RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-25 20:23:11 +01:00
Code Issues Projects Releases Wiki Activity
1fc1020e9e
llvm-mirror/unittests/CodeGen/GlobalISel/KnownBitsTest.cpp

581 lines
22 KiB
C++
Raw Normal View History

[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
//===- KnownBitsTest.cpp -------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "GISelMITest.h"
#include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsCst) {
[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
StringRef MIRString = " %3:_(s8) = G_CONSTANT i8 1\n"
" %4:_(s8) = COPY %3\n";
setUp(MIRString);
if (!TM)
return;
unsigned CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
unsigned SrcReg = FinalCopy->getOperand(1).getReg();
[globalisel][knownbits] Correct a typo that prevented a test working as intended llvm-svn: 371115
2019-09-05 22:25:52 +02:00
unsigned DstReg = FinalCopy->getOperand(0).getReg();
[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)1, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0xfe, Res.Zero.getZExtValue());
[globalisel] Support trivial COPY in GISelKnownBits Summary: Allow GISelKnownBits to look through the trivial case of TargetOpcode::COPY Reviewers: aditya_nandakumar Subscribers: rovka, hiraditya, volkan, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D67131 llvm-svn: 370955
2019-09-04 20:59:43 +02:00
KnownBits Res2 = Info.getKnownBits(DstReg);
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
}
[globalisel] Support trivial COPY in GISelKnownBits Summary: Allow GISelKnownBits to look through the trivial case of TargetOpcode::COPY Reviewers: aditya_nandakumar Subscribers: rovka, hiraditya, volkan, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D67131 llvm-svn: 370955
2019-09-04 20:59:43 +02:00
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsCstWithClass) {
[globalisel][knownbits] Account for missing type constraints Now that we look through copies, it's possible to visit registers that have a register class constraint but not a type constraint. Avoid looking through copies when this occurs as the SrcReg won't be able to determine it's bit width or any known bits. Along the same lines, if the initial query is on a register that doesn't have a type constraint then the result is a default-constructed KnownBits, that is, a 1-bit fully-unknown value. llvm-svn: 371116
2019-09-05 22:26:02 +02:00
StringRef MIRString = " %10:gpr32 = MOVi32imm 1\n"
" %4:_(s32) = COPY %10\n";
setUp(MIRString);
if (!TM)
return;
unsigned CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
unsigned SrcReg = FinalCopy->getOperand(1).getReg();
unsigned DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
// We can't analyze %3 due to the register class constraint. We will get a
// default-constructed KnownBits back.
EXPECT_EQ((uint64_t)1, Res.getBitWidth());
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
// We still don't know the values due to the register class constraint but %4
// did reveal the size of %3.
EXPECT_EQ((uint64_t)32, Res2.getBitWidth());
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
[GISelKnownBits] Add support for PHIs Teach the GISelKnowBits analysis how to deal with PHI operations. PHIs are essentially COPYs happening on edges, so we can just reuse the code for COPY. This is NFC COPY-wise has we leave Depth untouched when calling computeKnownBitsImpl for COPYs, like it was before this patch. Increasing Depth is however required for PHIs as they may loop back to themselves and we would end up in an infinite loop if we were not increasing Depth. Differential Revision: https://reviews.llvm.org/D73317
2020-01-25 01:15:43 +01:00
// Check that we are able to track bits through PHIs
// and get the intersections of everything we know on each operand.
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsCstPHI) {
[GISelKnownBits] Add support for PHIs Teach the GISelKnowBits analysis how to deal with PHI operations. PHIs are essentially COPYs happening on edges, so we can just reuse the code for COPY. This is NFC COPY-wise has we leave Depth untouched when calling computeKnownBitsImpl for COPYs, like it was before this patch. Increasing Depth is however required for PHIs as they may loop back to themselves and we would end up in an infinite loop if we were not increasing Depth. Differential Revision: https://reviews.llvm.org/D73317
2020-01-25 01:15:43 +01:00
StringRef MIRString = " bb.10:\n"
" %10:_(s8) = G_CONSTANT i8 3\n"
" %11:_(s1) = G_IMPLICIT_DEF\n"
" G_BRCOND %11(s1), %bb.11\n"
" G_BR %bb.12\n"
"\n"
" bb.11:\n"
" %12:_(s8) = G_CONSTANT i8 2\n"
" G_BR %bb.12\n"
"\n"
" bb.12:\n"
" %13:_(s8) = PHI %10(s8), %bb.10, %12(s8), %bb.11\n"
" %14:_(s8) = COPY %13\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
Register DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)2, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0xfc, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
// Check that we report we know nothing when we hit a
// non-generic register.
// Note: this could be improved though!
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsCstPHIToNonGenericReg) {
[GISelKnownBits] Add support for PHIs Teach the GISelKnowBits analysis how to deal with PHI operations. PHIs are essentially COPYs happening on edges, so we can just reuse the code for COPY. This is NFC COPY-wise has we leave Depth untouched when calling computeKnownBitsImpl for COPYs, like it was before this patch. Increasing Depth is however required for PHIs as they may loop back to themselves and we would end up in an infinite loop if we were not increasing Depth. Differential Revision: https://reviews.llvm.org/D73317
2020-01-25 01:15:43 +01:00
StringRef MIRString = " bb.10:\n"
" %10:gpr32 = MOVi32imm 3\n"
" %11:_(s1) = G_IMPLICIT_DEF\n"
" G_BRCOND %11(s1), %bb.11\n"
" G_BR %bb.12\n"
"\n"
" bb.11:\n"
" %12:_(s8) = G_CONSTANT i8 2\n"
" G_BR %bb.12\n"
"\n"
" bb.12:\n"
" %13:_(s8) = PHI %10, %bb.10, %12(s8), %bb.11\n"
" %14:_(s8) = COPY %13\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
Register DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
[GISel][KnownBits] Give up on PHI analysis as soon as we don't know anything When analyzing PHIs, we gather the known bits for every operand and merge them together to get the known bits of the result of the PHI. It is not unusual that merging the information leads to know nothing on the result (e.g., phi a: i8 3, b: i8 unknown, ..., after looking at the second argument we know we will know nothing on the result), thus, as soon as we reach that state, stop analyzing the following operand (i.e., on the previous example, we won't process anything after looking at `b`). This improves compile time in particular with PHIs with a large number of operands. NFC.
2020-02-20 20:27:36 +01:00
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
// Check that we know nothing when at least one value of a PHI
// comes from something we cannot analysis.
// This test is not particularly interesting, it is just
// here to cover the code that stops the analysis of PHIs
// earlier. In that case, we would not even look at the
// second incoming value.
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsUnknownPHI) {
[GISel][KnownBits] Give up on PHI analysis as soon as we don't know anything When analyzing PHIs, we gather the known bits for every operand and merge them together to get the known bits of the result of the PHI. It is not unusual that merging the information leads to know nothing on the result (e.g., phi a: i8 3, b: i8 unknown, ..., after looking at the second argument we know we will know nothing on the result), thus, as soon as we reach that state, stop analyzing the following operand (i.e., on the previous example, we won't process anything after looking at `b`). This improves compile time in particular with PHIs with a large number of operands. NFC.
2020-02-20 20:27:36 +01:00
StringRef MIRString =
" bb.10:\n"
" %10:_(s64) = COPY %0\n"
" %11:_(s1) = G_IMPLICIT_DEF\n"
" G_BRCOND %11(s1), %bb.11\n"
" G_BR %bb.12\n"
"\n"
" bb.11:\n"
" %12:_(s64) = G_CONSTANT i64 2\n"
" G_BR %bb.12\n"
"\n"
" bb.12:\n"
" %13:_(s64) = PHI %10(s64), %bb.10, %12(s64), %bb.11\n"
" %14:_(s64) = COPY %13\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
Register DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
[GISelKnownBits] Add support for PHIs Teach the GISelKnowBits analysis how to deal with PHI operations. PHIs are essentially COPYs happening on edges, so we can just reuse the code for COPY. This is NFC COPY-wise has we leave Depth untouched when calling computeKnownBitsImpl for COPYs, like it was before this patch. Increasing Depth is however required for PHIs as they may loop back to themselves and we would end up in an infinite loop if we were not increasing Depth. Differential Revision: https://reviews.llvm.org/D73317
2020-01-25 01:15:43 +01:00
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
// Check that we manage to process PHIs that loop on themselves.
// For now, the analysis just stops and assumes it knows nothing,
// eventually we could teach it how to properly track phis that
// loop back.
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsCstPHIWithLoop) {
[GISelKnownBits] Add support for PHIs Teach the GISelKnowBits analysis how to deal with PHI operations. PHIs are essentially COPYs happening on edges, so we can just reuse the code for COPY. This is NFC COPY-wise has we leave Depth untouched when calling computeKnownBitsImpl for COPYs, like it was before this patch. Increasing Depth is however required for PHIs as they may loop back to themselves and we would end up in an infinite loop if we were not increasing Depth. Differential Revision: https://reviews.llvm.org/D73317
2020-01-25 01:15:43 +01:00
StringRef MIRString =
" bb.10:\n"
" %10:_(s8) = G_CONSTANT i8 3\n"
" %11:_(s1) = G_IMPLICIT_DEF\n"
" G_BRCOND %11(s1), %bb.11\n"
" G_BR %bb.12\n"
"\n"
" bb.11:\n"
" %12:_(s8) = G_CONSTANT i8 2\n"
" G_BR %bb.12\n"
"\n"
" bb.12:\n"
" %13:_(s8) = PHI %10(s8), %bb.10, %12(s8), %bb.11, %14(s8), %bb.12\n"
" %14:_(s8) = COPY %13\n"
" G_BR %bb.12\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
Register DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
[GISel][KnownBits] Update a comment regarding the effect of cache on PHIs Unlike what I claimed in my previous commit. The caching is actually not NFC on PHIs. When we put a big enough max depth, we end up simulating loops. The cache is effectively cutting the simulation short and we get less information as a result. E.g., ``` v0 = G_CONSTANT i8 0xC0 jump v1 = G_PHI i8 v0, v2 v2 = G_LSHR i8 v1, 1 ``` Let say we want the known bits of v1. - With cache: Set v1 cache to we know nothing v1 is v0 & v2 v0 gives us 0xC0 v2 gives us known bits of v1 >> 1 v1 is in the cache => v1 is 0, thus v2 is 0x80 Finally v1 is v0 & v2 => 0x80 - Without cache and enough depth to do two iteration of the loop: v1 is v0 & v2 v0 gives us 0xC0 v2 gives us known bits of v1 >> 1 v1 is v0 & v2 v0 is 0xC0 v2 is v1 >> 1 Reach the max depth for v1... unwinding v1 is know nothing v2 is 0x80 v0 is 0xC0 v1 is 0x80 v2 is 0xC0 v0 is 0xC0 v1 is 0xC0 Thus now v1 is 0xC0 instead of 0x80. I've added a unittest demonstrating that. NFC
2020-02-25 23:03:21 +01:00
// Check that we don't try to analysis PHIs progression.
// Setting a deep enough max depth would allow to effectively simulate
// what happens in the loop.
// Thus, with a deep enough depth, we could actually figure out
// that %14's zero known bits are actually at least what we know
// for %10, right shifted by one.
// However, this process is super expensive compile-time wise and
// we don't want to reach that conclusion while playing with max depth.
// For now, the analysis just stops and assumes it knows nothing
// on PHIs, but eventually we could teach it how to properly track
// phis that loop back without relying on the luck effect of max
// depth.
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsDecreasingCstPHIWithLoop) {
[GISel][KnownBits] Update a comment regarding the effect of cache on PHIs Unlike what I claimed in my previous commit. The caching is actually not NFC on PHIs. When we put a big enough max depth, we end up simulating loops. The cache is effectively cutting the simulation short and we get less information as a result. E.g., ``` v0 = G_CONSTANT i8 0xC0 jump v1 = G_PHI i8 v0, v2 v2 = G_LSHR i8 v1, 1 ``` Let say we want the known bits of v1. - With cache: Set v1 cache to we know nothing v1 is v0 & v2 v0 gives us 0xC0 v2 gives us known bits of v1 >> 1 v1 is in the cache => v1 is 0, thus v2 is 0x80 Finally v1 is v0 & v2 => 0x80 - Without cache and enough depth to do two iteration of the loop: v1 is v0 & v2 v0 gives us 0xC0 v2 gives us known bits of v1 >> 1 v1 is v0 & v2 v0 is 0xC0 v2 is v1 >> 1 Reach the max depth for v1... unwinding v1 is know nothing v2 is 0x80 v0 is 0xC0 v1 is 0x80 v2 is 0xC0 v0 is 0xC0 v1 is 0xC0 Thus now v1 is 0xC0 instead of 0x80. I've added a unittest demonstrating that. NFC
2020-02-25 23:03:21 +01:00
StringRef MIRString = " bb.10:\n"
" %10:_(s8) = G_CONSTANT i8 5\n"
" %11:_(s8) = G_CONSTANT i8 1\n"
"\n"
" bb.12:\n"
" %13:_(s8) = PHI %10(s8), %bb.10, %14(s8), %bb.12\n"
" %14:_(s8) = G_LSHR %13, %11\n"
" %15:_(s8) = COPY %14\n"
" G_BR %bb.12\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
Register DstReg = FinalCopy->getOperand(0).getReg();
GISelKnownBits Info(*MF, /*MaxDepth=*/24);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
// A single iteration on the PHI (%13) gives:
// %10 has known zero of 0xFA
// %12 has known zero of 0x80 (we shift right by one so high bit is zero)
// Therefore, %14's known zero are 0x80 shifted by one 0xC0.
// If we had simulated the loop we could have more zero bits, basically
// up to 0xFC (count leading zero of 5, + 1).
EXPECT_EQ((uint64_t)0xC0, Res.Zero.getZExtValue());
KnownBits Res2 = Info.getKnownBits(DstReg);
EXPECT_EQ(Res.One.getZExtValue(), Res2.One.getZExtValue());
EXPECT_EQ(Res.Zero.getZExtValue(), Res2.Zero.getZExtValue());
}
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsPtrToIntViceVersa) {
[GISel]: Fix a bug in KnownBits where we should have been using SizeInBits https://reviews.llvm.org/D66039 We were using getIndexSize instead of getIndexSizeInBits(). Added test case for G_PTRTOINT and G_INTTOPTR. llvm-svn: 368618
2019-08-12 23:28:12 +02:00
StringRef MIRString = " %3:_(s16) = G_CONSTANT i16 256\n"
" %4:_(p0) = G_INTTOPTR %3\n"
" %5:_(s32) = G_PTRTOINT %4\n"
" %6:_(s32) = COPY %5\n";
setUp(MIRString);
if (!TM)
return;
unsigned CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
unsigned SrcReg = FinalCopy->getOperand(1).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ(256u, Res.One.getZExtValue());
EXPECT_EQ(0xfffffeffu, Res.Zero.getZExtValue());
}
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBitsXOR) {
[GlobalISel]: Add KnownBits for G_XOR https://reviews.llvm.org/D66119 llvm-svn: 368648
2019-08-13 06:32:33 +02:00
StringRef MIRString = " %3:_(s8) = G_CONSTANT i8 4\n"
" %4:_(s8) = G_CONSTANT i8 7\n"
" %5:_(s8) = G_XOR %3, %4\n"
" %6:_(s8) = COPY %5\n";
setUp(MIRString);
if (!TM)
return;
unsigned CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
unsigned SrcReg = FinalCopy->getOperand(1).getReg();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(SrcReg);
EXPECT_EQ(3u, Res.One.getZExtValue());
EXPECT_EQ(252u, Res.Zero.getZExtValue());
}
[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestKnownBits) {
[GISel]: Add GISelKnownBits analysis https://reviews.llvm.org/D65698 This adds a KnownBits analysis pass for GISel. This was done as a pass (compared to static functions) so that we can add other features such as caching queries(within a pass and across passes) in the future. This patch only adds the basic pass boiler plate, and implements a lazy non caching knownbits implementation (ported from SelectionDAG). I've also hooked up the AArch64PreLegalizerCombiner pass to use this - there should be no compile time regression as the analysis is lazy. llvm-svn: 368065
2019-08-06 19:18:29 +02:00
StringRef MIR = " %3:_(s32) = G_TRUNC %0\n"
" %4:_(s32) = G_TRUNC %1\n"
" %5:_(s32) = G_CONSTANT i32 5\n"
" %6:_(s32) = G_CONSTANT i32 24\n"
" %7:_(s32) = G_CONSTANT i32 28\n"
" %14:_(p0) = G_INTTOPTR %7\n"
" %16:_(s32) = G_PTRTOINT %14\n"
" %8:_(s32) = G_SHL %3, %5\n"
" %9:_(s32) = G_SHL %4, %5\n"
" %10:_(s32) = G_OR %8, %6\n"
" %11:_(s32) = G_OR %9, %16\n"
" %12:_(s32) = G_MUL %10, %11\n"
" %13:_(s32) = COPY %12\n";
setUp(MIR);
if (!TM)
return;
unsigned CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
unsigned SrcReg = FinalCopy->getOperand(1).getReg();
GISelKnownBits Info(*MF);
KnownBits Known = Info.getKnownBits(SrcReg);
EXPECT_FALSE(Known.hasConflict());
EXPECT_EQ(0u, Known.One.getZExtValue());
EXPECT_EQ(31u, Known.Zero.getZExtValue());
APInt Zeroes = Info.getKnownZeroes(SrcReg);
EXPECT_EQ(Known.Zero, Zeroes);
}
GlobalISel: Add maskedValueIsZero and signBitIsZero to known bits I dropped the DemandedElts since it seems to be missing from some of the new interfaces, but not others. llvm-svn: 370389
2019-08-29 19:24:36 +02:00
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestSignBitIsZero) {
[GISel][UnitTest] Fix a bunch of tests that were not doing anything After r368065, all the tests using GISelMITest must call setUp() before doing anything, otherwise the TargetMachine is not going to be set up. A few tests added after that commit were not doing that and ended up testing effectively nothing. Fix the setup of all the tests and fix the failing tests. llvm-svn: 374595
2019-10-11 22:58:26 +02:00
setUp();
GlobalISel: Add maskedValueIsZero and signBitIsZero to known bits I dropped the DemandedElts since it seems to be missing from some of the new interfaces, but not others. llvm-svn: 370389
2019-08-29 19:24:36 +02:00
if (!TM)
return;
const LLT S32 = LLT::scalar(32);
[GISel][UnitTest] Fix a bunch of tests that were not doing anything After r368065, all the tests using GISelMITest must call setUp() before doing anything, otherwise the TargetMachine is not going to be set up. A few tests added after that commit were not doing that and ended up testing effectively nothing. Fix the setup of all the tests and fix the failing tests. llvm-svn: 374595
2019-10-11 22:58:26 +02:00
auto SignBit = B.buildConstant(S32, 0x80000000);
GlobalISel: Add maskedValueIsZero and signBitIsZero to known bits I dropped the DemandedElts since it seems to be missing from some of the new interfaces, but not others. llvm-svn: 370389
2019-08-29 19:24:36 +02:00
auto Zero = B.buildConstant(S32, 0);
GISelKnownBits KnownBits(*MF);
EXPECT_TRUE(KnownBits.signBitIsZero(Zero.getReg(0)));
EXPECT_FALSE(KnownBits.signBitIsZero(SignBit.getReg(0)));
}
GlobalISel: Start adding computeNumSignBits to GISelKnownBits
2020-01-04 20:13:06 +01:00
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestNumSignBitsConstant) {
GlobalISel: Start adding computeNumSignBits to GISelKnownBits
2020-01-04 20:13:06 +01:00
StringRef MIRString = " %3:_(s8) = G_CONSTANT i8 1\n"
" %4:_(s8) = COPY %3\n"
" %5:_(s8) = G_CONSTANT i8 -1\n"
" %6:_(s8) = COPY %5\n"
" %7:_(s8) = G_CONSTANT i8 127\n"
" %8:_(s8) = COPY %7\n"
" %9:_(s8) = G_CONSTANT i8 32\n"
" %10:_(s8) = COPY %9\n"
" %11:_(s8) = G_CONSTANT i8 -32\n"
" %12:_(s8) = COPY %11\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg1 = Copies[Copies.size() - 5];
Register CopyRegNeg1 = Copies[Copies.size() - 4];
Register CopyReg127 = Copies[Copies.size() - 3];
Register CopyReg32 = Copies[Copies.size() - 2];
Register CopyRegNeg32 = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
EXPECT_EQ(7u, Info.computeNumSignBits(CopyReg1));
EXPECT_EQ(8u, Info.computeNumSignBits(CopyRegNeg1));
EXPECT_EQ(1u, Info.computeNumSignBits(CopyReg127));
EXPECT_EQ(2u, Info.computeNumSignBits(CopyReg32));
EXPECT_EQ(3u, Info.computeNumSignBits(CopyRegNeg32));
}
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestNumSignBitsSext) {
GlobalISel: Start adding computeNumSignBits to GISelKnownBits
2020-01-04 20:13:06 +01:00
StringRef MIRString = " %3:_(p0) = G_IMPLICIT_DEF\n"
" %4:_(s8) = G_LOAD %3 :: (load 1)\n"
" %5:_(s32) = G_SEXT %4\n"
" %6:_(s32) = COPY %5\n"
" %7:_(s8) = G_CONSTANT i8 -1\n"
" %8:_(s32) = G_SEXT %7\n"
" %9:_(s32) = COPY %8\n";
setUp(MIRString);
if (!TM)
return;
Register CopySextLoad = Copies[Copies.size() - 2];
Register CopySextNeg1 = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
EXPECT_EQ(25u, Info.computeNumSignBits(CopySextLoad));
EXPECT_EQ(32u, Info.computeNumSignBits(CopySextNeg1));
}
GlobalISel: Prepare to allow other target unit tests Currently all GlobalISel unittests use a hardcoded AArch64 target machine. Factor this so I can write some for AMDGPU specific known bits unittests.
2020-03-22 23:12:25 +01:00
TEST_F(AArch64GISelMITest, TestNumSignBitsTrunc) {
GlobalISel: Start adding computeNumSignBits to GISelKnownBits
2020-01-04 20:13:06 +01:00
StringRef MIRString = " %3:_(p0) = G_IMPLICIT_DEF\n"
" %4:_(s32) = G_LOAD %3 :: (load 4)\n"
" %5:_(s8) = G_TRUNC %4\n"
" %6:_(s8) = COPY %5\n"
" %7:_(s32) = G_CONSTANT i32 -1\n"
" %8:_(s8) = G_TRUNC %7\n"
" %9:_(s8) = COPY %8\n"
" %10:_(s32) = G_CONSTANT i32 7\n"
" %11:_(s8) = G_TRUNC %10\n"
" %12:_(s8) = COPY %11\n";
setUp(MIRString);
if (!TM)
return;
Register CopyTruncLoad = Copies[Copies.size() - 3];
Register CopyTruncNeg1 = Copies[Copies.size() - 2];
Register CopyTrunc7 = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
EXPECT_EQ(1u, Info.computeNumSignBits(CopyTruncLoad));
EXPECT_EQ(8u, Info.computeNumSignBits(CopyTruncNeg1));
EXPECT_EQ(5u, Info.computeNumSignBits(CopyTrunc7));
}
AMDGPU/GlobalISel: Implement computeNumSignBitsForTargetInstr
2020-03-23 00:06:30 +01:00
TEST_F(AMDGPUGISelMITest, TestNumSignBitsTrunc) {
StringRef MIRString =
" %3:_(<4 x s32>) = G_IMPLICIT_DEF\n"
" %4:_(s32) = G_IMPLICIT_DEF\n"
" %5:_(s32) = G_AMDGPU_BUFFER_LOAD_UBYTE %3, %4, %4, %4, 0, 0, 0 :: (load 1)\n"
" %6:_(s32) = COPY %5\n"
" %7:_(s32) = G_AMDGPU_BUFFER_LOAD_SBYTE %3, %4, %4, %4, 0, 0, 0 :: (load 1)\n"
" %8:_(s32) = COPY %7\n"
" %9:_(s32) = G_AMDGPU_BUFFER_LOAD_USHORT %3, %4, %4, %4, 0, 0, 0 :: (load 2)\n"
" %10:_(s32) = COPY %9\n"
" %11:_(s32) = G_AMDGPU_BUFFER_LOAD_SSHORT %3, %4, %4, %4, 0, 0, 0 :: (load 2)\n"
" %12:_(s32) = COPY %11\n";
setUp(MIRString);
if (!TM)
return;
Register CopyLoadUByte = Copies[Copies.size() - 4];
Register CopyLoadSByte = Copies[Copies.size() - 3];
Register CopyLoadUShort = Copies[Copies.size() - 2];
Register CopyLoadSShort = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
EXPECT_EQ(24u, Info.computeNumSignBits(CopyLoadUByte));
EXPECT_EQ(25u, Info.computeNumSignBits(CopyLoadSByte));
EXPECT_EQ(16u, Info.computeNumSignBits(CopyLoadUShort));
EXPECT_EQ(17u, Info.computeNumSignBits(CopyLoadSShort));
}
AMDGPU/GlobalISel: Implement computeKnownAlignForTargetInstr We probably need to move where intrinsics are lowered to copies to make this useful.
2020-06-05 21:29:18 +02:00
TEST_F(AMDGPUGISelMITest, TestTargetKnownAlign) {
StringRef MIRString =
" %5:_(p4) = G_INTRINSIC intrinsic(@llvm.amdgcn.dispatch.ptr)\n"
" %6:_(p4) = COPY %5\n"
" %7:_(p4) = G_INTRINSIC intrinsic(@llvm.amdgcn.queue.ptr)\n"
" %8:_(p4) = COPY %7\n"
" %9:_(p4) = G_INTRINSIC intrinsic(@llvm.amdgcn.kernarg.segment.ptr)\n"
" %10:_(p4) = COPY %9\n"
" %11:_(p4) = G_INTRINSIC intrinsic(@llvm.amdgcn.implicitarg.ptr)\n"
" %12:_(p4) = COPY %11\n"
" %13:_(p4) = G_INTRINSIC intrinsic(@llvm.amdgcn.implicit.buffer.ptr)\n"
" %14:_(p4) = COPY %13\n";
setUp(MIRString);
if (!TM)
return;
Register CopyDispatchPtr = Copies[Copies.size() - 5];
Register CopyQueuePtr = Copies[Copies.size() - 4];
Register CopyKernargSegmentPtr = Copies[Copies.size() - 3];
Register CopyImplicitArgPtr = Copies[Copies.size() - 2];
Register CopyImplicitBufferPtr = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
EXPECT_EQ(Align(4), Info.computeKnownAlignment(CopyDispatchPtr));
EXPECT_EQ(Align(4), Info.computeKnownAlignment(CopyQueuePtr));
EXPECT_EQ(Align(4), Info.computeKnownAlignment(CopyKernargSegmentPtr));
EXPECT_EQ(Align(4), Info.computeKnownAlignment(CopyImplicitArgPtr));
EXPECT_EQ(Align(4), Info.computeKnownAlignment(CopyImplicitBufferPtr));
}
[GlobalISel] Fix computing known bits for loads with range metadata In GlobalISel, if you have a load into a small type with a range, you'll hit an assert if you try to compute known bits on it starting at a larger type. e.g. ``` %x:_(s8) = G_LOAD %whatever(p0) :: (load 1 ... !range !n) ... %y:_(s32) = G_SOMETHING %x ``` When we walk through G_SOMETHING and hit the load, the width of our known bits is 32. However, the width of the range is going to be 8. This will cause us to hit an assert. To fix this, make computeKnownBitsFromRangeMetadata zero extend or truncate the range type to match the bitwidth of the known bits we're calculating. Add a testcase in CodeGen/GlobalISel/KnownBitsTest.cpp to reflect that this works now. https://reviews.llvm.org/D85375
2020-08-06 01:32:55 +02:00
TEST_F(AArch64GISelMITest, TestMetadata) {
StringRef MIRString = " %imp:_(p0) = G_IMPLICIT_DEF\n"
" %load:_(s8) = G_LOAD %imp(p0) :: (load 1)\n"
" %ext:_(s32) = G_ZEXT %load(s8)\n"
" %cst:_(s32) = G_CONSTANT i32 1\n"
" %and:_(s32) = G_AND %ext, %cst\n"
" %copy:_(s32) = COPY %and(s32)\n";
setUp(MIRString);
if (!TM)
return;
Register CopyReg = Copies[Copies.size() - 1];
MachineInstr *FinalCopy = MRI->getVRegDef(CopyReg);
Register SrcReg = FinalCopy->getOperand(1).getReg();
// We need a load with a metadata range for this to break. Fudge the load in
// the string and replace it with something we can work with.
MachineInstr *And = MRI->getVRegDef(SrcReg);
MachineInstr *Ext = MRI->getVRegDef(And->getOperand(1).getReg());
MachineInstr *Load = MRI->getVRegDef(Ext->getOperand(1).getReg());
IntegerType *Int8Ty = Type::getInt8Ty(Context);
// Value must be in [0, 2)
Metadata *LowAndHigh[] = {
ConstantAsMetadata::get(ConstantInt::get(Int8Ty, 0)),
ConstantAsMetadata::get(ConstantInt::get(Int8Ty, 2))};
auto NewMDNode = MDNode::get(Context, LowAndHigh);
const MachineMemOperand *OldMMO = *Load->memoperands_begin();
MachineMemOperand NewMMO(OldMMO->getPointerInfo(), OldMMO->getFlags(),
OldMMO->getSizeInBits(), OldMMO->getAlign(),
OldMMO->getAAInfo(), NewMDNode);
MachineIRBuilder MIB(*Load);
MIB.buildLoad(Load->getOperand(0), Load->getOperand(1), NewMMO);
Load->eraseFromParent();
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(And->getOperand(1).getReg());
// We don't know what the result of the load is, so we don't know any ones.
EXPECT_TRUE(Res.One.isNullValue());
// We know that the value is in [0, 2). So, we don't know if the first bit
// is 0 or not. However, we do know that every other bit must be 0.
APInt Mask(Res.getBitWidth(), 1);
Mask.flipAllBits();
EXPECT_EQ(Mask.getZExtValue(), Res.Zero.getZExtValue());
}
[GISel] Correct the known bits of G_ANYEXT Known bits for G_ANYEXT was incorrectly using KnownBits::zext, causing us to treat the high bits as zero even though they're (by definition) unknown. Differential Revision: https://reviews.llvm.org/D86323
2020-08-21 02:03:18 +02:00
TEST_F(AArch64GISelMITest, TestKnownBitsExt) {
StringRef MIRString = " %c1:_(s16) = G_CONSTANT i16 1\n"
" %x:_(s16) = G_IMPLICIT_DEF\n"
" %y:_(s16) = G_AND %x, %c1\n"
" %anyext:_(s32) = G_ANYEXT %y(s16)\n"
" %r1:_(s32) = COPY %anyext\n"
" %zext:_(s32) = G_ZEXT %y(s16)\n"
" %r2:_(s32) = COPY %zext\n"
" %sext:_(s32) = G_SEXT %y(s16)\n"
" %r3:_(s32) = COPY %sext\n";
setUp(MIRString);
if (!TM)
return;
Register CopyRegAny = Copies[Copies.size() - 3];
Register CopyRegZ = Copies[Copies.size() - 2];
Register CopyRegS = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
MachineInstr *Copy;
Register SrcReg;
KnownBits Res;
Copy = MRI->getVRegDef(CopyRegAny);
SrcReg = Copy->getOperand(1).getReg();
Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)32, Res.getBitWidth());
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0x0000fffe, Res.Zero.getZExtValue());
Copy = MRI->getVRegDef(CopyRegZ);
SrcReg = Copy->getOperand(1).getReg();
Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)32, Res.getBitWidth());
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0xfffffffe, Res.Zero.getZExtValue());
Copy = MRI->getVRegDef(CopyRegS);
SrcReg = Copy->getOperand(1).getReg();
Res = Info.getKnownBits(SrcReg);
EXPECT_EQ((uint64_t)32, Res.getBitWidth());
EXPECT_EQ((uint64_t)0, Res.One.getZExtValue());
EXPECT_EQ((uint64_t)0xfffffffe, Res.Zero.getZExtValue());
}
GlobalISel: Implement known bits for G_MERGE_VALUES
2020-08-27 18:15:16 +02:00
TEST_F(AArch64GISelMITest, TestKnownBitsMergeValues) {
StringRef MIRString = R"(
%val0:_(s16) = G_CONSTANT i16 35224
%val1:_(s16) = G_CONSTANT i16 17494
%val2:_(s16) = G_CONSTANT i16 4659
%val3:_(s16) = G_CONSTANT i16 43981
%merge:_(s64) = G_MERGE_VALUES %val0, %val1, %val2, %val3
%mergecopy:_(s64) = COPY %merge
)";
setUp(MIRString);
if (!TM)
return;
const uint64_t TestVal = UINT64_C(0xabcd123344568998);
Register CopyMerge = Copies[Copies.size() - 1];
GISelKnownBits Info(*MF);
KnownBits Res = Info.getKnownBits(CopyMerge);
EXPECT_EQ(64u, Res.getBitWidth());
EXPECT_EQ(TestVal, Res.One.getZExtValue());
EXPECT_EQ(~TestVal, Res.Zero.getZExtValue());
}
Reference in New Issue Copy Permalink