[WebAssembly] Fix scheduling dependencies in register-stackified code

Add physical register defs to instructions used from stackified
instructions to prevent them from being scheduled into the middle of
a stack sequence. This is a conservative measure which may be loosened
in the future.

Differential Revision: http://reviews.llvm.org/D15252

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@254811 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp b/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
index 7abc20a8387e9981f13ce8e46b66b360cda0a6a9..ac016a7b9b0aaa23fdf86b10195418a849076e3c 100644 (file)
--- a/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
+++ b/lib/Target/WebAssembly/WebAssemblyRegStackify.cpp
@@ -61,15 +61,41 @@ FunctionPass *llvm::createWebAssemblyRegStackify() {
 }
 
 // Decorate the given instruction with implicit operands that enforce the
-// expression stack ordering constraints.
-static void ImposeStackOrdering(MachineInstr *MI) {
-  // Read and write the opaque EXPR_STACK register.
-  MI->addOperand(MachineOperand::CreateReg(WebAssembly::EXPR_STACK,
-                                           /*isDef=*/true,
-                                           /*isImp=*/true));
+// expression stack ordering constraints needed for an instruction which is
+// consumed by an instruction using the expression stack.
+static void ImposeStackInputOrdering(MachineInstr *MI) {
+  // Write the opaque EXPR_STACK register.
+  if (!MI->definesRegister(WebAssembly::EXPR_STACK))
+    MI->addOperand(MachineOperand::CreateReg(WebAssembly::EXPR_STACK,
+                                             /*isDef=*/true,
+                                             /*isImp=*/true));
+}
+
+// Decorate the given instruction with implicit operands that enforce the
+// expression stack ordering constraints for an instruction which is on
+// the expression stack.
+static void ImposeStackOrdering(MachineInstr *MI, MachineRegisterInfo &MRI) {
+  ImposeStackInputOrdering(MI);
+
+  // Also read the opaque EXPR_STACK register.
   MI->addOperand(MachineOperand::CreateReg(WebAssembly::EXPR_STACK,
                                            /*isDef=*/false,
                                            /*isImp=*/true));
+
+  // Also, mark any inputs to this instruction as being consumed by an
+  // instruction on the expression stack.
+  // TODO: Find a lighter way to describe the appropriate constraints.
+  for (MachineOperand &MO : MI->uses()) {
+    if (!MO.isReg())
+      continue;
+    unsigned Reg = MO.getReg();
+    if (!TargetRegisterInfo::isVirtualRegister(Reg))
+      continue;
+    MachineInstr *Def = MRI.getVRegDef(Reg);
+    if (Def->getOpcode() == TargetOpcode::PHI)
+      continue;
+    ImposeStackInputOrdering(Def);
+  }
 }
 
 // Test whether it's safe to move Def to just before Insert. Note that this
@@ -126,8 +152,15 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
           continue;
 
         unsigned Reg = Op.getReg();
-        if (!TargetRegisterInfo::isVirtualRegister(Reg))
+        if (TargetRegisterInfo::isPhysicalRegister(Reg)) {
+          // An instruction with a physical register. Conservatively mark it as
+          // an expression stack input so that it isn't reordered with anything
+          // in an expression stack which might use it (physical registers
+          // aren't in SSA form so it's not trivial to determine this).
+          // TODO: Be less conservative.
+          ImposeStackInputOrdering(Insert);
           continue;
+        }
 
         // Only consider registers with a single definition.
         // TODO: Eventually we may relax this, to stackify phi transfers.
@@ -178,11 +211,11 @@ bool WebAssemblyRegStackify::runOnMachineFunction(MachineFunction &MF) {
         MBB.insert(MachineBasicBlock::instr_iterator(Insert),
                    Def->removeFromParent());
         MFI.stackifyVReg(Reg);
-        ImposeStackOrdering(Def);
+        ImposeStackOrdering(Def, MRI);
         Insert = Def;
       }
       if (AnyStackified)
-        ImposeStackOrdering(&MI);
+        ImposeStackOrdering(&MI, MRI);
     }
   }
 

diff --git a/test/CodeGen/WebAssembly/load-store-i1.ll b/test/CodeGen/WebAssembly/load-store-i1.ll
index 1acdfc0dbdeb5f66a6bcd705f88d30160cc21a46..37b5147294793c50faf40434901dfec16e4363fb 100644 (file)
--- a/test/CodeGen/WebAssembly/load-store-i1.ll
+++ b/test/CodeGen/WebAssembly/load-store-i1.ll
@@ -15,8 +15,8 @@ define i32 @load_u_i1_i32(i1* %p) {
 }
 
 ; CHECK-LABEL: load_s_i1_i32:
-; CHECK:      i32.load8_u $push[[NUM0:[0-9]+]]=, 0($0){{$}}
-; CHECK-NEXT: i32.const $[[NUM1:[0-9]+]]=, 31{{$}}
+; CHECK:      i32.const $[[NUM1:[0-9]+]]=, 31{{$}}
+; CHECK-NEXT: i32.load8_u $push[[NUM0:[0-9]+]]=, 0($0){{$}}
 ; CHECK-NEXT: shl $push[[NUM2:[0-9]+]]=, $pop[[NUM0]], $[[NUM1]]{{$}}
 ; CHECK-NEXT: shr_s $push[[NUM3:[0-9]+]]=, $pop[[NUM2]], $[[NUM1]]{{$}}
 ; CHECK-NEXT: return $pop[[NUM3]]{{$}}
@@ -36,8 +36,8 @@ define i64 @load_u_i1_i64(i1* %p) {
 }
 
 ; CHECK-LABEL: load_s_i1_i64:
-; CHECK:      i64.load8_u $push[[NUM0:[0-9]+]]=, 0($0){{$}}
-; CHECK-NEXT: i64.const $[[NUM1:[0-9]+]]=, 63{{$}}
+; CHECK:      i64.const $[[NUM1:[0-9]+]]=, 63{{$}}
+; CHECK-NEXT: i64.load8_u $push[[NUM0:[0-9]+]]=, 0($0){{$}}
 ; CHECK-NEXT: shl $push[[NUM2:[0-9]+]]=, $pop[[NUM0]], $[[NUM1]]{{$}}
 ; CHECK-NEXT: shr_s $push[[NUM3:[0-9]+]]=, $pop[[NUM2]], $[[NUM1]]{{$}}
 ; CHECK-NEXT: return $pop[[NUM3]]{{$}}

diff --git a/test/CodeGen/WebAssembly/reg-stackify.ll b/test/CodeGen/WebAssembly/reg-stackify.ll
index f3000aab70ad6d8a550c49c30973bc8001ccb0d6..180d70e2e4a7071cae3126432bc8bf9a11041e9c 100644 (file)
--- a/test/CodeGen/WebAssembly/reg-stackify.ll
+++ b/test/CodeGen/WebAssembly/reg-stackify.ll
@@ -44,4 +44,32 @@ define i32 @yes1(i32* %q) {
   ret i32 %t
 }
 
+; Don't schedule stack uses into the stack. To reduce register pressure, the
+; scheduler might be tempted to move the definition of $2 down. However, this
+; would risk getting incorrect liveness if the instructions are later
+; rearranged to make the stack contiguous.
+
+; CHECK-LABEL: stack_uses:
+; CHECK-NEXT: .param i32{{$}}
+; CHECK-NEXT: .result i32{{$}}
+; CHECK-NEXT: local i32, i32{{$}}
+; CHECK-NEXT: i32.const       $1=, 1{{$}}
+; CHECK-NEXT: i32.const       $2=, 0{{$}}
+; CHECK-NEXT: i32.and         $push0=, $0, $1{{$}}
+; CHECK-NEXT: i32.eq          $push1=, $pop0, $2{{$}}
+; CHECK-NEXT: block           BB4_2{{$}}
+; CHECK-NEXT: br_if           $pop1, BB4_2{{$}}
+; CHECK-NEXT: return          $2{{$}}
+; CHECK-NEXT: BB4_2:{{$}}
+; CHECK-NEXT: return          $1{{$}}
+define i32 @stack_uses(i32 %x) {
+entry:
+  %c = trunc i32 %x to i1
+  br i1 %c, label %true, label %false
+true:
+  ret i32 0
+false:
+  ret i32 1
+}
+
 !0 = !{}