[Hexagon] Minimize number of repeated constant extenders

Each constant extender requires an extra instruction, which adds to the
code size and also reduces the number of available slots in an instruction
packet. In most cases, the value of a repeated constant extender could be
loaded into a register, and the instructions using the extender could be
replaced with their counterparts that use that register instead.

This patch adds a pass that tries to reduce the number of constant
extenders, including extenders which differ only in an immediate offset
known at compile time, e.g. @global and @global+12.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@315735 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/Hexagon/CMakeLists.txt

@@ -20,6 +20,7 @@ add_llvm_target(HexagonCodeGen
   HexagonBranchRelaxation.cpp
   HexagonCFGOptimizer.cpp
   HexagonCommonGEP.cpp
+  HexagonConstExtenders.cpp
   HexagonConstPropagation.cpp
   HexagonCopyToCombine.cpp
   HexagonEarlyIfConv.cpp

lib/Target/Hexagon/HexagonTargetMachine.cpp

@@ -28,6 +28,9 @@
 
 using namespace llvm;
 
+static cl::opt<bool> EnableCExtOpt("hexagon-cext", cl::Hidden, cl::ZeroOrMore,
+  cl::init(true), cl::desc("Enable Hexagon constant-extender optimization"));
+
 static cl::opt<bool> EnableRDFOpt("rdf-opt", cl::Hidden, cl::ZeroOrMore,
   cl::init(true), cl::desc("Enable RDF-based optimizations"));
 
@@ -119,6 +122,7 @@ SchedCustomRegistry("hexagon", "Run Hexagon's custom scheduler",
 
 namespace llvm {
   extern char &HexagonExpandCondsetsID;
+  void initializeHexagonConstExtendersPass(PassRegistry&);
   void initializeHexagonEarlyIfConversionPass(PassRegistry&);
   void initializeHexagonExpandCondsetsPass(PassRegistry&);
   void initializeHexagonGenMuxPass(PassRegistry&);
@@ -135,6 +139,7 @@ namespace llvm {
   FunctionPass *createHexagonCallFrameInformation();
   FunctionPass *createHexagonCFGOptimizer();
   FunctionPass *createHexagonCommonGEP();
+  FunctionPass *createHexagonConstExtenders();
   FunctionPass *createHexagonConstPropagationPass();
   FunctionPass *createHexagonCopyToCombine();
   FunctionPass *createHexagonEarlyIfConversion();
@@ -176,6 +181,7 @@ extern "C" void LLVMInitializeHexagonTarget() {
   RegisterTargetMachine<HexagonTargetMachine> X(getTheHexagonTarget());
 
   PassRegistry &PR = *PassRegistry::getPassRegistry();
+  initializeHexagonConstExtendersPass(PR);
   initializeHexagonEarlyIfConversionPass(PR);
   initializeHexagonGenMuxPass(PR);
   initializeHexagonLoopIdiomRecognizePass(PR);
@@ -340,6 +346,8 @@ bool HexagonPassConfig::addInstSelector() {
 
 void HexagonPassConfig::addPreRegAlloc() {
   if (getOptLevel() != CodeGenOpt::None) {
+    if (EnableCExtOpt)
+      addPass(createHexagonConstExtenders());
     if (EnableExpandCondsets)
       insertPass(&RegisterCoalescerID, &HexagonExpandCondsetsID);
     if (!DisableStoreWidening)

test/CodeGen/Hexagon/cext-opt-basic.mir

@@ -0,0 +1,75 @@
+# RUN: llc -march=hexagon -run-pass hexagon-cext-opt -hexagon-cext-threshold=3 %s -o - | FileCheck %s
+
+--- |
+  define void @test0() { ret void }
+  define void @test1() { ret void }
+  define void @test2() { ret void }
+  @global_address = global [1024 x i32] zeroinitializer, align 8
+...
+
+# CHECK-LABEL: name: test0
+# CHECK: [[B:%[0-9]+]] = A2_tfrsi @global_address
+# CHECK: L2_loadri_io [[B]], 0
+# CHECK: L2_loadri_io [[B]], 4
+# CHECK: L2_loadri_io [[B]], 8
+---
+name: test0
+registers:
+  - { id: 0, class: intregs }
+  - { id: 1, class: intregs }
+  - { id: 2, class: intregs }
+body: |
+  bb.0:
+    %0 = PS_loadriabs @global_address
+    %1 = PS_loadriabs @global_address+4
+    %2 = PS_loadriabs @global_address+8
+...
+
+# CHECK-LABEL: name: test1
+# CHECK: [[C:%[0-9]+]] = COPY %r0
+# CHECK: [[B:%[0-9]+]] = A2_addi [[C]], @global_address
+# CHECK: L2_loadri_io [[B]], 0
+# CHECK: L2_loadri_io [[B]], 4
+# CHECK: L2_loadri_io [[B]], 8
+---
+name: test1
+registers:
+  - { id: 0, class: intregs }
+  - { id: 1, class: intregs }
+  - { id: 2, class: intregs }
+  - { id: 3, class: intregs }
+body: |
+  bb.0:
+    liveins: %r0
+    %0 = COPY %r0
+    %1 = L4_loadri_ur %0, 0, @global_address
+    %2 = L4_loadri_ur %0, 0, @global_address+4
+    %3 = L4_loadri_ur %0, 0, @global_address+8
+...
+
+# CHECK-LABEL: name: test2
+# CHECK: [[C:%[0-9]+]] = COPY %r0
+# CHECK: [[B:%[0-9]+]] = A2_tfrsi @global_address + 4
+# CHECK: [[T0:%[0-9]+]] = A2_addi [[B]], -4
+# CHECK: %r0 = COPY [[T0]]
+# CHECK: [[T1:%[0-9]+]] = A2_addi [[B]], -2
+# CHECK: %r1 = COPY [[T1]]
+# CHECK: L4_loadri_rr [[B]], [[C]], 0
+---
+name: test2
+registers:
+  - { id: 0, class: intregs }
+  - { id: 1, class: intregs }
+  - { id: 2, class: intregs }
+  - { id: 3, class: intregs }
+body: |
+  bb.0:
+    liveins: %r0
+    %0 = COPY %r0
+    %1 = A2_tfrsi @global_address
+    %r0 = COPY %1
+    %2 = A2_tfrsi @global_address+2
+    %r1 = COPY %2
+    %3 = L4_loadri_ur %0, 0, @global_address+4
+...
+

test/CodeGen/Hexagon/zextloadi1.ll

@@ -1,4 +1,4 @@
-; RUN: llc -march=hexagon < %s | FileCheck %s
+; RUN: llc -march=hexagon -hexagon-cext=0 < %s | FileCheck %s
 
 @i65_l = external global i65
 @i65_s = external global i65