mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-23 19:23:23 +01:00
0c16dcd701
This provides an optimized implementation of SADDO/SSUBO/UADDO/USUBO as well as ADDCARRY/SUBCARRY on top of the new CC implementation. In particular, multi-word arithmetic now uses UADDO/ADDCARRY instead of the old ADDC/ADDE logic, which means we no longer need to use "glue" links for those instructions. This also allows making full use of the memory-based instructions like ALSI, which couldn't be recognized due to limitations in the DAG matcher previously. Also, the llvm.sadd.with.overflow et.al. intrinsincs now expand to directly using the ADD instructions and checking for a CC 3 result. llvm-svn: 331203
254 lines
7.6 KiB
LLVM
254 lines
7.6 KiB
LLVM
; Test 32-bit subtraction in which the second operand is a sign-extended
|
|
; i16 memory value.
|
|
;
|
|
; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
|
|
|
|
declare i32 @foo()
|
|
|
|
; Check the low end of the SH range.
|
|
define zeroext i1 @f1(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f1:
|
|
; CHECK: sh %r3, 0(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%half = load i16, i16 *%src
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the high end of the aligned SH range.
|
|
define zeroext i1 @f2(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f2:
|
|
; CHECK: sh %r3, 4094(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 2047
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the next halfword up, which should use SHY instead of SH.
|
|
define zeroext i1 @f3(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f3:
|
|
; CHECK: shy %r3, 4096(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 2048
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the high end of the aligned SHY range.
|
|
define zeroext i1 @f4(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f4:
|
|
; CHECK: shy %r3, 524286(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 262143
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the next halfword up, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define zeroext i1 @f5(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f5:
|
|
; CHECK: agfi %r4, 524288
|
|
; CHECK: sh %r3, 0(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 262144
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the high end of the negative aligned SHY range.
|
|
define zeroext i1 @f6(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f6:
|
|
; CHECK: shy %r3, -2(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 -1
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the low end of the SHY range.
|
|
define zeroext i1 @f7(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f7:
|
|
; CHECK: shy %r3, -524288(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 -262144
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check the next halfword down, which needs separate address logic.
|
|
; Other sequences besides this one would be OK.
|
|
define zeroext i1 @f8(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f8:
|
|
; CHECK: agfi %r4, -524290
|
|
; CHECK: sh %r3, 0(%r4)
|
|
; CHECK-DAG: st %r3, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%ptr = getelementptr i16, i16 *%src, i64 -262145
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check that SH allows an index.
|
|
define zeroext i1 @f9(i64 %src, i64 %index, i32 %a, i32 *%res) {
|
|
; CHECK-LABEL: f9:
|
|
; CHECK: sh %r4, 4094({{%r3,%r2|%r2,%r3}})
|
|
; CHECK-DAG: st %r4, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4094
|
|
%ptr = inttoptr i64 %add2 to i16 *
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check that SHY allows an index.
|
|
define zeroext i1 @f10(i64 %src, i64 %index, i32 %a, i32 *%res) {
|
|
; CHECK-LABEL: f10:
|
|
; CHECK: shy %r4, 4096({{%r3,%r2|%r2,%r3}})
|
|
; CHECK-DAG: st %r4, 0(%r5)
|
|
; CHECK-DAG: ipm [[REG:%r[0-5]]]
|
|
; CHECK-DAG: afi [[REG]], 1342177280
|
|
; CHECK-DAG: risbg %r2, [[REG]], 63, 191, 33
|
|
; CHECK: br %r14
|
|
%add1 = add i64 %src, %index
|
|
%add2 = add i64 %add1, 4096
|
|
%ptr = inttoptr i64 %add2 to i16 *
|
|
%half = load i16, i16 *%ptr
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
ret i1 %obit
|
|
}
|
|
|
|
; Check using the overflow result for a branch.
|
|
define void @f11(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f11:
|
|
; CHECK: sh %r3, 0(%r4)
|
|
; CHECK: st %r3, 0(%r5)
|
|
; CHECK: jgo foo@PLT
|
|
; CHECK: br %r14
|
|
%half = load i16, i16 *%src
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
br i1 %obit, label %call, label %exit
|
|
|
|
call:
|
|
tail call i32 @foo()
|
|
br label %exit
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
; ... and the same with the inverted direction.
|
|
define void @f12(i32 %dummy, i32 %a, i16 *%src, i32 *%res) {
|
|
; CHECK-LABEL: f12:
|
|
; CHECK: sh %r3, 0(%r4)
|
|
; CHECK: st %r3, 0(%r5)
|
|
; CHECK: jgno foo@PLT
|
|
; CHECK: br %r14
|
|
%half = load i16, i16 *%src
|
|
%b = sext i16 %half to i32
|
|
%t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
|
|
%val = extractvalue {i32, i1} %t, 0
|
|
%obit = extractvalue {i32, i1} %t, 1
|
|
store i32 %val, i32 *%res
|
|
br i1 %obit, label %exit, label %call
|
|
|
|
call:
|
|
tail call i32 @foo()
|
|
br label %exit
|
|
|
|
exit:
|
|
ret void
|
|
}
|
|
|
|
|
|
declare {i32, i1} @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone
|
|
|