RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Code Issues Projects Releases Wiki Activity
5e32ef0acd
llvm-mirror/test/CodeGen/SystemZ/cond-store-02.ll
Richard Sandiford b3ecd3b03e [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 12:30:20 +00:00

398 lines
10 KiB
LLVM
Raw Blame History

; Test 16-bit conditional stores that are presented as selects.
;
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
declare void @foo(i16 *)
; Test the simple case, with the loaded value first.
define void @f1(i16 *%ptr, i16 %alt, i32 %limit) {
; CHECK-LABEL: f1:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; ...and with the loaded value second
define void @f2(i16 *%ptr, i16 %alt, i32 %limit) {
; CHECK-LABEL: f2:
; CHECK-NOT: %r2
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %alt, i16 %orig
store i16 %res, i16 *%ptr
ret void
}
; Test cases where the value is explicitly sign-extended to 32 bits, with the
; loaded value first.
define void @f3(i16 *%ptr, i32 %alt, i32 %limit) {
; CHECK-LABEL: f3:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = sext i16 %orig to i32
%res = select i1 %cond, i32 %ext, i32 %alt
%trunc = trunc i32 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; ...and with the loaded value second
define void @f4(i16 *%ptr, i32 %alt, i32 %limit) {
; CHECK-LABEL: f4:
; CHECK-NOT: %r2
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = sext i16 %orig to i32
%res = select i1 %cond, i32 %alt, i32 %ext
%trunc = trunc i32 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; Test cases where the value is explicitly zero-extended to 32 bits, with the
; loaded value first.
define void @f5(i16 *%ptr, i32 %alt, i32 %limit) {
; CHECK-LABEL: f5:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = zext i16 %orig to i32
%res = select i1 %cond, i32 %ext, i32 %alt
%trunc = trunc i32 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; ...and with the loaded value second
define void @f6(i16 *%ptr, i32 %alt, i32 %limit) {
; CHECK-LABEL: f6:
; CHECK-NOT: %r2
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = zext i16 %orig to i32
%res = select i1 %cond, i32 %alt, i32 %ext
%trunc = trunc i32 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; Test cases where the value is explicitly sign-extended to 64 bits, with the
; loaded value first.
define void @f7(i16 *%ptr, i64 %alt, i32 %limit) {
; CHECK-LABEL: f7:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = sext i16 %orig to i64
%res = select i1 %cond, i64 %ext, i64 %alt
%trunc = trunc i64 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; ...and with the loaded value second
define void @f8(i16 *%ptr, i64 %alt, i32 %limit) {
; CHECK-LABEL: f8:
; CHECK-NOT: %r2
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = sext i16 %orig to i64
%res = select i1 %cond, i64 %alt, i64 %ext
%trunc = trunc i64 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; Test cases where the value is explicitly zero-extended to 64 bits, with the
; loaded value first.
define void @f9(i16 *%ptr, i64 %alt, i32 %limit) {
; CHECK-LABEL: f9:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = zext i16 %orig to i64
%res = select i1 %cond, i64 %ext, i64 %alt
%trunc = trunc i64 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; ...and with the loaded value second
define void @f10(i16 *%ptr, i64 %alt, i32 %limit) {
; CHECK-LABEL: f10:
; CHECK-NOT: %r2
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%ext = zext i16 %orig to i64
%res = select i1 %cond, i64 %alt, i64 %ext
%trunc = trunc i64 %res to i16
store i16 %trunc, i16 *%ptr
ret void
}
; Check the high end of the aligned STH range.
define void @f11(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f11:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sth %r3, 4094(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 2047
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check the next halfword up, which should use STHY instead of STH.
define void @f12(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f12:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sthy %r3, 4096(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 2048
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check the high end of the aligned STHY range.
define void @f13(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f13:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sthy %r3, 524286(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 262143
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check the next halfword up, which needs separate address logic.
; Other sequences besides this one would be OK.
define void @f14(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f14:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: agfi %r2, 524288
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 262144
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check the low end of the STHY range.
define void @f15(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f15:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sthy %r3, -524288(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 -262144
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check the next halfword down, which needs separate address logic.
; Other sequences besides this one would be OK.
define void @f16(i16 *%base, i16 %alt, i32 %limit) {
; CHECK-LABEL: f16:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: agfi %r2, -524290
; CHECK: sth %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%ptr = getelementptr i16 *%base, i64 -262145
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check that STHY allows an index.
define void @f17(i64 %base, i64 %index, i16 %alt, i32 %limit) {
; CHECK-LABEL: f17:
; CHECK-NOT: %r2
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r2
; CHECK: sthy %r4, 4096(%r3,%r2)
; CHECK: [[LABEL]]:
; CHECK: br %r14
%add1 = add i64 %base, %index
%add2 = add i64 %add1, 4096
%ptr = inttoptr i64 %add2 to i16 *
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; Check that volatile loads are not matched.
define void @f18(i16 *%ptr, i16 %alt, i32 %limit) {
; CHECK-LABEL: f18:
; CHECK: lh {{%r[0-5]}}, 0(%r2)
; CHECK: {{jl|jnl}} [[LABEL:[^ ]*]]
; CHECK: [[LABEL]]:
; CHECK: sth {{%r[0-5]}}, 0(%r2)
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load volatile i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; ...likewise stores. In this case we should have a conditional load into %r3.
define void @f19(i16 *%ptr, i16 %alt, i32 %limit) {
; CHECK-LABEL: f19:
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK: lh %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: sth %r3, 0(%r2)
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store volatile i16 %res, i16 *%ptr
ret void
}
; Check that atomic loads are not matched. The transformation is OK for
; the "unordered" case tested here, but since we don't try to handle atomic
; operations at all in this context, it seems better to assert that than
; to restrict the test to a stronger ordering.
define void @f20(i16 *%ptr, i16 %alt, i32 %limit) {
; FIXME: should use a normal load instead of CS.
; CHECK-LABEL: f20:
; CHECK: cs {{%r[0-9]+}},
; CHECK: jl
; CHECK: {{jl|jnl}} [[LABEL:[^ ]*]]
; CHECK: [[LABEL]]:
; CHECK: sth {{%r[0-9]+}},
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load atomic i16 *%ptr unordered, align 2
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
ret void
}
; ...likewise stores.
define void @f21(i16 *%ptr, i16 %alt, i32 %limit) {
; FIXME: should use a normal store instead of CS.
; CHECK-LABEL: f21:
; CHECK: jhe [[LABEL:[^ ]*]]
; CHECK: lh %r3, 0(%r2)
; CHECK: [[LABEL]]:
; CHECK: cs {{%r[0-9]+}},
; CHECK: br %r14
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store atomic i16 %res, i16 *%ptr unordered, align 2
ret void
}
; Try a frame index base.
define void @f22(i16 %alt, i32 %limit) {
; CHECK-LABEL: f22:
; CHECK: brasl %r14, foo@PLT
; CHECK-NOT: %r15
; CHECK: jl [[LABEL:[^ ]*]]
; CHECK-NOT: %r15
; CHECK: sth {{%r[0-9]+}}, {{[0-9]+}}(%r15)
; CHECK: [[LABEL]]:
; CHECK: brasl %r14, foo@PLT
; CHECK: br %r14
%ptr = alloca i16
call void @foo(i16 *%ptr)
%cond = icmp ult i32 %limit, 42
%orig = load i16 *%ptr
%res = select i1 %cond, i16 %orig, i16 %alt
store i16 %res, i16 *%ptr
call void @foo(i16 *%ptr)
ret void
}
Reference in New Issue View Git Blame Copy Permalink