AMDGPU: Fix handling of 16-bit immediates

Since 32-bit instructions with 32-bit input immediate behavior
are used to materialize 16-bit constants in 32-bit registers
for 16-bit instructions, determining the legality based
on the size is incorrect. Change operands to have the size
specified in the type.

Also adds a workaround for a disassembler bug that
produces an immediate MCOperand for an operand that
is supposed to be OPERAND_REGISTER.

The assembler appears to accept out of bounds immediates and
truncates them, but this seems to be an issue for 32-bit
already.

llvm-svn: 289306

2 files changed, 76 insertions, 11 deletions

diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index 29cac2fbf6d..85cbadf0a57 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -329,25 +329,29 @@ unsigned getMCReg(unsigned Reg, const MCSubtargetInfo &STI) {
 
 bool isSISrcOperand(const MCInstrDesc &Desc, unsigned OpNo) {
   unsigned OpType = Desc.OpInfo[OpNo].OperandType;
-
-  return OpType == AMDGPU::OPERAND_REG_IMM32_INT ||
-         OpType == AMDGPU::OPERAND_REG_IMM32_FP ||
-         OpType == AMDGPU::OPERAND_REG_INLINE_C_INT ||
-         OpType == AMDGPU::OPERAND_REG_INLINE_C_FP;
+  return OpType >= AMDGPU::OPERAND_SRC_FIRST &&
+         OpType <= AMDGPU::OPERAND_SRC_LAST;
 }
 
 bool isSISrcFPOperand(const MCInstrDesc &Desc, unsigned OpNo) {
   unsigned OpType = Desc.OpInfo[OpNo].OperandType;
-
-  return OpType == AMDGPU::OPERAND_REG_IMM32_FP ||
-         OpType == AMDGPU::OPERAND_REG_INLINE_C_FP;
+  switch (OpType) {
+  case AMDGPU::OPERAND_REG_IMM_FP32:
+  case AMDGPU::OPERAND_REG_IMM_FP64:
+  case AMDGPU::OPERAND_REG_IMM_FP16:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP32:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP64:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP16:
+    return true;
+  default:
+    return false;
+  }
 }
 
 bool isSISrcInlinableOperand(const MCInstrDesc &Desc, unsigned OpNo) {
   unsigned OpType = Desc.OpInfo[OpNo].OperandType;
-
-  return OpType == AMDGPU::OPERAND_REG_INLINE_C_INT ||
-         OpType == AMDGPU::OPERAND_REG_INLINE_C_FP;
+  return OpType >= AMDGPU::OPERAND_REG_INLINE_C_FIRST &&
+         OpType <= AMDGPU::OPERAND_REG_INLINE_C_LAST;
 }
 
 // Avoid using MCRegisterClass::getSize, since that function will go away
@@ -413,6 +417,15 @@ bool isInlinableLiteral32(int32_t Literal, bool HasInv2Pi) {
   if (Literal >= -16 && Literal <= 64)
     return true;
 
+  // The actual type of the operand does not seem to matter as long
+  // as the bits match one of the inline immediate values.  For example:
+  //
+  // -nan has the hexadecimal encoding of 0xfffffffe which is -2 in decimal,
+  // so it is a legal inline immediate.
+  //
+  // 1065353216 has the hexadecimal encoding 0x3f800000 which is 1.0f in
+  // floating-point, so it is a legal inline immediate.
+
   uint32_t Val = static_cast<uint32_t>(Literal);
   return (Val == FloatToBits(0.0f)) ||
          (Val == FloatToBits(1.0f)) ||
@@ -426,6 +439,23 @@ bool isInlinableLiteral32(int32_t Literal, bool HasInv2Pi) {
          (Val == 0x3e22f983 && HasInv2Pi);
 }
 
+bool isInlinableLiteral16(int16_t Literal, bool HasInv2Pi) {
+  assert(HasInv2Pi);
+
+  if (Literal >= -16 && Literal <= 64)
+    return true;
+
+  uint16_t Val = static_cast<uint16_t>(Literal);
+  return Val == 0x3C00 || // 1.0
+         Val == 0xBC00 || // -1.0
+         Val == 0x3800 || // 0.5
+         Val == 0xB800 || // -0.5
+         Val == 0x4000 || // 2.0
+         Val == 0xC000 || // -2.0
+         Val == 0x4400 || // 4.0
+         Val == 0xC400 || // -4.0
+         Val == 0x3118;   // 1/2pi
+}
 
 } // End namespace AMDGPU
 } // End namespace llvm
diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 3101b96c8eb..ea5fc366d20 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -13,6 +13,8 @@
 #include "AMDKernelCodeT.h"
 #include "llvm/IR/CallingConv.h"
 
+#include "SIDefines.h"
+
 #define GET_INSTRINFO_OPERAND_ENUM
 #include "AMDGPUGenInstrInfo.inc"
 #undef GET_INSTRINFO_OPERAND_ENUM
@@ -167,6 +169,37 @@ unsigned getRegBitWidth(const MCRegisterClass &RC);
 unsigned getRegOperandSize(const MCRegisterInfo *MRI, const MCInstrDesc &Desc,
                            unsigned OpNo);
 
+LLVM_READNONE
+inline unsigned getOperandSize(const MCOperandInfo &OpInfo) {
+  switch (OpInfo.OperandType) {
+  case AMDGPU::OPERAND_REG_IMM_INT32:
+  case AMDGPU::OPERAND_REG_IMM_FP32:
+  case AMDGPU::OPERAND_REG_INLINE_C_INT32:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP32:
+    return 4;
+
+  case AMDGPU::OPERAND_REG_IMM_INT64:
+  case AMDGPU::OPERAND_REG_IMM_FP64:
+  case AMDGPU::OPERAND_REG_INLINE_C_INT64:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP64:
+    return 8;
+
+  case AMDGPU::OPERAND_REG_IMM_INT16:
+  case AMDGPU::OPERAND_REG_IMM_FP16:
+  case AMDGPU::OPERAND_REG_INLINE_C_INT16:
+  case AMDGPU::OPERAND_REG_INLINE_C_FP16:
+    return 2;
+
+  default:
+    llvm_unreachable("unhandled operand type");
+  }
+}
+
+LLVM_READNONE
+inline unsigned getOperandSize(const MCInstrDesc &Desc, unsigned OpNo) {
+  return getOperandSize(Desc.OpInfo[OpNo]);
+}
+
 /// \brief Is this literal inlinable
 LLVM_READNONE
 bool isInlinableLiteral64(int64_t Literal, bool HasInv2Pi);
@@ -174,6 +207,8 @@ bool isInlinableLiteral64(int64_t Literal, bool HasInv2Pi);
 LLVM_READNONE
 bool isInlinableLiteral32(int32_t Literal, bool HasInv2Pi);
 
+LLVM_READNONE
+bool isInlinableLiteral16(int16_t Literal, bool HasInv2Pi);
 
 } // end namespace AMDGPU
 } // end namespace llvm