diff options
author | Richard Sandiford <rsandifo@linux.vnet.ibm.com> | 2013-07-03 10:10:02 +0000 |
---|---|---|
committer | Richard Sandiford <rsandifo@linux.vnet.ibm.com> | 2013-07-03 10:10:02 +0000 |
commit | fa487e83a83c260d6a50f3df00a0eb012553a912 (patch) | |
tree | f6ddd72df044eaa9cabbce37fd4b04f64b978139 /test/CodeGen/SystemZ/int-sub-02.ll | |
parent | b81b477cd4392a51112c3af0659ea9fc176e74f1 (diff) | |
download | llvm-fa487e83a83c260d6a50f3df00a0eb012553a912.tar.gz llvm-fa487e83a83c260d6a50f3df00a0eb012553a912.tar.bz2 llvm-fa487e83a83c260d6a50f3df00a0eb012553a912.tar.xz |
[SystemZ] Fold more spills
Add a mapping from register-based <INSN>R instructions to the corresponding
memory-based <INSN>. Use it to cut down on the number of spill loads.
Some instructions extend their operands from smaller fields, so this
required a new TSFlags field to say how big the unextended operand is.
This optimisation doesn't trigger for C(G)R and CL(G)R because in practice
we always combine those instructions with a branch. Adding a test for every
other case probably seems excessive, but it did catch a missed optimisation
for DSGF (fixed in r185435).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185529 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test/CodeGen/SystemZ/int-sub-02.ll')
-rw-r--r-- | test/CodeGen/SystemZ/int-sub-02.ll | 78 |
1 files changed, 78 insertions, 0 deletions
diff --git a/test/CodeGen/SystemZ/int-sub-02.ll b/test/CodeGen/SystemZ/int-sub-02.ll index 5150a960a5..99d1c7b7c8 100644 --- a/test/CodeGen/SystemZ/int-sub-02.ll +++ b/test/CodeGen/SystemZ/int-sub-02.ll @@ -2,6 +2,8 @@ ; ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s +declare i64 @foo() + ; Check SGFR. define i64 @f1(i64 %a, i32 %b) { ; CHECK: f1: @@ -100,3 +102,79 @@ define i64 @f8(i64 %a, i64 %src, i64 %index) { %sub = sub i64 %a, %bext ret i64 %sub } + +; Check that subtractions of spilled values can use SGF rather than SGFR. +define i64 @f9(i32 *%ptr0) { +; CHECK: f9: +; CHECK: brasl %r14, foo@PLT +; CHECK: sgf %r2, 160(%r15) +; CHECK: br %r14 + %ptr1 = getelementptr i32 *%ptr0, i64 2 + %ptr2 = getelementptr i32 *%ptr0, i64 4 + %ptr3 = getelementptr i32 *%ptr0, i64 6 + %ptr4 = getelementptr i32 *%ptr0, i64 8 + %ptr5 = getelementptr i32 *%ptr0, i64 10 + %ptr6 = getelementptr i32 *%ptr0, i64 12 + %ptr7 = getelementptr i32 *%ptr0, i64 14 + %ptr8 = getelementptr i32 *%ptr0, i64 16 + %ptr9 = getelementptr i32 *%ptr0, i64 18 + + %val0 = load i32 *%ptr0 + %val1 = load i32 *%ptr1 + %val2 = load i32 *%ptr2 + %val3 = load i32 *%ptr3 + %val4 = load i32 *%ptr4 + %val5 = load i32 *%ptr5 + %val6 = load i32 *%ptr6 + %val7 = load i32 *%ptr7 + %val8 = load i32 *%ptr8 + %val9 = load i32 *%ptr9 + + %frob0 = add i32 %val0, 100 + %frob1 = add i32 %val1, 100 + %frob2 = add i32 %val2, 100 + %frob3 = add i32 %val3, 100 + %frob4 = add i32 %val4, 100 + %frob5 = add i32 %val5, 100 + %frob6 = add i32 %val6, 100 + %frob7 = add i32 %val7, 100 + %frob8 = add i32 %val8, 100 + %frob9 = add i32 %val9, 100 + + store i32 %frob0, i32 *%ptr0 + store i32 %frob1, i32 *%ptr1 + store i32 %frob2, i32 *%ptr2 + store i32 %frob3, i32 *%ptr3 + store i32 %frob4, i32 *%ptr4 + store i32 %frob5, i32 *%ptr5 + store i32 %frob6, i32 *%ptr6 + store i32 %frob7, i32 *%ptr7 + store i32 %frob8, i32 *%ptr8 + store i32 %frob9, i32 *%ptr9 + + %ret = call i64 @foo() + + %ext0 = sext i32 %frob0 to i64 + %ext1 = sext i32 %frob1 to i64 + %ext2 = sext i32 %frob2 to i64 + %ext3 = sext i32 %frob3 to i64 + %ext4 = sext i32 %frob4 to i64 + %ext5 = sext i32 %frob5 to i64 + %ext6 = sext i32 %frob6 to i64 + %ext7 = sext i32 %frob7 to i64 + %ext8 = sext i32 %frob8 to i64 + %ext9 = sext i32 %frob9 to i64 + + %sub0 = sub i64 %ret, %ext0 + %sub1 = sub i64 %sub0, %ext1 + %sub2 = sub i64 %sub1, %ext2 + %sub3 = sub i64 %sub2, %ext3 + %sub4 = sub i64 %sub3, %ext4 + %sub5 = sub i64 %sub4, %ext5 + %sub6 = sub i64 %sub5, %ext6 + %sub7 = sub i64 %sub6, %ext7 + %sub8 = sub i64 %sub7, %ext8 + %sub9 = sub i64 %sub8, %ext9 + + ret i64 %sub9 +} |