[llvm] [X86] Set up the framework for optimization of CCMP/CTEST (PR #84603)

[Evgenii Kudriashov via llvm-commits]

================
@@ -0,0 +1,887 @@
+//=========-- X86ConditionalCompares.cpp --- CCMP formation for X86 -------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the X86ConditionalCompares pass which reduces
+// branching by using the conditional compare instructions CCMP, CTEST.
+//
+// The CFG transformations for forming conditional compares are very similar to
+// if-conversion, and this pass should run immediately before the early
+// if-conversion pass.
+//
+//===----------------------------------------------------------------------===//
+
+#include "X86.h"
+#include "X86InstrInfo.h"
+#include "X86Subtarget.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/MachineTraceMetrics.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/TargetInstrInfo.h"
+#include "llvm/CodeGen/TargetRegisterInfo.h"
+#include "llvm/CodeGen/TargetSubtargetInfo.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "x86-ccmp"
+
+// Absolute maximum number of instructions allowed per speculated block.
+// This bypasses all other heuristics, so it should be set fairly high.
+static cl::opt<unsigned> BlockInstrLimit(
+    "x86-ccmp-limit", cl::init(30), cl::Hidden,
+    cl::desc("Maximum number of instructions per speculated block."));
+
+STATISTIC(NumConsidered, "Number of ccmps considered");
+STATISTIC(NumPhiRejs, "Number of ccmps rejected (PHI)");
+STATISTIC(NumPhysRejs, "Number of ccmps rejected (Physregs)");
+STATISTIC(NumPhi2Rejs, "Number of ccmps rejected (PHI2)");
+STATISTIC(NumHeadBranchRejs, "Number of ccmps rejected (Head branch)");
+STATISTIC(NumCmpBranchRejs, "Number of ccmps rejected (CmpBB branch)");
+STATISTIC(NumCmpTermRejs, "Number of ccmps rejected (CmpBB is cbz...)");
+STATISTIC(NumMultEFLAGSUses, "Number of ccmps rejected (EFLAGS used)");
+STATISTIC(NumUnknEFLAGSDefs, "Number of ccmps rejected (EFLAGS def unknown)");
+STATISTIC(NumSpeculateRejs, "Number of ccmps rejected (Can't speculate)");
+STATISTIC(NumConverted, "Number of ccmp instructions created");
+
+//===----------------------------------------------------------------------===//
+//                                 SSACCmpConv
+//===----------------------------------------------------------------------===//
+//
+// The SSACCmpConv class performs ccmp-conversion on SSA form machine code
+// after determining if it is possible. The class contains no heuristics;
+// external code should be used to determine when ccmp-conversion is a good
+// idea.
+//
+// CCmp-formation works on a CFG representing chained conditions, typically
+// from C's short-circuit || and && operators:
+//
+//   From:         Head            To:         Head
+//                 / |                         CmpBB
+//                /  |                         / |
+//               |  CmpBB                     /  |
+//               |  / |                    Tail  |
+//               | /  |                      |   |
+//              Tail  |                      |   |
+//                |   |                      |   |
+//               ... ...                    ... ...
+//
+// The Head block is terminated by a br.cond instruction, and the CmpBB block
+// contains compare + br.cond. Tail must be a successor of both.
+//
+// The cmp-conversion turns the compare instruction in CmpBB into a conditional
+// compare, and merges CmpBB into Head, speculatively executing its
+// instructions. The X86 conditional compare instructions have an operand that
+// specifies the conditional flags to set values when the condition is false and
+// the compare isn't executed. This makes it possible to chain compares with
+// different condition codes.
+//
+// Example:
+//
+// void f(int a, int b) {
+//   if (a == 5 || b == 17)
+//     foo();
+// }
+//
+//    Head:
+//      cmpl  $5, $edi
+//      je Tail
+//    CmpBB:
+//      cmpl  $17, $esi
+//      je Tail
+//    ...
+//    Tail:
+//      call foo
+//
+//  Becomes:
+//
+//    Head:
+//      cmpl  $5, $edi
+//      ccmpel {dfv=zf} $17, $edi
+//      je Tail
+//    ...
+//    Tail:
+//      call foo
+//
+// The ccmp condition code is the one that would cause the Head terminator to
----------------
e-kud wrote:

My bad. I've found tests cases where cross block combine happens.

https://github.com/llvm/llvm-project/pull/84603