76ea5feb1f
This generalizes the existing combine for concat(radd, radd) to any binops. For much the same reason as the existing code, pushing the concat up through the tree are hopefully quicker (or the same) as the existing two half-vector operations, and can help combine away the concat.
202 lines
8.3 KiB
LLVM
202 lines
8.3 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
|
|
; RUN: llc -mtriple=aarch64 -mattr=+fullfp16 -verify-machineinstrs -o - %s | FileCheck %s
|
|
|
|
|
|
define <8 x i16> @concat_add(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) {
|
|
; CHECK-LABEL: concat_add:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: add v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = add <4 x i16> %a, %b
|
|
%y = add <4 x i16> %c, %d
|
|
%z = shufflevector <4 x i16> %x, <4 x i16> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x i16> @concat_addtunc(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, <4 x i32> %d) {
|
|
; CHECK-LABEL: concat_addtunc:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: add v2.4s, v2.4s, v3.4s
|
|
; CHECK-NEXT: add v0.4s, v0.4s, v1.4s
|
|
; CHECK-NEXT: uzp1 v0.8h, v0.8h, v2.8h
|
|
; CHECK-NEXT: ret
|
|
%x = add <4 x i32> %a, %b
|
|
%y = add <4 x i32> %c, %d
|
|
%xt = trunc <4 x i32> %x to <4 x i16>
|
|
%yt = trunc <4 x i32> %y to <4 x i16>
|
|
%z = shufflevector <4 x i16> %xt, <4 x i16> %yt, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x i16> @concat_addtunc2(<4 x i32> %a, <4 x i32> %b, <4 x i32> %c, <4 x i32> %d) {
|
|
; CHECK-LABEL: concat_addtunc2:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: uzp1 v1.8h, v1.8h, v3.8h
|
|
; CHECK-NEXT: uzp1 v0.8h, v0.8h, v2.8h
|
|
; CHECK-NEXT: add v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%at = trunc <4 x i32> %a to <4 x i16>
|
|
%bt = trunc <4 x i32> %b to <4 x i16>
|
|
%ct = trunc <4 x i32> %c to <4 x i16>
|
|
%dt = trunc <4 x i32> %d to <4 x i16>
|
|
%x = add <4 x i16> %at, %bt
|
|
%y = add <4 x i16> %ct, %dt
|
|
%z = shufflevector <4 x i16> %x, <4 x i16> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x i16> @concat_sub(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) {
|
|
; CHECK-LABEL: concat_sub:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: sub v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = sub <4 x i16> %a, %b
|
|
%y = sub <4 x i16> %c, %d
|
|
%z = shufflevector <4 x i16> %x, <4 x i16> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x i16> @concat_mul(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) {
|
|
; CHECK-LABEL: concat_mul:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: mul v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = mul <4 x i16> %a, %b
|
|
%y = mul <4 x i16> %c, %d
|
|
%z = shufflevector <4 x i16> %x, <4 x i16> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x i16> @concat_xor(<4 x i16> %a, <4 x i16> %b, <4 x i16> %c, <4 x i16> %d) {
|
|
; CHECK-LABEL: concat_xor:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: eor v0.16b, v0.16b, v1.16b
|
|
; CHECK-NEXT: ret
|
|
%x = xor <4 x i16> %a, %b
|
|
%y = xor <4 x i16> %c, %d
|
|
%z = shufflevector <4 x i16> %x, <4 x i16> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x i16> %z
|
|
}
|
|
|
|
define <8 x half> @concat_fadd(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x half> %d) {
|
|
; CHECK-LABEL: concat_fadd:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: fadd v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = fadd <4 x half> %a, %b
|
|
%y = fadd <4 x half> %c, %d
|
|
%z = shufflevector <4 x half> %x, <4 x half> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x half> %z
|
|
}
|
|
|
|
define <8 x half> @concat_fmul(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x half> %d) {
|
|
; CHECK-LABEL: concat_fmul:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: fmul v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = fmul <4 x half> %a, %b
|
|
%y = fmul <4 x half> %c, %d
|
|
%z = shufflevector <4 x half> %x, <4 x half> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x half> %z
|
|
}
|
|
|
|
define <8 x half> @concat_min(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x half> %d) {
|
|
; CHECK-LABEL: concat_min:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: // kill: def $d1 killed $d1 def $q1
|
|
; CHECK-NEXT: // kill: def $d0 killed $d0 def $q0
|
|
; CHECK-NEXT: // kill: def $d3 killed $d3 def $q3
|
|
; CHECK-NEXT: // kill: def $d2 killed $d2 def $q2
|
|
; CHECK-NEXT: mov v1.d[1], v3.d[0]
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: fminnm v0.8h, v0.8h, v1.8h
|
|
; CHECK-NEXT: ret
|
|
%x = call <4 x half> @llvm.minnum.v4f16(<4 x half> %a, <4 x half> %b)
|
|
%y = call <4 x half> @llvm.minnum.v4f16(<4 x half> %c, <4 x half> %d)
|
|
%z = shufflevector <4 x half> %x, <4 x half> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x half> %z
|
|
}
|
|
|
|
define <8 x half> @concat_minmax(<4 x half> %a, <4 x half> %b, <4 x half> %c, <4 x half> %d) {
|
|
; CHECK-LABEL: concat_minmax:
|
|
; CHECK: // %bb.0:
|
|
; CHECK-NEXT: fmaxnm v2.4h, v2.4h, v3.4h
|
|
; CHECK-NEXT: fminnm v0.4h, v0.4h, v1.4h
|
|
; CHECK-NEXT: mov v0.d[1], v2.d[0]
|
|
; CHECK-NEXT: ret
|
|
%x = call <4 x half> @llvm.minnum.v4f16(<4 x half> %a, <4 x half> %b)
|
|
%y = call <4 x half> @llvm.maxnum.v4f16(<4 x half> %c, <4 x half> %d)
|
|
%z = shufflevector <4 x half> %x, <4 x half> %y, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
|
|
ret <8 x half> %z
|
|
}
|
|
|
|
define <16 x i8> @signOf_neon(ptr nocapture noundef readonly %a, ptr nocapture noundef readonly %b) {
|
|
; CHECK-LABEL: signOf_neon:
|
|
; CHECK: // %bb.0: // %entry
|
|
; CHECK-NEXT: ldp q1, q2, [x0]
|
|
; CHECK-NEXT: movi v0.16b, #1
|
|
; CHECK-NEXT: ldp q3, q4, [x1]
|
|
; CHECK-NEXT: cmhi v5.8h, v1.8h, v3.8h
|
|
; CHECK-NEXT: cmhi v6.8h, v2.8h, v4.8h
|
|
; CHECK-NEXT: cmhi v1.8h, v3.8h, v1.8h
|
|
; CHECK-NEXT: cmhi v2.8h, v4.8h, v2.8h
|
|
; CHECK-NEXT: uzp1 v3.16b, v5.16b, v6.16b
|
|
; CHECK-NEXT: uzp1 v1.16b, v1.16b, v2.16b
|
|
; CHECK-NEXT: and v0.16b, v3.16b, v0.16b
|
|
; CHECK-NEXT: orr v0.16b, v0.16b, v1.16b
|
|
; CHECK-NEXT: ret
|
|
entry:
|
|
%0 = load <8 x i16>, ptr %a, align 2
|
|
%add.ptr = getelementptr inbounds i8, ptr %a, i64 16
|
|
%1 = load <8 x i16>, ptr %add.ptr, align 2
|
|
%2 = load <8 x i16>, ptr %b, align 2
|
|
%add.ptr6 = getelementptr inbounds i8, ptr %b, i64 16
|
|
%3 = load <8 x i16>, ptr %add.ptr6, align 2
|
|
%cmp.i33 = icmp ugt <8 x i16> %0, %2
|
|
%cmp.i31 = icmp ugt <8 x i16> %1, %3
|
|
%cmp.i29 = icmp ugt <8 x i16> %2, %0
|
|
%cmp.i = icmp ugt <8 x i16> %3, %1
|
|
%vmovn.i38.neg = zext <8 x i1> %cmp.i33 to <8 x i8>
|
|
%vmovn.i37.neg = zext <8 x i1> %cmp.i31 to <8 x i8>
|
|
%4 = select <8 x i1> %cmp.i29, <8 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <8 x i8> %vmovn.i38.neg
|
|
%5 = select <8 x i1> %cmp.i, <8 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <8 x i8> %vmovn.i37.neg
|
|
%or.i = shufflevector <8 x i8> %4, <8 x i8> %5, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
|
|
ret <16 x i8> %or.i
|
|
}
|