AMDGPU: Fix packed 16-bit inline constants (#76522)

Consistently treat packed 16-bit operands as 32-bit values, because
that's really what they are. The attempt to treat them differently was
ultimately incorrect and lead to miscompiles, e.g. when using non-splat
constants such as (1, 0) as operands.

Recognize 32-bit float constants for i/u16 instructions. This is a bit
odd conceptually, but it matches HW behavior and SP3.

Remove isFoldableLiteralV216; there was too much magic in the dependency
between it and its use in SIFoldOperands. Instead, we now simply rely on
checking whether a constant is an inline constant, and trying a bunch of
permutations of the low and high halves. This is more obviously correct
and leads to some new cases where inline constants are used as shown by
tests.

Move the logic for switching packed add vs. sub into SIFoldOperands.
This has two benefits: all logic that optimizes for inline constants in
packed math is now in one place; and it applies to both SelectionDAG and
GISel paths.

Disable the use of opsel with v_dot* instructions on gfx11. They are
documented to ignore opsel on src0 and src1. It may be interesting to
re-enable to use of opsel on src2 as a future optimization.

A similar "proper" fix of what inline constants mean could potentially
be applied to unpacked 16-bit ops. However, it's less clear what the
benefit would be, and there are surely places where we'd have to
carefully audit whether values are properly sign- or zero-extended. It
is best to keep such a change separate.

Fixes: Corruption in FSR 2.0 (latent bug exposed by an LLPC change)

1 files changed, 13 insertions, 8 deletions

diff --git a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index 5f2b7c0..b7f0438 100644
--- a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -1865,6 +1865,9 @@ static const fltSemantics *getOpFltSemantics(uint8_t OperandType) {
   case AMDGPU::OPERAND_REG_IMM_V2FP32:
   case AMDGPU::OPERAND_REG_INLINE_C_V2INT32:
   case AMDGPU::OPERAND_REG_IMM_V2INT32:
+  case AMDGPU::OPERAND_REG_IMM_V2INT16:
+  case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
+  case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
   case AMDGPU::OPERAND_KIMM32:
   case AMDGPU::OPERAND_INLINE_SPLIT_BARRIER_INT32:
     return &APFloat::IEEEsingle();
@@ -1879,13 +1882,10 @@ static const fltSemantics *getOpFltSemantics(uint8_t OperandType) {
   case AMDGPU::OPERAND_REG_IMM_FP16_DEFERRED:
   case AMDGPU::OPERAND_REG_INLINE_C_INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_FP16:
-  case AMDGPU::OPERAND_REG_INLINE_C_V2INT16:
   case AMDGPU::OPERAND_REG_INLINE_C_V2FP16:
   case AMDGPU::OPERAND_REG_INLINE_AC_INT16:
   case AMDGPU::OPERAND_REG_INLINE_AC_FP16:
-  case AMDGPU::OPERAND_REG_INLINE_AC_V2INT16:
   case AMDGPU::OPERAND_REG_INLINE_AC_V2FP16:
-  case AMDGPU::OPERAND_REG_IMM_V2INT16:
   case AMDGPU::OPERAND_REG_IMM_V2FP16:
   case AMDGPU::OPERAND_KIMM16:
     return &APFloat::IEEEhalf();
@@ -2033,9 +2033,14 @@ bool AMDGPUOperand::isLiteralImm(MVT type) const {
   // We allow fp literals with f16x2 operands assuming that the specified
   // literal goes into the lower half and the upper half is zero. We also
   // require that the literal may be losslessly converted to f16.
-  MVT ExpectedType = (type == MVT::v2f16)? MVT::f16 :
-                     (type == MVT::v2i16)? MVT::i16 :
-                     (type == MVT::v2f32)? MVT::f32 : type;
+  //
+  // For i16x2 operands, we assume that the specified literal is encoded as a
+  // single-precision float. This is pretty odd, but it matches SP3 and what
+  // happens in hardware.
+  MVT ExpectedType = (type == MVT::v2f16)   ? MVT::f16
+                     : (type == MVT::v2i16) ? MVT::f32
+                     : (type == MVT::v2f32) ? MVT::f32
+                                            : type;
 
   APFloat FPLiteral(APFloat::IEEEdouble(), APInt(64, Imm.Val));
   return canLosslesslyConvertToFPType(FPLiteral, ExpectedType);
@@ -3401,12 +3406,12 @@ bool AMDGPUAsmParser::isInlineConstant(const MCInst &Inst,
     if (OperandType == AMDGPU::OPERAND_REG_INLINE_C_V2INT16 ||
         OperandType == AMDGPU::OPERAND_REG_INLINE_AC_V2INT16 ||
         OperandType == AMDGPU::OPERAND_REG_IMM_V2INT16)
-      return AMDGPU::isInlinableIntLiteralV216(Val);
+      return AMDGPU::isInlinableLiteralV2I16(Val);
 
     if (OperandType == AMDGPU::OPERAND_REG_INLINE_C_V2FP16 ||
         OperandType == AMDGPU::OPERAND_REG_INLINE_AC_V2FP16 ||
         OperandType == AMDGPU::OPERAND_REG_IMM_V2FP16)
-      return AMDGPU::isInlinableLiteralV216(Val, hasInv2PiInlineImm());
+      return AMDGPU::isInlinableLiteralV2F16(Val);
 
     return AMDGPU::isInlinableLiteral16(Val, hasInv2PiInlineImm());
   }