2013-05-07 00:17:29 +08:00
|
|
|
; Test 8-bit atomic min/max operations.
|
|
|
|
;
|
2016-04-20 07:51:52 +08:00
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
2013-05-07 00:17:29 +08:00
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-SHIFT1
|
|
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-SHIFT2
|
|
|
|
|
|
|
|
; Check signed minimum.
|
|
|
|
; - CHECK is for the main loop.
|
|
|
|
; - CHECK-SHIFT1 makes sure that the negated shift count used by the second
|
|
|
|
; RLL is set up correctly. The negation is independent of the NILL and L
|
|
|
|
; tested in CHECK.
|
|
|
|
; - CHECK-SHIFT2 makes sure that %b is shifted into the high part of the word
|
|
|
|
; before being used, and that the low bits are set to 1. This sequence is
|
|
|
|
; independent of the other loop prologue instructions.
|
|
|
|
define i16 @f1(i16 *%src, i16 %b) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f1:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: risbg [[RISBG:%r[1-9]+]], %r2, 0, 189, 0{{$}}
|
2017-10-06 21:59:28 +08:00
|
|
|
; CHECK-DAG: sll %r2, 3
|
|
|
|
; CHECK-DAG: l [[OLD:%r[0-9]+]], 0([[RISBG]])
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: [[LOOP:\.[^:]*]]:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll [[ROT:%r[0-9]+]], [[OLD]], 0(%r2)
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK: crjle [[ROT]], %r3, [[KEEP:\..*]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], %r3, 32, 47, 0
|
|
|
|
; CHECK: [[KEEP]]:
|
|
|
|
; CHECK: rll [[NEW:%r[0-9]+]], [[ROT]], 0({{%r[1-9]+}})
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: cs [[OLD]], [[NEW]], 0([[RISBG]])
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jl [[LOOP]]
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll %r2, [[OLD]], 16(%r2)
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f1:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK-SHIFT1: sll %r2, 3
|
|
|
|
; CHECK-SHIFT1: lcr [[NEGSHIFT:%r[1-9]+]], %r2
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: rll {{%r[0-9]+}}, {{%r[0-9]+}}, 0([[NEGSHIFT]])
|
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f1:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: sll %r3, 16
|
|
|
|
; CHECK-SHIFT2: rll
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK-SHIFT2: crjle {{%r[0-9]+}}, %r3
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw min i16 *%src, i16 %b seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check signed maximum.
|
|
|
|
define i16 @f2(i16 *%src, i16 %b) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f2:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: risbg [[RISBG:%r[1-9]+]], %r2, 0, 189, 0{{$}}
|
2017-10-06 21:59:28 +08:00
|
|
|
; CHECK-DAG: sll %r2, 3
|
|
|
|
; CHECK-DAG: l [[OLD:%r[0-9]+]], 0([[RISBG]])
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: [[LOOP:\.[^:]*]]:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll [[ROT:%r[0-9]+]], [[OLD]], 0(%r2)
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK: crjhe [[ROT]], %r3, [[KEEP:\..*]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], %r3, 32, 47, 0
|
|
|
|
; CHECK: [[KEEP]]:
|
|
|
|
; CHECK: rll [[NEW:%r[0-9]+]], [[ROT]], 0({{%r[1-9]+}})
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: cs [[OLD]], [[NEW]], 0([[RISBG]])
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jl [[LOOP]]
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll %r2, [[OLD]], 16(%r2)
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f2:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK-SHIFT1: sll %r2, 3
|
|
|
|
; CHECK-SHIFT1: lcr [[NEGSHIFT:%r[1-9]+]], %r2
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: rll {{%r[0-9]+}}, {{%r[0-9]+}}, 0([[NEGSHIFT]])
|
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f2:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: sll %r3, 16
|
|
|
|
; CHECK-SHIFT2: rll
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK-SHIFT2: crjhe {{%r[0-9]+}}, %r3
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw max i16 *%src, i16 %b seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check unsigned minimum.
|
|
|
|
define i16 @f3(i16 *%src, i16 %b) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f3:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: risbg [[RISBG:%r[1-9]+]], %r2, 0, 189, 0{{$}}
|
2017-10-06 21:59:28 +08:00
|
|
|
; CHECK-DAG: sll %r2, 3
|
|
|
|
; CHECK-DAG: l [[OLD:%r[0-9]+]], 0([[RISBG]])
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: [[LOOP:\.[^:]*]]:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll [[ROT:%r[0-9]+]], [[OLD]], 0(%r2)
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK: clrjle [[ROT]], %r3, [[KEEP:\..*]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], %r3, 32, 47, 0
|
|
|
|
; CHECK: [[KEEP]]:
|
|
|
|
; CHECK: rll [[NEW:%r[0-9]+]], [[ROT]], 0({{%r[1-9]+}})
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: cs [[OLD]], [[NEW]], 0([[RISBG]])
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jl [[LOOP]]
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll %r2, [[OLD]], 16(%r2)
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f3:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK-SHIFT1: sll %r2, 3
|
|
|
|
; CHECK-SHIFT1: lcr [[NEGSHIFT:%r[1-9]+]], %r2
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: rll {{%r[0-9]+}}, {{%r[0-9]+}}, 0([[NEGSHIFT]])
|
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f3:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: sll %r3, 16
|
|
|
|
; CHECK-SHIFT2: rll
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK-SHIFT2: clrjle {{%r[0-9]+}}, %r3,
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw umin i16 *%src, i16 %b seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check unsigned maximum.
|
|
|
|
define i16 @f4(i16 *%src, i16 %b) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f4:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: risbg [[RISBG:%r[1-9]+]], %r2, 0, 189, 0{{$}}
|
2017-10-06 21:59:28 +08:00
|
|
|
; CHECK-DAG: sll %r2, 3
|
|
|
|
; CHECK-DAG: l [[OLD:%r[0-9]+]], 0([[RISBG]])
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: [[LOOP:\.[^:]*]]:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll [[ROT:%r[0-9]+]], [[OLD]], 0(%r2)
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK: clrjhe [[ROT]], %r3, [[KEEP:\..*]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], %r3, 32, 47, 0
|
|
|
|
; CHECK: [[KEEP]]:
|
|
|
|
; CHECK: rll [[NEW:%r[0-9]+]], [[ROT]], 0({{%r[1-9]+}})
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: cs [[OLD]], [[NEW]], 0([[RISBG]])
|
[SystemZ] Be more careful about inverting CC masks
System z branches have a mask to select which of the 4 CC values should
cause the branch to be taken. We can invert a branch by inverting the mask.
However, not all instructions can produce all 4 CC values, so inverting
the branch like this can lead to some oddities. For example, integer
comparisons only produce a CC of 0 (equal), 1 (less) or 2 (greater).
If an integer EQ is reversed to NE before instruction selection,
the branch will test for 1 or 2. If instead the branch is reversed
after instruction selection (by inverting the mask), it will test for
1, 2 or 3. Both are correct, but the second isn't really canonical.
This patch therefore keeps track of which CC values are possible
and uses this when inverting a mask.
Although this is mostly cosmestic, it fixes undefined behavior
for the CIJNLH in branch-08.ll. Another fix would have been
to mask out bit 0 when generating the fused compare and branch,
but the point of this patch is that we shouldn't need to do that
in the first place.
The patch also makes it easier to reuse CC results from other instructions.
llvm-svn: 187495
2013-07-31 20:30:20 +08:00
|
|
|
; CHECK: jl [[LOOP]]
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK: rll %r2, [[OLD]], 16(%r2)
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f4:
|
2016-04-30 03:53:16 +08:00
|
|
|
; CHECK-SHIFT1: sll %r2, 3
|
|
|
|
; CHECK-SHIFT1: lcr [[NEGSHIFT:%r[1-9]+]], %r2
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: rll {{%r[0-9]+}}, {{%r[0-9]+}}, 0([[NEGSHIFT]])
|
|
|
|
; CHECK-SHIFT1: rll
|
|
|
|
; CHECK-SHIFT1: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f4:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: sll %r3, 16
|
|
|
|
; CHECK-SHIFT2: rll
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK-SHIFT2: clrjhe {{%r[0-9]+}}, %r3,
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: rll
|
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw umax i16 *%src, i16 %b seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the lowest useful signed minimum value. We need to load 0x80010000
|
|
|
|
; into the source register.
|
|
|
|
define i16 @f5(i16 *%src) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f5:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: llilh [[SRC2:%r[0-9]+]], 32769
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK: crjle [[ROT:%r[0-9]+]], [[SRC2]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], [[SRC2]], 32, 47, 0
|
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f5:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: br %r14
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f5:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw min i16 *%src, i16 -32767 seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the highest useful signed maximum value. We need to load 0x7ffe0000
|
|
|
|
; into the source register.
|
|
|
|
define i16 @f6(i16 *%src) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f6:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: llilh [[SRC2:%r[0-9]+]], 32766
|
2013-05-28 18:41:11 +08:00
|
|
|
; CHECK: crjhe [[ROT:%r[0-9]+]], [[SRC2]]
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], [[SRC2]], 32, 47, 0
|
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f6:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: br %r14
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f6:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw max i16 *%src, i16 32766 seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the lowest useful unsigned maximum value. We need to load 0x00010000
|
|
|
|
; into the source register.
|
|
|
|
define i16 @f7(i16 *%src) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f7:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: llilh [[SRC2:%r[0-9]+]], 1
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK: clrjle [[ROT:%r[0-9]+]], [[SRC2]],
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], [[SRC2]], 32, 47, 0
|
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f7:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: br %r14
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f7:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw umin i16 *%src, i16 1 seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|
|
|
|
|
|
|
|
; Check the highest useful unsigned maximum value. We need to load 0xfffe0000
|
|
|
|
; into the source register.
|
|
|
|
define i16 @f8(i16 *%src) {
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-LABEL: f8:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: llilh [[SRC2:%r[0-9]+]], 65534
|
2013-09-18 17:56:40 +08:00
|
|
|
; CHECK: clrjhe [[ROT:%r[0-9]+]], [[SRC2]],
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK: risbg [[ROT]], [[SRC2]], 32, 47, 0
|
|
|
|
; CHECK: br %r14
|
|
|
|
;
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT1-LABEL: f8:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT1: br %r14
|
2013-07-14 14:24:09 +08:00
|
|
|
; CHECK-SHIFT2-LABEL: f8:
|
2013-05-07 00:17:29 +08:00
|
|
|
; CHECK-SHIFT2: br %r14
|
|
|
|
%res = atomicrmw umax i16 *%src, i16 65534 seq_cst
|
|
|
|
ret i16 %res
|
|
|
|
}
|