bext = zext i32 %b to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 true)
+ %trunc = trunc i64 %abs to i32
+ ret i32 %trunc
+}
+
+define i64 @abd_ext_i64(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_ext_i64:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i64:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %aext = zext i64 %a to i128
+ %bext = zext i64 %b to i128
+ %sub = sub i128 %aext, %bext
+ %abs = call i128 @llvm.abs.i128(i128 %sub, i1 false)
+ %trunc = trunc i128 %abs to i64
+ ret i64 %trunc
+}
+
+define i64 @abd_ext_i64_undef(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_ext_i64_undef:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i64_undef:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %aext = zext i64 %a to i128
+ %bext = zext i64 %b to i128
+ %sub = sub i128 %aext, %bext
+ %abs = call i128 @llvm.abs.i128(i128 %sub, i1 true)
+ %trunc = trunc i128 %abs to i64
+ ret i64 %trunc
+}
+
+define i128 @abd_ext_i128(i128 %a, i128 %b) nounwind {
+; PPC32-LABEL: abd_ext_i128:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 6, 6, 10
+; PPC32-NEXT: subfe 5, 9, 5
+; PPC32-NEXT: subfe 4, 8, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: mfxer 7
+; PPC32-NEXT: li 11, 0
+; PPC32-NEXT: addze 8, 11
+; PPC32-NEXT: addic 8, 8, -1
+; PPC32-NEXT: mtxer 7
+; PPC32-NEXT: subfe 7, 8, 8
+; PPC32-NEXT: xor 6, 6, 7
+; PPC32-NEXT: xor 5, 5, 7
+; PPC32-NEXT: subc 6, 6, 7
+; PPC32-NEXT: xor 4, 4, 7
+; PPC32-NEXT: subfe 5, 7, 5
+; PPC32-NEXT: xor 3, 3, 7
+; PPC32-NEXT: subfe 4, 7, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i128:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 4, 4, 6
+; PPC64-NEXT: li 7, 0
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: mfxer 5
+; PPC64-NEXT: addze 6, 7
+; PPC64-NEXT: addic 6, 6, -1
+; PPC64-NEXT: mtxer 5
+; PPC64-NEXT: subfe 5, 6, 6
+; PPC64-NEXT: xor 4, 4, 5
+; PPC64-NEXT: xor 3, 3, 5
+; PPC64-NEXT: subc 4, 4, 5
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: blr
+ %aext = zext i128 %a to i256
+ %bext = zext i128 %b to i256
+ %sub = sub i256 %aext, %bext
+ %abs = call i256 @llvm.abs.i256(i256 %sub, i1 false)
+ %trunc = trunc i256 %abs to i128
+ ret i128 %trunc
+}
+
+define i128 @abd_ext_i128_undef(i128 %a, i128 %b) nounwind {
+; PPC32-LABEL: abd_ext_i128_undef:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 6, 6, 10
+; PPC32-NEXT: subfe 5, 9, 5
+; PPC32-NEXT: subfe 4, 8, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: mfxer 7
+; PPC32-NEXT: li 11, 0
+; PPC32-NEXT: addze 8, 11
+; PPC32-NEXT: addic 8, 8, -1
+; PPC32-NEXT: mtxer 7
+; PPC32-NEXT: subfe 7, 8, 8
+; PPC32-NEXT: xor 6, 6, 7
+; PPC32-NEXT: xor 5, 5, 7
+; PPC32-NEXT: subc 6, 6, 7
+; PPC32-NEXT: xor 4, 4, 7
+; PPC32-NEXT: subfe 5, 7, 5
+; PPC32-NEXT: xor 3, 3, 7
+; PPC32-NEXT: subfe 4, 7, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_ext_i128_undef:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 4, 4, 6
+; PPC64-NEXT: li 7, 0
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: mfxer 5
+; PPC64-NEXT: addze 6, 7
+; PPC64-NEXT: addic 6, 6, -1
+; PPC64-NEXT: mtxer 5
+; PPC64-NEXT: subfe 5, 6, 6
+; PPC64-NEXT: xor 4, 4, 5
+; PPC64-NEXT: xor 3, 3, 5
+; PPC64-NEXT: subc 4, 4, 5
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: blr
+ %aext = zext i128 %a to i256
+ %bext = zext i128 %b to i256
+ %sub = sub i256 %aext, %bext
+ %abs = call i256 @llvm.abs.i256(i256 %sub, i1 true)
+ %trunc = trunc i256 %abs to i128
+ ret i128 %trunc
+}
+
+;
+; sub(umax(a,b),umin(a,b)) -> abdu(a,b)
+;
+
+define i8 @abd_minmax_i8(i8 %a, i8 %b) nounwind {
+; PPC32-LABEL: abd_minmax_i8:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 24
+; PPC32-NEXT: clrlwi 4, 4, 24
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_minmax_i8:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 56
+; PPC64-NEXT: clrldi 4, 4, 56
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %min = call i8 @llvm.umin.i8(i8 %a, i8 %b)
+ %max = call i8 @llvm.umax.i8(i8 %a, i8 %b)
+ %sub = sub i8 %max, %min
+ ret i8 %sub
+}
+
+define i16 @abd_minmax_i16(i16 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_minmax_i16:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_minmax_i16:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %min = call i16 @llvm.umin.i16(i16 %a, i16 %b)
+ %max = call i16 @llvm.umax.i16(i16 %a, i16 %b)
+ %sub = sub i16 %max, %min
+ ret i16 %sub
+}
+
+define i32 @abd_minmax_i32(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_minmax_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_minmax_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %min = call i32 @llvm.umin.i32(i32 %a, i32 %b)
+ %max = call i32 @llvm.umax.i32(i32 %a, i32 %b)
+ %sub = sub i32 %max, %min
+ ret i32 %sub
+}
+
+define i64 @abd_minmax_i64(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_minmax_i64:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_minmax_i64:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %min = call i64 @llvm.umin.i64(i64 %a, i64 %b)
+ %max = call i64 @llvm.umax.i64(i64 %a, i64 %b)
+ %sub = sub i64 %max, %min
+ ret i64 %sub
+}
+
+define i128 @abd_minmax_i128(i128 %a, i128 %b) nounwind {
+; PPC32-LABEL: abd_minmax_i128:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 6, 6, 10
+; PPC32-NEXT: subfe 5, 9, 5
+; PPC32-NEXT: subfe 4, 8, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: mfxer 7
+; PPC32-NEXT: li 11, 0
+; PPC32-NEXT: addze 8, 11
+; PPC32-NEXT: addic 8, 8, -1
+; PPC32-NEXT: mtxer 7
+; PPC32-NEXT: subfe 7, 8, 8
+; PPC32-NEXT: xor 6, 6, 7
+; PPC32-NEXT: xor 5, 5, 7
+; PPC32-NEXT: subc 6, 6, 7
+; PPC32-NEXT: xor 4, 4, 7
+; PPC32-NEXT: subfe 5, 7, 5
+; PPC32-NEXT: xor 3, 3, 7
+; PPC32-NEXT: subfe 4, 7, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_minmax_i128:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 4, 4, 6
+; PPC64-NEXT: li 7, 0
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: mfxer 5
+; PPC64-NEXT: addze 6, 7
+; PPC64-NEXT: addic 6, 6, -1
+; PPC64-NEXT: mtxer 5
+; PPC64-NEXT: subfe 5, 6, 6
+; PPC64-NEXT: xor 4, 4, 5
+; PPC64-NEXT: xor 3, 3, 5
+; PPC64-NEXT: subc 4, 4, 5
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: blr
+ %min = call i128 @llvm.umin.i128(i128 %a, i128 %b)
+ %max = call i128 @llvm.umax.i128(i128 %a, i128 %b)
+ %sub = sub i128 %max, %min
+ ret i128 %sub
+}
+
+;
+; select(icmp(a,b),sub(a,b),sub(b,a)) -> abdu(a,b)
+;
+
+define i8 @abd_cmp_i8(i8 %a, i8 %b) nounwind {
+; PPC32-LABEL: abd_cmp_i8:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 24
+; PPC32-NEXT: clrlwi 4, 4, 24
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_cmp_i8:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 56
+; PPC64-NEXT: clrldi 4, 4, 56
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp ugt i8 %a, %b
+ %ab = sub i8 %a, %b
+ %ba = sub i8 %b, %a
+ %sel = select i1 %cmp, i8 %ab, i8 %ba
+ ret i8 %sel
+}
+
+define i16 @abd_cmp_i16(i16 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_cmp_i16:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_cmp_i16:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp uge i16 %a, %b
+ %ab = sub i16 %a, %b
+ %ba = sub i16 %b, %a
+ %sel = select i1 %cmp, i16 %ab, i16 %ba
+ ret i16 %sel
+}
+
+define i32 @abd_cmp_i32(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_cmp_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_cmp_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp ult i32 %a, %b
+ %ab = sub i32 %a, %b
+ %ba = sub i32 %b, %a
+ %sel = select i1 %cmp, i32 %ba, i32 %ab
+ ret i32 %sel
+}
+
+define i64 @abd_cmp_i64(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_cmp_i64:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_cmp_i64:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp uge i64 %a, %b
+ %ab = sub i64 %a, %b
+ %ba = sub i64 %b, %a
+ %sel = select i1 %cmp, i64 %ab, i64 %ba
+ ret i64 %sel
+}
+
+define i128 @abd_cmp_i128(i128 %a, i128 %b) nounwind {
+; PPC32-LABEL: abd_cmp_i128:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 6, 6, 10
+; PPC32-NEXT: subfe 5, 9, 5
+; PPC32-NEXT: subfe 4, 8, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: mfxer 7
+; PPC32-NEXT: li 11, 0
+; PPC32-NEXT: addze 8, 11
+; PPC32-NEXT: addic 8, 8, -1
+; PPC32-NEXT: mtxer 7
+; PPC32-NEXT: subfe 7, 8, 8
+; PPC32-NEXT: xor 6, 6, 7
+; PPC32-NEXT: xor 5, 5, 7
+; PPC32-NEXT: subc 6, 6, 7
+; PPC32-NEXT: xor 4, 4, 7
+; PPC32-NEXT: subfe 5, 7, 5
+; PPC32-NEXT: xor 3, 3, 7
+; PPC32-NEXT: subfe 4, 7, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_cmp_i128:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 4, 4, 6
+; PPC64-NEXT: li 7, 0
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: mfxer 5
+; PPC64-NEXT: addze 6, 7
+; PPC64-NEXT: addic 6, 6, -1
+; PPC64-NEXT: mtxer 5
+; PPC64-NEXT: subfe 5, 6, 6
+; PPC64-NEXT: xor 4, 4, 5
+; PPC64-NEXT: xor 3, 3, 5
+; PPC64-NEXT: subc 4, 4, 5
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: blr
+ %cmp = icmp uge i128 %a, %b
+ %ab = sub i128 %a, %b
+ %ba = sub i128 %b, %a
+ %sel = select i1 %cmp, i128 %ab, i128 %ba
+ ret i128 %sel
+}
+
+;
+; sub(select(icmp(a,b),a,b),select(icmp(a,b),b,a)) -> abdu(a,b)
+;
+
+define i8 @abd_select_i8(i8 %a, i8 %b) nounwind {
+; PPC32-LABEL: abd_select_i8:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 24
+; PPC32-NEXT: clrlwi 4, 4, 24
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_select_i8:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 56
+; PPC64-NEXT: clrldi 4, 4, 56
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp ult i8 %a, %b
+ %ab = select i1 %cmp, i8 %a, i8 %b
+ %ba = select i1 %cmp, i8 %b, i8 %a
+ %sub = sub i8 %ba, %ab
+ ret i8 %sub
+}
+
+define i16 @abd_select_i16(i16 %a, i16 %b) nounwind {
+; PPC32-LABEL: abd_select_i16:
+; PPC32: # %bb.0:
+; PPC32-NEXT: clrlwi 3, 3, 16
+; PPC32-NEXT: clrlwi 4, 4, 16
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: srawi 4, 3, 31
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_select_i16:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 48
+; PPC64-NEXT: clrldi 4, 4, 48
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp ule i16 %a, %b
+ %ab = select i1 %cmp, i16 %a, i16 %b
+ %ba = select i1 %cmp, i16 %b, i16 %a
+ %sub = sub i16 %ba, %ab
+ ret i16 %sub
+}
+
+define i32 @abd_select_i32(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_select_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_select_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp ugt i32 %a, %b
+ %ab = select i1 %cmp, i32 %a, i32 %b
+ %ba = select i1 %cmp, i32 %b, i32 %a
+ %sub = sub i32 %ab, %ba
+ ret i32 %sub
+}
+
+define i64 @abd_select_i64(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_select_i64:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_select_i64:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %cmp = icmp uge i64 %a, %b
+ %ab = select i1 %cmp, i64 %a, i64 %b
+ %ba = select i1 %cmp, i64 %b, i64 %a
+ %sub = sub i64 %ab, %ba
+ ret i64 %sub
+}
+
+define i128 @abd_select_i128(i128 %a, i128 %b) nounwind {
+; PPC32-LABEL: abd_select_i128:
+; PPC32: # %bb.0:
+; PPC32-NEXT: subc 6, 6, 10
+; PPC32-NEXT: subfe 5, 9, 5
+; PPC32-NEXT: subfe 4, 8, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: mfxer 7
+; PPC32-NEXT: li 11, 0
+; PPC32-NEXT: addze 8, 11
+; PPC32-NEXT: addic 8, 8, -1
+; PPC32-NEXT: mtxer 7
+; PPC32-NEXT: subfe 7, 8, 8
+; PPC32-NEXT: xor 6, 6, 7
+; PPC32-NEXT: xor 5, 5, 7
+; PPC32-NEXT: subc 6, 6, 7
+; PPC32-NEXT: xor 4, 4, 7
+; PPC32-NEXT: subfe 5, 7, 5
+; PPC32-NEXT: xor 3, 3, 7
+; PPC32-NEXT: subfe 4, 7, 4
+; PPC32-NEXT: subfe 3, 7, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_select_i128:
+; PPC64: # %bb.0:
+; PPC64-NEXT: subc 4, 4, 6
+; PPC64-NEXT: li 7, 0
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: mfxer 5
+; PPC64-NEXT: addze 6, 7
+; PPC64-NEXT: addic 6, 6, -1
+; PPC64-NEXT: mtxer 5
+; PPC64-NEXT: subfe 5, 6, 6
+; PPC64-NEXT: xor 4, 4, 5
+; PPC64-NEXT: xor 3, 3, 5
+; PPC64-NEXT: subc 4, 4, 5
+; PPC64-NEXT: subfe 3, 5, 3
+; PPC64-NEXT: blr
+ %cmp = icmp ult i128 %a, %b
+ %ab = select i1 %cmp, i128 %a, i128 %b
+ %ba = select i1 %cmp, i128 %b, i128 %a
+ %sub = sub i128 %ba, %ab
+ ret i128 %sub
+}
+
+;
+; Operand is guaranteed non-zero: -(x | 1) cannot be 0
+;
+
+define i32 @abd_nonzero_rhs_i32(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_nonzero_rhs_i32:
+; PPC32: # %bb.0:
+; PPC32-NEXT: ori 4, 4, 1
+; PPC32-NEXT: addc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_nonzero_rhs_i32:
+; PPC64: # %bb.0:
+; PPC64-NEXT: ori 4, 4, 1
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: neg 4, 4
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: blr
+ %b1 = or i32 %b, 1
+ %nb = sub i32 0, %b1
+ %aext = zext i32 %a to i64
+ %bext = zext i32 %nb to i64
+ %sub = sub i64 %aext, %bext
+ %abs = call i64 @llvm.abs.i64(i64 %sub, i1 false)
+ %trunc = trunc i64 %abs to i32
+ ret i32 %trunc
+}
+
+define i64 @abd_nonzero_rhs_i64(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_nonzero_rhs_i64:
+; PPC32: # %bb.0:
+; PPC32-NEXT: ori 6, 6, 1
+; PPC32-NEXT: subfic 6, 6, 0
+; PPC32-NEXT: subfze 5, 5
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_nonzero_rhs_i64:
+; PPC64: # %bb.0:
+; PPC64-NEXT: ori 4, 4, 1
+; PPC64-NEXT: addc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %b1 = or i64 %b, 1
+ %nb = sub i64 0, %b1
+ %aext = zext i64 %a to i128
+ %bext = zext i64 %nb to i128
+ %sub = sub i128 %aext, %bext
+ %abs = call i128 @llvm.abs.i128(i128 %sub, i1 false)
+ %trunc = trunc i128 %abs to i64
+ ret i64 %trunc
+}
+
+define i32 @abd_nonzero_rhs_i32_select(i32 %a, i32 %b) nounwind {
+; PPC32-LABEL: abd_nonzero_rhs_i32_select:
+; PPC32: # %bb.0:
+; PPC32-NEXT: ori 4, 4, 1
+; PPC32-NEXT: addc 3, 3, 4
+; PPC32-NEXT: subfe 4, 3, 3
+; PPC32-NEXT: xor 3, 3, 4
+; PPC32-NEXT: sub 3, 3, 4
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_nonzero_rhs_i32_select:
+; PPC64: # %bb.0:
+; PPC64-NEXT: ori 4, 4, 1
+; PPC64-NEXT: clrldi 3, 3, 32
+; PPC64-NEXT: neg 4, 4
+; PPC64-NEXT: clrldi 4, 4, 32
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: sradi 4, 3, 63
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %b1 = or i32 %b, 1
+ %nb = sub i32 0, %b1
+ %cmp = icmp ugt i32 %a, %nb
+ %ab = sub i32 %a, %nb
+ %ba = sub i32 %nb, %a
+ %sel = select i1 %cmp, i32 %ab, i32 %ba
+ ret i32 %sel
+}
+
+define i64 @abd_nonzero_rhs_i64_select(i64 %a, i64 %b) nounwind {
+; PPC32-LABEL: abd_nonzero_rhs_i64_select:
+; PPC32: # %bb.0:
+; PPC32-NEXT: ori 6, 6, 1
+; PPC32-NEXT: subfic 6, 6, 0
+; PPC32-NEXT: subfze 5, 5
+; PPC32-NEXT: subc 4, 4, 6
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: mfxer 5
+; PPC32-NEXT: li 7, 0
+; PPC32-NEXT: addze 6, 7
+; PPC32-NEXT: addic 6, 6, -1
+; PPC32-NEXT: mtxer 5
+; PPC32-NEXT: subfe 5, 6, 6
+; PPC32-NEXT: xor 4, 4, 5
+; PPC32-NEXT: xor 3, 3, 5
+; PPC32-NEXT: subc 4, 4, 5
+; PPC32-NEXT: subfe 3, 5, 3
+; PPC32-NEXT: blr
+;
+; PPC64-LABEL: abd_nonzero_rhs_i64_select:
+; PPC64: # %bb.0:
+; PPC64-NEXT: ori 4, 4, 1
+; PPC64-NEXT: addc 3, 3, 4
+; PPC64-NEXT: subfe 4, 3, 3
+; PPC64-NEXT: xor 3, 3, 4
+; PPC64-NEXT: sub 3, 3, 4
+; PPC64-NEXT: blr
+ %b1 = or i64 %b, 1
+ %nb = sub i64 0, %b1
+ %cmp = icmp ugt i64 %a, %nb
+ %ab = sub i64 %a, %nb
+ %ba = sub i64 %nb, %a
+ %sel = select i1 %cmp, i64 %ab, i64 %ba
+ ret i64 %sel
+}
+
+declare i8 @llvm.abs.i8(i8, i1)
+declare i16 @llvm.abs.i16(i16, i1)
+declare i32 @llvm.abs.i32(i32, i1)
----------------
lei137 wrote:
same comment as for abdu-neg.ll
https://github.com/llvm/llvm-project/pull/182704
More information about the llvm-commits
mailing list