LLVM › LLVM - Dev

[llvm-dev] InstrEmitter::CreateVirtualRegisters handling of CopyToReg

‹ Previous Topic Next Topic ›
 
classic Classic list List threaded Threaded
♦
♦
 3 messages Options
Tim Northover via llvm-dev
Reply | Threaded
Open this post in threaded view
♦
♦
|

[llvm-dev] InstrEmitter::CreateVirtualRegisters handling of CopyToReg

Tim Northover via llvm-dev
Hi,

I wonder if anyone has any comment on a patch like:

diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 65ee3816f84..4780f6f0e59 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -243,18 +243,21 @@ void InstrEmitter::CreateVirtualRegisters(SDNode
*Node,

      if (!VRBase && !IsClone && !IsCloned)
        for (SDNode *User : Node->uses()) {
          if (User->getOpcode() == ISD::CopyToReg &&
              User->getOperand(2).getNode() == Node &&
              User->getOperand(2).getResNo() == i) {
            unsigned Reg =
cast<RegisterSDNode>(User->getOperand(1))->getReg();
            if (TargetRegisterInfo::isVirtualRegister(Reg)) {
-            const TargetRegisterClass *RegRC = MRI->getRegClass(Reg);
-            if (RegRC == RC) {
+            // Allow constraining the virtual register's class within
reason,
+            // just like what AddRegisterOperand will allow.
+            const TargetRegisterClass *ConstrainedRC
+              = MRI->constrainRegClass(Reg, RC, MinRCSize);
+            if (ConstrainedRC) {
                VRBase = Reg;
                MIB.addReg(VRBase, RegState::Define);
                break;
              }
            }
          }
        }

Why do the register classes currently have to match exactly in this case?

It seems that these COPYs that now remain may end up in the same
register class, if the users require it. So why not constrain also here
directly, if this is done generally when the register is used as input?

/Jonas


_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Tim Northover via llvm-dev
Reply | Threaded
Open this post in threaded view
♦
♦
|

[llvm-dev] Fwd: InstrEmitter::CreateVirtualRegisters handling of CopyToReg

Tim Northover via llvm-dev

Hi,

It would still be nice to get some comment about this patch which seems to help on SystemZ in the case where GRX32 pseudo instructions are used. This means that the register eventually ends up either in GR32 or GRH32 (low or high 32 bit subregs).

Basically, I wonder if anyone has tried this before and has any argument against this?

Thanks,

Jonas


-------- Forwarded Message --------
Subject: InstrEmitter::CreateVirtualRegisters handling of CopyToReg
Date: Wed, 30 May 2018 15:02:57 +0200
From: Jonas Paulsson [hidden email]
To: llvm-dev [hidden email]
CC: Ulrich Weigand [hidden email] 


Hi,

I wonder if anyone has any comment on a patch like:

diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp 
b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 65ee3816f84..4780f6f0e59 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -243,18 +243,21 @@ void InstrEmitter::CreateVirtualRegisters(SDNode 
*Node,

     if (!VRBase && !IsClone && !IsCloned)
       for (SDNode *User : Node->uses()) {
         if (User->getOpcode() == ISD::CopyToReg &&
             User->getOperand(2).getNode() == Node &&
             User->getOperand(2).getResNo() == i) {
           unsigned Reg = 
cast<RegisterSDNode>(User->getOperand(1))->getReg();
           if (TargetRegisterInfo::isVirtualRegister(Reg)) {
-            const TargetRegisterClass *RegRC = MRI->getRegClass(Reg);
-            if (RegRC == RC) {
+            // Allow constraining the virtual register's class within 
reason,
+            // just like what AddRegisterOperand will allow.
+            const TargetRegisterClass *ConstrainedRC
+              = MRI->constrainRegClass(Reg, RC, MinRCSize);
+            if (ConstrainedRC) {
               VRBase = Reg;
               MIB.addReg(VRBase, RegState::Define);
               break;
             }
           }
         }
       }

Why do the register classes currently have to match exactly in this case?

It seems that these COPYs that now remain may end up in the same 
register class, if the users require it. So why not constrain also here 
directly, if this is done generally when the register is used as input?

/Jonas

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Tim Northover via llvm-dev
Reply | Threaded
Open this post in threaded view
♦
♦
|

Re: [llvm-dev] Fwd: InstrEmitter::CreateVirtualRegisters handling of CopyToReg

Tim Northover via llvm-dev
I haven't had the occasion to touch this particular part of InstrEmitter, but the patch looks reasonable.

-Eli

On 6/8/2018 11:37 PM, Jonas Paulsson wrote:

Hi,

It would still be nice to get some comment about this patch which seems to help on SystemZ in the case where GRX32 pseudo instructions are used. This means that the register eventually ends up either in GR32 or GRH32 (low or high 32 bit subregs).

Basically, I wonder if anyone has tried this before and has any argument against this?

Thanks,

Jonas


-------- Forwarded Message --------
Subject: InstrEmitter::CreateVirtualRegisters handling of CopyToReg
Date: Wed, 30 May 2018 15:02:57 +0200
From: Jonas Paulsson [hidden email]
To: llvm-dev [hidden email]
CC: Ulrich Weigand [hidden email] 


Hi,

I wonder if anyone has any comment on a patch like:

diff --git a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp 
b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 65ee3816f84..4780f6f0e59 100644
--- a/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -243,18 +243,21 @@ void InstrEmitter::CreateVirtualRegisters(SDNode 
*Node,

     if (!VRBase && !IsClone && !IsCloned)
       for (SDNode *User : Node->uses()) {
         if (User->getOpcode() == ISD::CopyToReg &&
             User->getOperand(2).getNode() == Node &&
             User->getOperand(2).getResNo() == i) {
           unsigned Reg = 
cast<RegisterSDNode>(User->getOperand(1))->getReg();
           if (TargetRegisterInfo::isVirtualRegister(Reg)) {
-            const TargetRegisterClass *RegRC = MRI->getRegClass(Reg);
-            if (RegRC == RC) {
+            // Allow constraining the virtual register's class within 
reason,
+            // just like what AddRegisterOperand will allow.
+            const TargetRegisterClass *ConstrainedRC
+              = MRI->constrainRegClass(Reg, RC, MinRCSize);
+            if (ConstrainedRC) {
               VRBase = Reg;
               MIB.addReg(VRBase, RegState::Define);
               break;
             }
           }
         }
       }

Why do the register classes currently have to match exactly in this case?

It seems that these COPYs that now remain may end up in the same 
register class, if the users require it. So why not constrain also here 
directly, if this is done generally when the register is used as input?

/Jonas





-- 
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project

_______________________________________________
LLVM Developers mailing list
[hidden email]
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
Free forum by Nabble Edit this page