| Ulrich Weigand | c3ec80f | 2018-04-30 17:54:28 +0000 | [diff] [blame] | 1 | ; Test the three-operand form of 32-bit subtraction. |
| 2 | ; |
| 3 | ; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z196 | FileCheck %s |
| 4 | |
| 5 | declare i32 @foo(i32, i32, i32) |
| 6 | |
| 7 | ; Check SLRK. |
| 8 | define i32 @f1(i32 %dummy, i32 %a, i32 %b, i32 *%flag) { |
| 9 | ; CHECK-LABEL: f1: |
| 10 | ; CHECK: slrk %r2, %r3, %r4 |
| 11 | ; CHECK: ipm [[REG:%r[0-5]]] |
| 12 | ; CHECK: afi [[REG]], -536870912 |
| 13 | ; CHECK: srl [[REG]], 31 |
| 14 | ; CHECK: st [[REG]], 0(%r5) |
| 15 | ; CHECK: br %r14 |
| 16 | %t = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) |
| 17 | %val = extractvalue {i32, i1} %t, 0 |
| 18 | %obit = extractvalue {i32, i1} %t, 1 |
| 19 | %ext = zext i1 %obit to i32 |
| 20 | store i32 %ext, i32 *%flag |
| 21 | ret i32 %val |
| 22 | } |
| 23 | |
| 24 | ; Check using the overflow result for a branch. |
| 25 | define i32 @f2(i32 %dummy, i32 %a, i32 %b) { |
| 26 | ; CHECK-LABEL: f2: |
| 27 | ; CHECK: slrk %r2, %r3, %r4 |
| 28 | ; CHECK-NEXT: bnler %r14 |
| 29 | ; CHECK: lhi %r2, 0 |
| 30 | ; CHECK: jg foo@PLT |
| 31 | %t = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) |
| 32 | %val = extractvalue {i32, i1} %t, 0 |
| 33 | %obit = extractvalue {i32, i1} %t, 1 |
| 34 | br i1 %obit, label %call, label %exit |
| 35 | |
| 36 | call: |
| 37 | %res = tail call i32 @foo(i32 0, i32 %a, i32 %b) |
| 38 | ret i32 %res |
| 39 | |
| 40 | exit: |
| 41 | ret i32 %val |
| 42 | } |
| 43 | |
| 44 | ; ... and the same with the inverted direction. |
| 45 | define i32 @f3(i32 %dummy, i32 %a, i32 %b) { |
| 46 | ; CHECK-LABEL: f3: |
| 47 | ; CHECK: slrk %r2, %r3, %r4 |
| 48 | ; CHECK-NEXT: bler %r14 |
| 49 | ; CHECK: lhi %r2, 0 |
| 50 | ; CHECK: jg foo@PLT |
| 51 | %t = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) |
| 52 | %val = extractvalue {i32, i1} %t, 0 |
| 53 | %obit = extractvalue {i32, i1} %t, 1 |
| 54 | br i1 %obit, label %exit, label %call |
| 55 | |
| 56 | call: |
| 57 | %res = tail call i32 @foo(i32 0, i32 %a, i32 %b) |
| 58 | ret i32 %res |
| 59 | |
| 60 | exit: |
| 61 | ret i32 %val |
| 62 | } |
| 63 | |
| 64 | ; Check that we can still use SLR in obvious cases. |
| 65 | define i32 @f4(i32 %a, i32 %b, i32 *%flag) { |
| 66 | ; CHECK-LABEL: f4: |
| 67 | ; CHECK: slr %r2, %r3 |
| 68 | ; CHECK: ipm [[REG:%r[0-5]]] |
| 69 | ; CHECK: afi [[REG]], -536870912 |
| 70 | ; CHECK: srl [[REG]], 31 |
| 71 | ; CHECK: st [[REG]], 0(%r4) |
| 72 | ; CHECK: br %r14 |
| 73 | %t = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b) |
| 74 | %val = extractvalue {i32, i1} %t, 0 |
| 75 | %obit = extractvalue {i32, i1} %t, 1 |
| 76 | %ext = zext i1 %obit to i32 |
| 77 | store i32 %ext, i32 *%flag |
| 78 | ret i32 %val |
| 79 | } |
| 80 | |
| 81 | declare {i32, i1} @llvm.usub.with.overflow.i32(i32, i32) nounwind readnone |
| 82 | |