From c3ec80fea186b6dd956001467a1076c6c2ee4fb6 Mon Sep 17 00:00:00 2001 From: Ulrich Weigand Date: Mon, 30 Apr 2018 17:54:28 +0000 Subject: [SystemZ] Handle SADDO et.al. and ADD/SUBCARRY 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 --- llvm/test/CodeGen/SystemZ/int-ssub-05.ll | 186 +++++++++++++++++++++++++++++++ 1 file changed, 186 insertions(+) create mode 100644 llvm/test/CodeGen/SystemZ/int-ssub-05.ll (limited to 'llvm/test/CodeGen/SystemZ/int-ssub-05.ll') diff --git a/llvm/test/CodeGen/SystemZ/int-ssub-05.ll b/llvm/test/CodeGen/SystemZ/int-ssub-05.ll new file mode 100644 index 00000000000..d33f034a93c --- /dev/null +++ b/llvm/test/CodeGen/SystemZ/int-ssub-05.ll @@ -0,0 +1,186 @@ +; Test subtractions between an i64 and a sign-extended i16 on z14. +; +; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z14 | FileCheck %s + +declare i64 @foo() + +; Check SGH with no displacement. +define zeroext i1 @f1(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f1: +; CHECK: sgh %r3, 0(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %half = load i16, i16 *%src + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + ret i1 %obit +} + +; Check the high end of the aligned SGH range. +define zeroext i1 @f4(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f4: +; CHECK: sgh %r3, 524286(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %ptr = getelementptr i16, i16 *%src, i64 262143 + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%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(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f5: +; CHECK: agfi %r4, 524288 +; CHECK: sgh %r3, 0(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %ptr = getelementptr i16, i16 *%src, i64 262144 + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + ret i1 %obit +} + +; Check the high end of the negative aligned SGH range. +define zeroext i1 @f6(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f6: +; CHECK: sgh %r3, -2(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %ptr = getelementptr i16, i16 *%src, i64 -1 + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + ret i1 %obit +} + +; Check the low end of the SGH range. +define zeroext i1 @f7(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f7: +; CHECK: sgh %r3, -524288(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %ptr = getelementptr i16, i16 *%src, i64 -262144 + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%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(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f8: +; CHECK: agfi %r4, -524290 +; CHECK: sgh %r3, 0(%r4) +; CHECK-DAG: stg %r3, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %ptr = getelementptr i16, i16 *%src, i64 -262145 + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + ret i1 %obit +} + +; Check that SGH allows an index. +define zeroext i1 @f9(i64 %src, i64 %index, i64 %a, i64 *%res) { +; CHECK-LABEL: f9: +; CHECK: sgh %r4, 524284({{%r3,%r2|%r2,%r3}}) +; CHECK-DAG: stg %r4, 0(%r5) +; CHECK-DAG: lghi %r2, 0 +; CHECK-DAG: locghio %r2, 1 +; CHECK: br %r14 + %add1 = add i64 %src, %index + %add2 = add i64 %add1, 524284 + %ptr = inttoptr i64 %add2 to i16 * + %half = load i16, i16 *%ptr + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + ret i1 %obit +} + +; Check using the overflow result for a branch. +define void @f11(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f11: +; CHECK: sgh %r3, 0(%r4) +; CHECK: stg %r3, 0(%r5) +; CHECK: jgo foo@PLT +; CHECK: br %r14 + %half = load i16, i16 *%src + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + br i1 %obit, label %call, label %exit + +call: + tail call i64 @foo() + br label %exit + +exit: + ret void +} + +; ... and the same with the inverted direction. +define void @f12(i64 %dummy, i64 %a, i16 *%src, i64 *%res) { +; CHECK-LABEL: f12: +; CHECK: sgh %r3, 0(%r4) +; CHECK: stg %r3, 0(%r5) +; CHECK: jgno foo@PLT +; CHECK: br %r14 + %half = load i16, i16 *%src + %b = sext i16 %half to i64 + %t = call {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b) + %val = extractvalue {i64, i1} %t, 0 + %obit = extractvalue {i64, i1} %t, 1 + store i64 %val, i64 *%res + br i1 %obit, label %exit, label %call + +call: + tail call i64 @foo() + br label %exit + +exit: + ret void +} + + +declare {i64, i1} @llvm.ssub.with.overflow.i64(i64, i64) nounwind readnone + -- cgit v1.2.3