maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/unittests/Support/BranchProbabilityTest.cpp

406 lines
16 KiB
C++
Raw Normal View History

Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
//===- unittest/Support/BranchProbabilityTest.cpp - BranchProbability tests -=//
//
Update the file headers across all of the LLVM projects in the monorepo to reflect the new license. We understand that people may be surprised that we're moving the header entirely to discuss the new license. We checked this carefully with the Foundation's lawyer and we believe this is the correct approach. Essentially, all code in the project is now made available by the LLVM project under our new license, so you will see that the license headers include that license only. Some of our contributors have contributed code under our old license, and accordingly, we have retained a copy of our old license notice in the top-level files in each project and repository. llvm-svn: 351636
2019-01-19 16:50:56 +08:00
// 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
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
//
//===----------------------------------------------------------------------===//
#include "llvm/Support/BranchProbability.h"
#include "llvm/Support/raw_ostream.h"
#include "gtest/gtest.h"
using namespace llvm;
namespace llvm {
Pass BranchProbability/BlockMass by value instead of const& as they are small. NFC. llvm-svn: 247357
2015-09-11 07:10:42 +08:00
void PrintTo(BranchProbability P, ::std::ostream *os) {
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
*os << P.getNumerator() << "/" << P.getDenominator();
}
}
namespace {
typedef BranchProbability BP;
TEST(BranchProbabilityTest, Accessors) {
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(306783378u, BP(1, 7).getNumerator());
EXPECT_EQ(1u << 31, BP(1, 7).getDenominator());
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
EXPECT_EQ(0u, BP::getZero().getNumerator());
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(1u << 31, BP::getZero().getDenominator());
EXPECT_EQ(1u << 31, BP::getOne().getNumerator());
EXPECT_EQ(1u << 31, BP::getOne().getDenominator());
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
}
TEST(BranchProbabilityTest, Operators) {
EXPECT_TRUE(BP(1, 7) < BP(2, 7));
EXPECT_TRUE(BP(1, 7) < BP(1, 4));
EXPECT_TRUE(BP(5, 7) < BP(3, 4));
EXPECT_FALSE(BP(1, 7) < BP(1, 7));
EXPECT_FALSE(BP(1, 7) < BP(2, 14));
EXPECT_FALSE(BP(4, 7) < BP(1, 2));
EXPECT_FALSE(BP(4, 7) < BP(3, 7));
EXPECT_FALSE(BP(1, 7) > BP(2, 7));
EXPECT_FALSE(BP(1, 7) > BP(1, 4));
EXPECT_FALSE(BP(5, 7) > BP(3, 4));
EXPECT_FALSE(BP(1, 7) > BP(1, 7));
EXPECT_FALSE(BP(1, 7) > BP(2, 14));
EXPECT_TRUE(BP(4, 7) > BP(1, 2));
EXPECT_TRUE(BP(4, 7) > BP(3, 7));
EXPECT_TRUE(BP(1, 7) <= BP(2, 7));
EXPECT_TRUE(BP(1, 7) <= BP(1, 4));
EXPECT_TRUE(BP(5, 7) <= BP(3, 4));
EXPECT_TRUE(BP(1, 7) <= BP(1, 7));
EXPECT_TRUE(BP(1, 7) <= BP(2, 14));
EXPECT_FALSE(BP(4, 7) <= BP(1, 2));
EXPECT_FALSE(BP(4, 7) <= BP(3, 7));
EXPECT_FALSE(BP(1, 7) >= BP(2, 7));
EXPECT_FALSE(BP(1, 7) >= BP(1, 4));
EXPECT_FALSE(BP(5, 7) >= BP(3, 4));
EXPECT_TRUE(BP(1, 7) >= BP(1, 7));
EXPECT_TRUE(BP(1, 7) >= BP(2, 14));
EXPECT_TRUE(BP(4, 7) >= BP(1, 2));
EXPECT_TRUE(BP(4, 7) >= BP(3, 7));
EXPECT_FALSE(BP(1, 7) == BP(2, 7));
EXPECT_FALSE(BP(1, 7) == BP(1, 4));
EXPECT_FALSE(BP(5, 7) == BP(3, 4));
EXPECT_TRUE(BP(1, 7) == BP(1, 7));
EXPECT_TRUE(BP(1, 7) == BP(2, 14));
EXPECT_FALSE(BP(4, 7) == BP(1, 2));
EXPECT_FALSE(BP(4, 7) == BP(3, 7));
EXPECT_TRUE(BP(1, 7) != BP(2, 7));
EXPECT_TRUE(BP(1, 7) != BP(1, 4));
EXPECT_TRUE(BP(5, 7) != BP(3, 4));
EXPECT_FALSE(BP(1, 7) != BP(1, 7));
EXPECT_FALSE(BP(1, 7) != BP(2, 14));
EXPECT_TRUE(BP(4, 7) != BP(1, 2));
EXPECT_TRUE(BP(4, 7) != BP(3, 7));
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_TRUE(BP(1, 7) == BP(2, 14));
EXPECT_TRUE(BP(1, 7) == BP(3, 21));
EXPECT_TRUE(BP(5, 7) == BP(25, 35));
EXPECT_TRUE(BP(99999998, 100000000) < BP(99999999, 100000000));
EXPECT_TRUE(BP(4, 8) == BP(400000000, 800000000));
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
}
SupportTest: Fix test names harder r207552, r207553 and r207554 all had bad test names. llvm-svn: 207560
2014-04-30 01:07:45 +08:00
TEST(BranchProbabilityTest, MoreOperators) {
Support: More BlockFrequencyTest => BranchProbabilityTest llvm-svn: 207554
2014-04-30 00:44:59 +08:00
BP A(4, 5);
BP B(4U << 29, 5U << 29);
BP C(3, 4);
EXPECT_TRUE(A == B);
EXPECT_FALSE(A != B);
EXPECT_FALSE(A < B);
EXPECT_FALSE(A > B);
EXPECT_TRUE(A <= B);
EXPECT_TRUE(A >= B);
EXPECT_FALSE(B == C);
EXPECT_TRUE(B != C);
EXPECT_FALSE(B < C);
EXPECT_TRUE(B > C);
EXPECT_FALSE(B <= C);
EXPECT_TRUE(B >= C);
BP BigZero(0, UINT32_MAX);
BP BigOne(UINT32_MAX, UINT32_MAX);
EXPECT_FALSE(BigZero == BigOne);
EXPECT_TRUE(BigZero != BigOne);
EXPECT_TRUE(BigZero < BigOne);
EXPECT_FALSE(BigZero > BigOne);
EXPECT_TRUE(BigZero <= BigOne);
EXPECT_FALSE(BigZero >= BigOne);
}
[BPI] Add multiplication by scalar operators to BranchProbability This patch just adds two operators to BranchProbability class: (BP * scalar) and (BP *= scalar). Reviewers: junbuml, chandlerc, sanjoy, vsk Reviewed By: chandlerc Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D32334 llvm-svn: 300945
2017-04-21 11:14:30 +08:00
TEST(BranchProbabilityTest, ArithmeticOperators) {
BP Z(0, 1);
BP O(1, 1);
BP H(1, 2);
BP Q(1, 4);
BP Q3(3, 4);
EXPECT_EQ(Z + O, O);
EXPECT_EQ(H + Z, H);
EXPECT_EQ(H + H, O);
EXPECT_EQ(Q + H, Q3);
EXPECT_EQ(Q + Q3, O);
EXPECT_EQ(H + Q3, O);
EXPECT_EQ(Q3 + Q3, O);
EXPECT_EQ(Z - O, Z);
EXPECT_EQ(O - Z, O);
EXPECT_EQ(O - H, H);
EXPECT_EQ(O - Q, Q3);
EXPECT_EQ(Q3 - H, Q);
EXPECT_EQ(Q - H, Z);
EXPECT_EQ(Q - Q3, Z);
EXPECT_EQ(Z * O, Z);
EXPECT_EQ(H * H, Q);
EXPECT_EQ(Q * O, Q);
EXPECT_EQ(O * O, O);
EXPECT_EQ(Z * Z, Z);
EXPECT_EQ(Z * 3, Z);
EXPECT_EQ(Q * 3, Q3);
EXPECT_EQ(H * 3, O);
EXPECT_EQ(Q3 * 2, O);
EXPECT_EQ(O * UINT32_MAX, O);
EXPECT_EQ(Z / 4, Z);
EXPECT_EQ(O / 4, Q);
EXPECT_EQ(Q3 / 3, Q);
EXPECT_EQ(H / 2, Q);
EXPECT_EQ(O / 2, H);
EXPECT_EQ(H / UINT32_MAX, Z);
BP Min(1, 1u << 31);
EXPECT_EQ(O / UINT32_MAX, Z);
EXPECT_EQ(Min * UINT32_MAX, O);
}
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
TEST(BranchProbabilityTest, getCompl) {
EXPECT_EQ(BP(5, 7), BP(2, 7).getCompl());
EXPECT_EQ(BP(2, 7), BP(5, 7).getCompl());
EXPECT_EQ(BP::getZero(), BP(7, 7).getCompl());
EXPECT_EQ(BP::getOne(), BP(0, 7).getCompl());
}
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
TEST(BranchProbabilityTest, scale) {
// Multiply by 1.0.
EXPECT_EQ(UINT64_MAX, BP(1, 1).scale(UINT64_MAX));
EXPECT_EQ(UINT64_MAX, BP(7, 7).scale(UINT64_MAX));
EXPECT_EQ(UINT32_MAX, BP(1, 1).scale(UINT32_MAX));
EXPECT_EQ(UINT32_MAX, BP(7, 7).scale(UINT32_MAX));
EXPECT_EQ(0u, BP(1, 1).scale(0));
EXPECT_EQ(0u, BP(7, 7).scale(0));
// Multiply by 0.0.
EXPECT_EQ(0u, BP(0, 1).scale(UINT64_MAX));
EXPECT_EQ(0u, BP(0, 1).scale(UINT64_MAX));
EXPECT_EQ(0u, BP(0, 1).scale(0));
auto Two63 = UINT64_C(1) << 63;
auto Two31 = UINT64_C(1) << 31;
// Multiply by 0.5.
EXPECT_EQ(Two63 - 1, BP(1, 2).scale(UINT64_MAX));
// Big fractions.
EXPECT_EQ(1u, BP(Two31, UINT32_MAX).scale(2));
EXPECT_EQ(Two31, BP(Two31, UINT32_MAX).scale(Two31 * 2));
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(9223372036854775807ULL, BP(Two31, UINT32_MAX).scale(UINT64_MAX));
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
// High precision.
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(UINT64_C(9223372045444710399),
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
BP(Two31 + 1, UINT32_MAX - 2).scale(UINT64_MAX));
}
TEST(BranchProbabilityTest, scaleByInverse) {
// Divide by 1.0.
EXPECT_EQ(UINT64_MAX, BP(1, 1).scaleByInverse(UINT64_MAX));
EXPECT_EQ(UINT64_MAX, BP(7, 7).scaleByInverse(UINT64_MAX));
EXPECT_EQ(UINT32_MAX, BP(1, 1).scaleByInverse(UINT32_MAX));
EXPECT_EQ(UINT32_MAX, BP(7, 7).scaleByInverse(UINT32_MAX));
EXPECT_EQ(0u, BP(1, 1).scaleByInverse(0));
EXPECT_EQ(0u, BP(7, 7).scaleByInverse(0));
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
auto MAX_DENOMINATOR = BP::getDenominator();
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
// Divide by something very small.
EXPECT_EQ(UINT64_MAX, BP(1, UINT32_MAX).scaleByInverse(UINT64_MAX));
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(uint64_t(UINT32_MAX) * MAX_DENOMINATOR,
BP(1, MAX_DENOMINATOR).scaleByInverse(UINT32_MAX));
EXPECT_EQ(MAX_DENOMINATOR, BP(1, MAX_DENOMINATOR).scaleByInverse(1));
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
auto Two63 = UINT64_C(1) << 63;
auto Two31 = UINT64_C(1) << 31;
// Divide by 0.5.
EXPECT_EQ(UINT64_MAX - 1, BP(1, 2).scaleByInverse(Two63 - 1));
EXPECT_EQ(UINT64_MAX, BP(1, 2).scaleByInverse(Two63));
// Big fractions.
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(2u, BP(Two31, UINT32_MAX).scaleByInverse(1));
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
EXPECT_EQ(2u, BP(Two31 - 1, UINT32_MAX).scaleByInverse(1));
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(Two31 * 2, BP(Two31, UINT32_MAX).scaleByInverse(Two31));
EXPECT_EQ(Two31 * 2, BP(Two31 - 1, UINT32_MAX).scaleByInverse(Two31));
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
EXPECT_EQ(UINT64_MAX, BP(Two31, UINT32_MAX).scaleByInverse(Two63 + Two31));
// High precision. The exact answers to these are close to the successors of
// the floor. If we were rounding, these would round up.
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
EXPECT_EQ(UINT64_C(18446744060824649767),
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
BP(Two31 + 2, UINT32_MAX - 2)
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
.scaleByInverse(UINT64_C(9223372047592194056)));
EXPECT_EQ(UINT64_C(18446744060824649739),
Support: Add BranchProbability::scale() and ::scaleByInverse() Add API to `BranchProbability` for scaling big integers. Next job is to rip the logic out of `BlockMass` and `BlockFrequency`. llvm-svn: 207544
2014-04-30 00:15:35 +08:00
BP(Two31 + 1, UINT32_MAX).scaleByInverse(Two63 + Two31));
}
SupportTest: Fix test names harder r207552, r207553 and r207554 all had bad test names. llvm-svn: 207560
2014-04-30 01:07:45 +08:00
TEST(BranchProbabilityTest, scaleBruteForce) {
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
struct {
uint64_t Num;
uint32_t Prob[2];
uint64_t Result;
} Tests[] = {
// Data for scaling that results in <= 64 bit division.
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x1423e2a50ULL, { 0x64819521, 0x7765dd13 }, 0x10f418888ULL },
{ 0x35ef14ceULL, { 0x28ade3c7, 0x304532ae }, 0x2d73c33bULL },
{ 0xd03dbfbe24ULL, { 0x790079, 0xe419f3 }, 0x6e776fc2c4ULL },
{ 0x21d67410bULL, { 0x302a9dc2, 0x3ddb4442 }, 0x1a5948fd4ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x8664aeadULL, { 0x3d523513, 0x403523b1 }, 0x805a04cfULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x201db0cf4ULL, { 0x35112a7b, 0x79fc0c74 }, 0xdf8b07f8ULL },
{ 0x13f1e4430aULL, { 0x21c92bf, 0x21e63aae }, 0x13e0cba26ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x16c83229ULL, { 0x3793f66f, 0x53180dea }, 0xf3ce7b6ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0xc62415be8ULL, { 0x9cc4a63, 0x4327ae9b }, 0x1ce8b71c1ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x6fac5e434ULL, { 0xe5f9170, 0x1115e10b }, 0x5df23dd4cULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x1929375f2ULL, { 0x3a851375, 0x76c08456 }, 0xc662b083ULL },
{ 0x243c89db6ULL, { 0x354ebfc0, 0x450ef197 }, 0x1bf8c1663ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x310e9b31aULL, { 0x1b1b8acf, 0x2d3629f0 }, 0x1d69c93f9ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0xa1fae921dULL, { 0xa7a098c, 0x10469f44 }, 0x684413d6eULL },
{ 0xc1582d957ULL, { 0x498e061, 0x59856bc }, 0x9edc5f4ecULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x57cfee75ULL, { 0x1d061dc3, 0x7c8bfc17 }, 0x1476a220ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x139220080ULL, { 0x294a6c71, 0x2a2b07c9 }, 0x1329e1c75ULL },
{ 0x1665d353cULL, { 0x7080db5, 0xde0d75c }, 0xb590d9faULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0xe8f14541ULL, { 0x5188e8b2, 0x736527ef }, 0xa4971be5ULL },
{ 0x2f4775f29ULL, { 0x254ef0fe, 0x435fcf50 }, 0x1a2e449c1ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x27b85d8d7ULL, { 0x304c8220, 0x5de678f2 }, 0x146e3befbULL },
{ 0x1d362e36bULL, { 0x36c85b12, 0x37a66f55 }, 0x1cc19b8e7ULL },
{ 0x155fd48c7ULL, { 0xf5894d, 0x1256108 }, 0x11e383604ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0xb5db2d15ULL, { 0x39bb26c5, 0x5bdcda3e }, 0x72499259ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x153990298ULL, { 0x48921c09, 0x706eb817 }, 0xdb3268e7ULL },
{ 0x28a7c3ed7ULL, { 0x1f776fd7, 0x349f7a70 }, 0x184f73ae2ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x724dbeabULL, { 0x1bd149f5, 0x253a085e }, 0x5569c0b3ULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0xd8f0c513ULL, { 0x18c8cc4c, 0x1b72bad0 }, 0xc3e30642ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x17ce3dcbULL, { 0x1e4c6260, 0x233b359e }, 0x1478f4afULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x1ce036ce0ULL, { 0x29e3c8af, 0x5318dd4a }, 0xe8e76195ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x1473ae2aULL, { 0x29b897ba, 0x2be29378 }, 0x13718185ULL },
{ 0x1dd41aa68ULL, { 0x3d0a4441, 0x5a0e8f12 }, 0x1437b6bbfULL },
{ 0x1b49e4a53ULL, { 0x3430c1fe, 0x5a204aed }, 0xfcd6852fULL },
{ 0x217941b19ULL, { 0x12ced2bd, 0x21b68310 }, 0x12aca65b1ULL },
{ 0xac6a4dc8ULL, { 0x3ed68da8, 0x6fdca34c }, 0x60da926dULL },
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x1c503a4e7ULL, { 0xfcbbd32, 0x11e48d17 }, 0x18fec7d37ULL },
{ 0x1c885855ULL, { 0x213e919d, 0x25941897 }, 0x193de742ULL },
{ 0x29b9c168eULL, { 0x2b644aea, 0x45725ee7 }, 0x1a122e5d4ULL },
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
{ 0x806a33f2ULL, { 0x30a80a23, 0x5063733a }, 0x4db9a264ULL },
{ 0x282afc96bULL, { 0x143ae554, 0x1a9863ff }, 0x1e8de5204ULL },
// Data for scaling that results in > 64 bit division.
Use fixed-point representation for BranchProbability. BranchProbability now is represented by its numerator and denominator in uint32_t type. This patch changes this representation into a fixed point that is represented by the numerator in uint32_t type and a constant denominator 1<<31. This is quite similar to the representation of BlockMass in BlockFrequencyInfoImpl.h. There are several pros and cons of this change: Pros: 1. It uses only a half space of the current one. 2. Some operations are much faster like plus, subtraction, comparison, and scaling by an integer. Cons: 1. Constructing a probability using arbitrary numerator and denominator needs additional calculations. 2. It is a little less precise than before as we use a fixed denominator. For example, 1 - 1/3 may not be exactly identical to 1 / 3 (this will lead to many BranchProbability unit test failures). This should not matter when we only use it for branch probability. If we use it like a rational value for some precise calculations we may need another construct like ValueRatio. One important reason for this change is that we propose to store branch probabilities instead of edge weights in MachineBasicBlock. We also want clients to use probability instead of weight when adding successors to a MBB. The current BranchProbability has more space which may be a concern. Differential revision: http://reviews.llvm.org/D12603 llvm-svn: 248633
2015-09-26 07:09:59 +08:00
{ 0x23ca5f2f672ca41cULL, { 0xecbc641, 0x111373f7 }, 0x1f0301e5c76869c6ULL },
{ 0x5e4f2468142265e3ULL, { 0x1ddf5837, 0x32189233 }, 0x383ca7bad6053ac9ULL },
{ 0x277a1a6f6b266bf6ULL, { 0x415d81a8, 0x61eb5e1e }, 0x1a5a3e1d1c9e8540ULL },
{ 0x1bdbb49a237035cbULL, { 0xea5bf17, 0x1d25ffb3 }, 0xdffc51c5cb51cf1ULL },
{ 0x2bce6d29b64fb8ULL, { 0x3bfd5631, 0x7525c9bb }, 0x166ebedd9581fdULL },
{ 0x3a02116103df5013ULL, { 0x2ee18a83, 0x3299aea8 }, 0x35be89227276f105ULL },
{ 0x7b5762390799b18cULL, { 0x12f8e5b9, 0x2563bcd4 }, 0x3e960077695655a3ULL },
{ 0x69cfd72537021579ULL, { 0x4c35f468, 0x6a40feee }, 0x4be4cb38695a4f30ULL },
{ 0x49dfdf835120f1c1ULL, { 0x8cb3759, 0x559eb891 }, 0x79663f6e3c8d8f6ULL },
{ 0x74b5be5c27676381ULL, { 0x47e4c5e0, 0x7c7b19ff }, 0x4367d2dfb22b3265ULL },
{ 0x4f50f97075e7f431ULL, { 0x9a50a17, 0x11cd1185 }, 0x2af952b30374f382ULL },
{ 0x2f8b0d712e393be4ULL, { 0x1487e386, 0x15aa356e }, 0x2d0df3649b2b19fcULL },
{ 0x224c1c75999d3deULL, { 0x3b2df0ea, 0x4523b100 }, 0x1d5b481d160dd8bULL },
{ 0x2bcbcea22a399a76ULL, { 0x28b58212, 0x48dd013e }, 0x187814d0610c8a56ULL },
{ 0x1dbfca91257cb2d1ULL, { 0x1a8c04d9, 0x5e92502c }, 0x859cf7d19e83ad0ULL },
{ 0x7f20039b57cda935ULL, { 0xeccf651, 0x323f476e }, 0x25720cd9054634bdULL },
{ 0x40512c6a586aa087ULL, { 0x113b0423, 0x398c9eab }, 0x1341c03dbb662054ULL },
{ 0x63d802693f050a11ULL, { 0xf50cdd6, 0xfce2a44 }, 0x60c0177b667a4feaULL },
{ 0x2d956b422838de77ULL, { 0xb2d345b, 0x1321e557 }, 0x1aa0ed16b094575cULL },
{ 0x5a1cdf0c1657bc91ULL, { 0x1d77bb0c, 0x1f991ff1 }, 0x54097ee9907290eaULL },
{ 0x3801b26d7e00176bULL, { 0xeed25da, 0x1a819d8b }, 0x1f89e96a616b9abeULL },
{ 0x37655e74338e1e45ULL, { 0x300e170a, 0x5a1595fe }, 0x1d8cfb55ff6a6dbcULL },
{ 0x7b38703f2a84e6ULL, { 0x66d9053, 0xc79b6b9 }, 0x3f7d4c91b9afb9ULL },
{ 0x2245063c0acb3215ULL, { 0x30ce2f5b, 0x610e7271 }, 0x113b916455fe2560ULL },
{ 0x6bc195877b7b8a7eULL, { 0x392004aa, 0x4a24e60c }, 0x530594fabfc81cc3ULL },
{ 0x40a3fde23c7b43dbULL, { 0x4e712195, 0x6553e56e }, 0x320a799bc205c78dULL },
{ 0x1d3dfc2866fbccbaULL, { 0x5075b517, 0x5fc42245 }, 0x18917f00745cb781ULL },
{ 0x19aeb14045a61121ULL, { 0x1bf6edec, 0x707e2f4b }, 0x6626672aa2ba10aULL },
{ 0x44ff90486c531e9fULL, { 0x66598a, 0x8a90dc }, 0x32f6f2b097001598ULL },
{ 0x3f3e7121092c5bcbULL, { 0x1c754df7, 0x5951a1b9 }, 0x14267f50d4971583ULL },
{ 0x60e2dafb7e50a67eULL, { 0x4d96c66e, 0x65bd878d }, 0x49e317155d75e883ULL },
{ 0x656286667e0e6e29ULL, { 0x9d971a2, 0xacda23b }, 0x5c6ee3159e1deac3ULL },
{ 0x1114e0974255d507ULL, { 0x1c693, 0x2d6ff }, 0xaae42e4be5f9f8dULL },
{ 0x508c8baf3a70ff5aULL, { 0x3b26b779, 0x6ad78745 }, 0x2c983876178ed5b1ULL },
{ 0x5b47bc666bf1f9cfULL, { 0x10a87ed6, 0x187d358a }, 0x3e1767153bea720aULL },
{ 0x50954e3744460395ULL, { 0x7a42263, 0xcdaa048 }, 0x2fe739f0944a023cULL },
{ 0x20020b406550dd8fULL, { 0x3318539, 0x42eead0 }, 0x186f326307c0d985ULL },
{ 0x5bcb0b872439ffd5ULL, { 0x6f61fb2, 0x9af7344 }, 0x41fa1e3c47f0f80dULL },
{ 0x7a670f365db87a53ULL, { 0x417e102, 0x3bb54c67 }, 0x8642a551d0f41b0ULL },
{ 0x1ef0db1e7bab1cd0ULL, { 0x2b60cf38, 0x4188f78f }, 0x147ae0d63fc0575aULL }
Support: BlockFrequencyTest => BranchProbabilityTest Move a detailed test of `BranchProbability::scale()` from `BlockFrequencyTest` over to `BranchProbabilityTest`. llvm-svn: 207552
2014-04-30 00:40:17 +08:00
};
for (const auto &T : Tests) {
EXPECT_EQ(T.Result, BP(T.Prob[0], T.Prob[1]).scale(T.Num));
}
}
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
TEST(BranchProbabilityTest, NormalizeProbabilities) {
[BranchProbability] Remove the restriction that known and unknown probabilities cannot coexist when being normalized. The current BranchProbability::normalizeProbabilities() forbids known and unknown probabilities to coexist in the list. This was once used to help capture probability exceptions but has caused some reported build failures (https://llvm.org/bugs/show_bug.cgi?id=25838). This patch removes this restriction by evenly distributing the complement of the sum of all known probabilities to unknown ones. We could still treat this as an abnormal behavior, but it is better to emit warnings in our future profile validator. Differential revision: http://reviews.llvm.org/D15548 llvm-svn: 255934
2015-12-18 06:27:07 +08:00
const auto UnknownProb = BranchProbability::getUnknown();
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
{
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
SmallVector<BranchProbability, 2> Probs{{0, 1}, {0, 1}};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[1].getNumerator());
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
{
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
SmallVector<BranchProbability, 2> Probs{{0, 1}, {1, 1}};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(0u, Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator(), Probs[1].getNumerator());
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
{
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
SmallVector<BranchProbability, 2> Probs{{1, 100}, {1, 100}};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[1].getNumerator());
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
{
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
SmallVector<BranchProbability, 2> Probs{{1, 1}, {1, 1}};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[1].getNumerator());
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
{
Replace the unit test of BranchProbability::normalizeEdgeWeights() with BranchProbability::normalizeProbabilities(). BranchProbability::normalizeEdgeWeights() is going to be retired soon. llvm-svn: 255618
2015-12-15 13:25:27 +08:00
SmallVector<BranchProbability, 3> Probs{{1, 1}, {1, 1}, {1, 1}};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[1].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[2].getNumerator());
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
[BranchProbability] Remove the restriction that known and unknown probabilities cannot coexist when being normalized. The current BranchProbability::normalizeProbabilities() forbids known and unknown probabilities to coexist in the list. This was once used to help capture probability exceptions but has caused some reported build failures (https://llvm.org/bugs/show_bug.cgi?id=25838). This patch removes this restriction by evenly distributing the complement of the sum of all known probabilities to unknown ones. We could still treat this as an abnormal behavior, but it is better to emit warnings in our future profile validator. Differential revision: http://reviews.llvm.org/D15548 llvm-svn: 255934
2015-12-18 06:27:07 +08:00
{
SmallVector<BranchProbability, 2> Probs{{0, 1}, UnknownProb};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
BranchProbabilityTest.cpp: Suppress warnings. [-Wsign-compare] llvm-svn: 255940
2015-12-18 08:18:18 +08:00
EXPECT_EQ(0U, Probs[0].getNumerator());
[BranchProbability] Remove the restriction that known and unknown probabilities cannot coexist when being normalized. The current BranchProbability::normalizeProbabilities() forbids known and unknown probabilities to coexist in the list. This was once used to help capture probability exceptions but has caused some reported build failures (https://llvm.org/bugs/show_bug.cgi?id=25838). This patch removes this restriction by evenly distributing the complement of the sum of all known probabilities to unknown ones. We could still treat this as an abnormal behavior, but it is better to emit warnings in our future profile validator. Differential revision: http://reviews.llvm.org/D15548 llvm-svn: 255934
2015-12-18 06:27:07 +08:00
EXPECT_EQ(BranchProbability::getDenominator(), Probs[1].getNumerator());
}
{
SmallVector<BranchProbability, 2> Probs{{1, 1}, UnknownProb};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator(), Probs[0].getNumerator());
BranchProbabilityTest.cpp: Suppress warnings. [-Wsign-compare] llvm-svn: 255940
2015-12-18 08:18:18 +08:00
EXPECT_EQ(0U, Probs[1].getNumerator());
[BranchProbability] Remove the restriction that known and unknown probabilities cannot coexist when being normalized. The current BranchProbability::normalizeProbabilities() forbids known and unknown probabilities to coexist in the list. This was once used to help capture probability exceptions but has caused some reported build failures (https://llvm.org/bugs/show_bug.cgi?id=25838). This patch removes this restriction by evenly distributing the complement of the sum of all known probabilities to unknown ones. We could still treat this as an abnormal behavior, but it is better to emit warnings in our future profile validator. Differential revision: http://reviews.llvm.org/D15548 llvm-svn: 255934
2015-12-18 06:27:07 +08:00
}
{
SmallVector<BranchProbability, 2> Probs{{1, 2}, UnknownProb};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 2, Probs[1].getNumerator());
}
{
SmallVector<BranchProbability, 4> Probs{
{1, 2}, {1, 2}, {1, 2}, UnknownProb};
BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[0].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[1].getNumerator());
EXPECT_EQ(BranchProbability::getDenominator() / 3 + 1,
Probs[2].getNumerator());
BranchProbabilityTest.cpp: Suppress warnings. [-Wsign-compare] llvm-svn: 255940
2015-12-18 08:18:18 +08:00
EXPECT_EQ(0U, Probs[3].getNumerator());
[BranchProbability] Remove the restriction that known and unknown probabilities cannot coexist when being normalized. The current BranchProbability::normalizeProbabilities() forbids known and unknown probabilities to coexist in the list. This was once used to help capture probability exceptions but has caused some reported build failures (https://llvm.org/bugs/show_bug.cgi?id=25838). This patch removes this restriction by evenly distributing the complement of the sum of all known probabilities to unknown ones. We could still treat this as an abnormal behavior, but it is better to emit warnings in our future profile validator. Differential revision: http://reviews.llvm.org/D15548 llvm-svn: 255934
2015-12-18 06:27:07 +08:00
}
Recommit r250345, it was reverted in r250366 to investigate a bot failure. Our internal bot is still red after r250366. llvm-svn: 250415
2015-10-15 22:59:40 +08:00
}
Support: Add unit tests for BranchProbability llvm-svn: 207540
2014-04-30 00:12:13 +08:00
}