; Test the handling of base + displacement addresses for large frames,
; in cases where both 12-bit and 20-bit displacements are allowed.
+; The tests here assume z10 register pressure, without the high words
+; being available.
+;
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | \
+; RUN: FileCheck -check-prefix=CHECK-NOFP %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 -disable-fp-elim | \
+; RUN: FileCheck -check-prefix=CHECK-FP %s
;
-; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck -check-prefix=CHECK-NOFP %s
-; RUN: llc < %s -mtriple=s390x-linux-gnu -disable-fp-elim | FileCheck -check-prefix=CHECK-FP %s
-
; This file tests what happens when a displacement is converted from
; being relative to the start of a frame object to being relative to
; the frame itself. In some cases the test is only possible if two
; emergency spill slots at 160(%r15), the amount that we need to allocate
; in order to put another object at offset 4088 is 4088 - 176 = 3912 bytes.
define void @f1() {
-; CHECK-NOFP: f1:
+; CHECK-NOFP-LABEL: f1:
; CHECK-NOFP: mvi 4095(%r15), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f1:
+; CHECK-FP-LABEL: f1:
; CHECK-FP: mvi 4095(%r11), 42
; CHECK-FP: br %r14
%region1 = alloca [3912 x i8], align 8
; Test the first offset that is out-of-range of the 12-bit form.
define void @f2() {
-; CHECK-NOFP: f2:
+; CHECK-NOFP-LABEL: f2:
; CHECK-NOFP: mviy 4096(%r15), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f2:
+; CHECK-FP-LABEL: f2:
; CHECK-FP: mviy 4096(%r11), 42
; CHECK-FP: br %r14
%region1 = alloca [3912 x i8], align 8
; The last in-range doubleword offset is 524280, so by the same reasoning
; as above, we need to allocate objects of 524280 - 176 = 524104 bytes.
define void @f3() {
-; CHECK-NOFP: f3:
+; CHECK-NOFP-LABEL: f3:
; CHECK-NOFP: mviy 524287(%r15), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f3:
+; CHECK-FP-LABEL: f3:
; CHECK-FP: mviy 524287(%r11), 42
; CHECK-FP: br %r14
%region1 = alloca [524104 x i8], align 8
; and the offset is also out of LAY's range, so expect a constant load
; followed by an addition.
define void @f4() {
-; CHECK-NOFP: f4:
+; CHECK-NOFP-LABEL: f4:
; CHECK-NOFP: llilh %r1, 8
; CHECK-NOFP: agr %r1, %r15
; CHECK-NOFP: mvi 0(%r1), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f4:
+; CHECK-FP-LABEL: f4:
; CHECK-FP: llilh %r1, 8
; CHECK-FP: agr %r1, %r11
; CHECK-FP: mvi 0(%r1), 42
; Add 4095 to the previous offset, to test the other end of the MVI range.
; The instruction will actually be STCY before frame lowering.
define void @f5() {
-; CHECK-NOFP: f5:
+; CHECK-NOFP-LABEL: f5:
; CHECK-NOFP: llilh %r1, 8
; CHECK-NOFP: agr %r1, %r15
; CHECK-NOFP: mvi 4095(%r1), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f5:
+; CHECK-FP-LABEL: f5:
; CHECK-FP: llilh %r1, 8
; CHECK-FP: agr %r1, %r11
; CHECK-FP: mvi 4095(%r1), 42
; Test the next offset after that, which uses MVIY instead of MVI.
define void @f6() {
-; CHECK-NOFP: f6:
+; CHECK-NOFP-LABEL: f6:
; CHECK-NOFP: llilh %r1, 8
; CHECK-NOFP: agr %r1, %r15
; CHECK-NOFP: mviy 4096(%r1), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f6:
+; CHECK-FP-LABEL: f6:
; CHECK-FP: llilh %r1, 8
; CHECK-FP: agr %r1, %r11
; CHECK-FP: mviy 4096(%r1), 42
; anchors 0x10000 bytes apart, so that the high part can be loaded using
; LLILH while still using MVI in more cases than 0x40000 anchors would.
define void @f7() {
-; CHECK-NOFP: f7:
+; CHECK-NOFP-LABEL: f7:
; CHECK-NOFP: llilh %r1, 23
; CHECK-NOFP: agr %r1, %r15
; CHECK-NOFP: mviy 65535(%r1), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f7:
+; CHECK-FP-LABEL: f7:
; CHECK-FP: llilh %r1, 23
; CHECK-FP: agr %r1, %r11
; CHECK-FP: mviy 65535(%r1), 42
; Keep the object-relative offset the same but bump the size of the
; objects by one doubleword.
define void @f8() {
-; CHECK-NOFP: f8:
+; CHECK-NOFP-LABEL: f8:
; CHECK-NOFP: llilh %r1, 24
; CHECK-NOFP: agr %r1, %r15
; CHECK-NOFP: mvi 7(%r1), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f8:
+; CHECK-FP-LABEL: f8:
; CHECK-FP: llilh %r1, 24
; CHECK-FP: agr %r1, %r11
; CHECK-FP: mvi 7(%r1), 42
; The LA then gets lowered into the LLILH/LA form. The exact sequence
; isn't that important though.
define void @f9() {
-; CHECK-NOFP: f9:
+; CHECK-NOFP-LABEL: f9:
; CHECK-NOFP: llilh [[R1:%r[1-5]]], 16
; CHECK-NOFP: la [[R2:%r[1-5]]], 8([[R1]],%r15)
; CHECK-NOFP: agfi [[R2]], 524288
; CHECK-NOFP: mvi 0([[R2]]), 42
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f9:
+; CHECK-FP-LABEL: f9:
; CHECK-FP: llilh [[R1:%r[1-5]]], 16
; CHECK-FP: la [[R2:%r[1-5]]], 8([[R1]],%r11)
; CHECK-FP: agfi [[R2]], 524288
; call-clobbered registers are live and no call-saved ones have been
; allocated).
define void @f10(i32 *%vptr) {
-; CHECK-NOFP: f10:
+; CHECK-NOFP-LABEL: f10:
; CHECK-NOFP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r15)
; CHECK-NOFP: llilh [[REGISTER]], 8
; CHECK-NOFP: agr [[REGISTER]], %r15
; CHECK-NOFP: lg [[REGISTER]], [[OFFSET]](%r15)
; CHECK-NOFP: br %r14
;
-; CHECK-FP: f10:
+; CHECK-FP-LABEL: f10:
; CHECK-FP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r11)
; CHECK-FP: llilh [[REGISTER]], 8
; CHECK-FP: agr [[REGISTER]], %r11
; And again with maximum register pressure. The only spill slots that the
; NOFP case needs are the emergency ones, so the offsets are the same as for f4.
-; However, the FP case uses %r11 as the frame pointer and must therefore
-; spill a second register. This leads to an extra displacement of 8.
+; The FP case needs to spill an extra register and is too dependent on
+; register allocation heuristics for a stable test.
define void @f11(i32 *%vptr) {
-; CHECK-NOFP: f11:
+; CHECK-NOFP-LABEL: f11:
; CHECK-NOFP: stmg %r6, %r15,
; CHECK-NOFP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r15)
; CHECK-NOFP: llilh [[REGISTER]], 8
; CHECK-NOFP: lg [[REGISTER]], [[OFFSET]](%r15)
; CHECK-NOFP: lmg %r6, %r15,
; CHECK-NOFP: br %r14
-;
-; CHECK-FP: f11:
-; CHECK-FP: stmg %r6, %r15,
-; CHECK-FP: stg [[REGISTER:%r[1-9][0-4]?]], [[OFFSET:160|168]](%r11)
-; CHECK-FP: llilh [[REGISTER]], 8
-; CHECK-FP: agr [[REGISTER]], %r11
-; CHECK-FP: mvi 8([[REGISTER]]), 42
-; CHECK-FP: lg [[REGISTER]], [[OFFSET]](%r11)
-; CHECK-FP: lmg %r6, %r15,
-; CHECK-FP: br %r14
%i0 = load volatile i32 *%vptr
%i1 = load volatile i32 *%vptr
%i3 = load volatile i32 *%vptr