R600/SI: Fix broken encoding of DS_WRITE_B32

The logic in SIInsertWaits::getHwCounts() only really made sense for SMRD
instructions, and trying to shoehorn it into handling DS_WRITE_B32 caused
it to corrupt the encoding of that by clobbering the first operand with
the second one.

Undo that damage and only apply the SMRD logic to that.

Fixes some derivates related piglit regressions with radeonsi.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188558 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/SIDefines.h b/lib/Target/R600/SIDefines.h
index 572ed6a..f5445ad 100644 (file)
--- a/lib/Target/R600/SIDefines.h
+++ b/lib/Target/R600/SIDefines.h
@@ -13,7 +13,8 @@
 
 namespace SIInstrFlags {
 enum {
-  MIMG = 1 << 3
+  MIMG = 1 << 3,
+  SMRD = 1 << 4
 };
 }
 

diff --git a/lib/Target/R600/SIInsertWaits.cpp b/lib/Target/R600/SIInsertWaits.cpp
index ba202e3..7e42fb7 100644 (file)
--- a/lib/Target/R600/SIInsertWaits.cpp
+++ b/lib/Target/R600/SIInsertWaits.cpp
@@ -134,14 +134,19 @@ Counters SIInsertWaits::getHwCounts(MachineInstr &MI) {
   // LGKM may uses larger values
   if (TSFlags & SIInstrFlags::LGKM_CNT) {
 
-    MachineOperand &Op = MI.getOperand(0);
-    if (!Op.isReg())
-      Op = MI.getOperand(1);
-    assert(Op.isReg() && "First LGKM operand must be a register!");
-
-    unsigned Reg = Op.getReg();
-    unsigned Size = TRI->getMinimalPhysRegClass(Reg)->getSize();
-    Result.Named.LGKM = Size > 4 ? 2 : 1;
+    if (TII->isSMRD(MI.getOpcode())) {
+
+      MachineOperand &Op = MI.getOperand(0);
+      assert(Op.isReg() && "First LGKM operand must be a register!");
+
+      unsigned Reg = Op.getReg();
+      unsigned Size = TRI->getMinimalPhysRegClass(Reg)->getSize();
+      Result.Named.LGKM = Size > 4 ? 2 : 1;
+
+    } else {
+      // DS
+      Result.Named.LGKM = 1;
+    }
 
   } else {
     Result.Named.LGKM = 0;

diff --git a/lib/Target/R600/SIInstrFormats.td b/lib/Target/R600/SIInstrFormats.td
index cd1bbcd..9576c05 100644 (file)
--- a/lib/Target/R600/SIInstrFormats.td
+++ b/lib/Target/R600/SIInstrFormats.td
@@ -18,11 +18,13 @@ class InstSI <dag outs, dag ins, string asm, list<dag> pattern> :
   field bits<1> EXP_CNT = 0;
   field bits<1> LGKM_CNT = 0;
   field bits<1> MIMG = 0;
+  field bits<1> SMRD = 0;
 
   let TSFlags{0} = VM_CNT;
   let TSFlags{1} = EXP_CNT;
   let TSFlags{2} = LGKM_CNT;
   let TSFlags{3} = MIMG;
+  let TSFlags{4} = SMRD;
 }
 
 class Enc32 <dag outs, dag ins, string asm, list<dag> pattern> :
@@ -142,6 +144,7 @@ class SMRD <bits<5> op, bits<1> imm, dag outs, dag ins, string asm,
   let Inst{31-27} = 0x18; //encoding
 
   let LGKM_CNT = 1;
+  let SMRD = 1;
 }
 
 //===----------------------------------------------------------------------===//

diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index 9bb4ad9..2719ea2 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -229,6 +229,10 @@ int SIInstrInfo::isMIMG(uint16_t Opcode) const {
   return get(Opcode).TSFlags & SIInstrFlags::MIMG;
 }
 
+int SIInstrInfo::isSMRD(uint16_t Opcode) const {
+  return get(Opcode).TSFlags & SIInstrFlags::SMRD;
+}
+
 //===----------------------------------------------------------------------===//
 // Indirect addressing callbacks
 //===----------------------------------------------------------------------===//

diff --git a/lib/Target/R600/SIInstrInfo.h b/lib/Target/R600/SIInstrInfo.h
index 8d24ab4..87b8063 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.h
+++ b/lib/Target/R600/SIInstrInfo.h
@@ -48,6 +48,7 @@ public:
 
   virtual bool isSafeToMoveRegClassDefs(const TargetRegisterClass *RC) const;
   int isMIMG(uint16_t Opcode) const;
+  int isSMRD(uint16_t Opcode) const;
 
   virtual int getIndirectIndexBegin(const MachineFunction &MF) const;
 

diff --git a/test/CodeGen/R600/local-memory.ll b/test/CodeGen/R600/local-memory.ll
index 5458fb9..9ebb769 100644 (file)
--- a/test/CodeGen/R600/local-memory.ll
+++ b/test/CodeGen/R600/local-memory.ll
@@ -13,7 +13,7 @@
 ; SI-CHECK-NEXT: .long 32768
 
 ; EG-CHECK: LDS_WRITE
-; SI-CHECK: DS_WRITE_B32
+; SI-CHECK: DS_WRITE_B32 0
 
 ; GROUP_BARRIER must be the last instruction in a clause
 ; EG-CHECK: GROUP_BARRIER
@@ -21,7 +21,7 @@
 ; SI-CHECK: S_BARRIER
 
 ; EG-CHECK: LDS_READ_RET
-; SI-CHECK: DS_READ_B32
+; SI-CHECK: DS_READ_B32 {{VGPR[0-9]+}}, 0
 
 define void @local_memory(i32 addrspace(1)* %out) {
 entry: