These can be handled via the VECTOR SHIFT LEFT/RIGHT DOUBLE family of instructions, depending on architecture level. Fixes: https://github.com/llvm/llvm-project/issues/129955
252 lines
7.4 KiB
LLVM
252 lines
7.4 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
|
|
; Test 128-bit funnel shifts.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z15 | FileCheck %s --check-prefix=Z15
|
|
|
|
declare i128 @llvm.fshl.i128(i128, i128, i128)
|
|
declare i128 @llvm.fshr.i128(i128, i128, i128)
|
|
|
|
; Funnel shift left by constant multiple of 8.
|
|
define i128 @f1(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v0, 0(%r4), 3
|
|
; CHECK-NEXT: vl %v1, 0(%r3), 3
|
|
; CHECK-NEXT: vsldb %v0, %v1, %v0, 4
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f1:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsldb %v0, %v1, %v0, 4
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshl.i128(i128 %a, i128 %b, i128 32)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift left by constant smaller than 8.
|
|
define i128 @f2(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v1, 0(%r3), 3
|
|
; CHECK-NEXT: vl %v0, 0(%r4), 3
|
|
; CHECK-NEXT: vrepib %v2, 5
|
|
; CHECK-NEXT: vsl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 123
|
|
; CHECK-NEXT: vsrlb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsrl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v1, %v0
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f2:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsld %v0, %v1, %v0, 5
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshl.i128(i128 %a, i128 %b, i128 5)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift left by some other constant.
|
|
define i128 @f3(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v1, 0(%r4), 3
|
|
; CHECK-NEXT: vrepib %v2, 86
|
|
; CHECK-NEXT: vsrlb %v1, %v1, %v2
|
|
; CHECK-NEXT: vl %v0, 0(%r3), 3
|
|
; CHECK-NEXT: vsrl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 42
|
|
; CHECK-NEXT: vslb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v0, %v1
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f3:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsldb %v1, %v1, %v0, 5
|
|
; Z15-NEXT: vsldb %v0, %v0, %v0, 5
|
|
; Z15-NEXT: vsld %v0, %v1, %v0, 2
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshl.i128(i128 %a, i128 %b, i128 42)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift left by a variable amount.
|
|
define i128 @f4(i128 %a, i128 %b, i128 %sh) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: l %r0, 12(%r5)
|
|
; CHECK-NEXT: vlvgp %v2, %r0, %r0
|
|
; CHECK-NEXT: vl %v1, 0(%r3), 3
|
|
; CHECK-NEXT: vrepb %v2, %v2, 15
|
|
; CHECK-NEXT: vslb %v1, %v1, %v2
|
|
; CHECK-NEXT: vl %v0, 0(%r4), 3
|
|
; CHECK-NEXT: vsl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 1
|
|
; CHECK-NEXT: xilf %r0, 4294967295
|
|
; CHECK-NEXT: vsrl %v0, %v0, %v2
|
|
; CHECK-NEXT: vlvgp %v2, %r0, %r0
|
|
; CHECK-NEXT: vrepb %v2, %v2, 15
|
|
; CHECK-NEXT: vsrlb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsrl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v1, %v0
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f4:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: l %r0, 12(%r5)
|
|
; Z15-NEXT: vlvgp %v2, %r0, %r0
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vrepb %v2, %v2, 15
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vslb %v1, %v1, %v2
|
|
; Z15-NEXT: vsl %v1, %v1, %v2
|
|
; Z15-NEXT: vrepib %v2, 1
|
|
; Z15-NEXT: xilf %r0, 4294967295
|
|
; Z15-NEXT: vsrl %v0, %v0, %v2
|
|
; Z15-NEXT: vlvgp %v2, %r0, %r0
|
|
; Z15-NEXT: vrepb %v2, %v2, 15
|
|
; Z15-NEXT: vsrlb %v0, %v0, %v2
|
|
; Z15-NEXT: vsrl %v0, %v0, %v2
|
|
; Z15-NEXT: vo %v0, %v1, %v0
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshl.i128(i128 %a, i128 %b, i128 %sh)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift right by constant multiple of 8.
|
|
define i128 @f5(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f5:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v0, 0(%r4), 3
|
|
; CHECK-NEXT: vl %v1, 0(%r3), 3
|
|
; CHECK-NEXT: vsldb %v0, %v1, %v0, 12
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f5:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsldb %v0, %v1, %v0, 12
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshr.i128(i128 %a, i128 %b, i128 32)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift right by constant smaller than 8.
|
|
define i128 @f6(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f6:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v1, 0(%r4), 3
|
|
; CHECK-NEXT: vl %v0, 0(%r3), 3
|
|
; CHECK-NEXT: vrepib %v2, 5
|
|
; CHECK-NEXT: vsrl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 123
|
|
; CHECK-NEXT: vslb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v0, %v1
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f6:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsrd %v0, %v1, %v0, 5
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshr.i128(i128 %a, i128 %b, i128 5)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift right by some other constant.
|
|
define i128 @f7(i128 %a, i128 %b) {
|
|
; CHECK-LABEL: f7:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: vl %v1, 0(%r4), 3
|
|
; CHECK-NEXT: vrepib %v2, 42
|
|
; CHECK-NEXT: vsrlb %v1, %v1, %v2
|
|
; CHECK-NEXT: vl %v0, 0(%r3), 3
|
|
; CHECK-NEXT: vsrl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 86
|
|
; CHECK-NEXT: vslb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v0, %v1
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f7:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: vl %v0, 0(%r4), 3
|
|
; Z15-NEXT: vl %v1, 0(%r3), 3
|
|
; Z15-NEXT: vsldb %v0, %v1, %v0, 11
|
|
; Z15-NEXT: vsldb %v1, %v1, %v1, 11
|
|
; Z15-NEXT: vsrd %v0, %v1, %v0, 2
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshr.i128(i128 %a, i128 %b, i128 42)
|
|
ret i128 %res
|
|
}
|
|
|
|
; Funnel shift right by a variable amount.
|
|
define i128 @f8(i128 %a, i128 %b, i128 %sh) {
|
|
; CHECK-LABEL: f8:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: l %r0, 12(%r5)
|
|
; CHECK-NEXT: vlvgp %v2, %r0, %r0
|
|
; CHECK-NEXT: vl %v1, 0(%r4), 3
|
|
; CHECK-NEXT: vrepb %v2, %v2, 15
|
|
; CHECK-NEXT: vsrlb %v1, %v1, %v2
|
|
; CHECK-NEXT: vl %v0, 0(%r3), 3
|
|
; CHECK-NEXT: vsrl %v1, %v1, %v2
|
|
; CHECK-NEXT: vrepib %v2, 1
|
|
; CHECK-NEXT: xilf %r0, 4294967295
|
|
; CHECK-NEXT: vsl %v0, %v0, %v2
|
|
; CHECK-NEXT: vlvgp %v2, %r0, %r0
|
|
; CHECK-NEXT: vrepb %v2, %v2, 15
|
|
; CHECK-NEXT: vslb %v0, %v0, %v2
|
|
; CHECK-NEXT: vsl %v0, %v0, %v2
|
|
; CHECK-NEXT: vo %v0, %v0, %v1
|
|
; CHECK-NEXT: vst %v0, 0(%r2), 3
|
|
; CHECK-NEXT: br %r14
|
|
;
|
|
; Z15-LABEL: f8:
|
|
; Z15: # %bb.0:
|
|
; Z15-NEXT: l %r0, 12(%r5)
|
|
; Z15-NEXT: vlvgp %v2, %r0, %r0
|
|
; Z15-NEXT: vl %v1, 0(%r4), 3
|
|
; Z15-NEXT: vrepb %v2, %v2, 15
|
|
; Z15-NEXT: vl %v0, 0(%r3), 3
|
|
; Z15-NEXT: vsrlb %v1, %v1, %v2
|
|
; Z15-NEXT: vsrl %v1, %v1, %v2
|
|
; Z15-NEXT: vrepib %v2, 1
|
|
; Z15-NEXT: xilf %r0, 4294967295
|
|
; Z15-NEXT: vsl %v0, %v0, %v2
|
|
; Z15-NEXT: vlvgp %v2, %r0, %r0
|
|
; Z15-NEXT: vrepb %v2, %v2, 15
|
|
; Z15-NEXT: vslb %v0, %v0, %v2
|
|
; Z15-NEXT: vsl %v0, %v0, %v2
|
|
; Z15-NEXT: vo %v0, %v0, %v1
|
|
; Z15-NEXT: vst %v0, 0(%r2), 3
|
|
; Z15-NEXT: br %r14
|
|
%res = tail call i128 @llvm.fshr.i128(i128 %a, i128 %b, i128 %sh)
|
|
ret i128 %res
|
|
}
|
|
|