[GlobalISel] Remove semantic operand of G_IS_FPCLASS

Instruction G_IS_FPCLASS had an operand that represented floating-point
semantics of its first operand. It allowed types that have the same length,
like `bfloat16` and `half`, to be distinguished. Unfortunately, it is
not sufficient, as other operation still cannot distinguish such types.
Solution of this problem must be more general, so now this operand is removed.

Differential Revision: https://reviews.llvm.org/D138004
This commit is contained in:
Serge Pavlov 2022-11-15 12:02:06 +07:00
parent b97911c37b
commit ec893da990
4 changed files with 9 additions and 23 deletions

View File

@ -484,8 +484,7 @@ G_IS_FPCLASS
^^^^^^^^^^^^
Tests if the first operand, which must be floating-point scalar or vector, has
floating-point class specified by the second operand. The third operand
specifies floating-point semantics of the tested value. Returns non-zero (true)
floating-point class specified by the second operand. Returns non-zero (true)
or zero (false). It's target specific whether a true value is 1, ~0U, or some
other non-zero value. If the first operand is a vector, the returned value is a
vector of the same length.

View File

@ -748,7 +748,7 @@ def G_FCANONICALIZE : GenericInstruction {
// Generic opcode equivalent to the llvm.is_fpclass intrinsic.
def G_IS_FPCLASS: GenericInstruction {
let OutOperandList = (outs type0:$dst);
let InOperandList = (ins type1:$src, unknown:$test, unknown:$fpsem);
let InOperandList = (ins type1:$src, unknown:$test);
let hasSideEffects = false;
}

View File

@ -1714,16 +1714,6 @@ void MachineVerifier::verifyPreISelGenericInstruction(const MachineInstr *MI) {
report("Incorrect floating-point class set (operand 2)", MI);
break;
}
const MachineOperand &SemanticsMO = MI->getOperand(3);
if (!SemanticsMO.isImm()) {
report("floating-point semantics (operand 3) must be an immediate", MI);
break;
}
int64_t Semantics = SemanticsMO.getImm();
if (Semantics < 0 || Semantics > APFloat::S_MaxSemantics) {
report("Incorrect floating-point semantics (operand 3)", MI);
break;
}
break;
}
case TargetOpcode::G_ASSERT_ALIGN: {

View File

@ -15,26 +15,23 @@ body: |
%ptr:_(p0) = COPY $x0
%vector:_(<4 x s32>) = COPY $q0
%val1:_(s1) = G_IS_FPCLASS %s32, 1
%val1:_(s1) = G_IS_FPCLASS %s32
; CHECK: *** Bad machine code: Too few operands ***
; CHECK: 4 operands expected, but 3 given.
; CHECK: 3 operands expected, but 2 given.
%val2:_(p0) = G_IS_FPCLASS %s32, 3, 2
%val2:_(p0) = G_IS_FPCLASS %s32, 3
; CHECK: *** Bad machine code: Destination must be a scalar or vector of scalars ***
%val3:_(s1) = G_IS_FPCLASS %s32, 1, 66
; CHECK: *** Bad machine code: Incorrect floating-point semantics (operand 3) ***
%val4:_(s1) = G_IS_FPCLASS %s32, 7777, 2
%val4:_(s1) = G_IS_FPCLASS %s32, 7777
; CHECK: *** Bad machine code: Incorrect floating-point class set (operand 2) ***
%val5:_(s1) = G_IS_FPCLASS %ptr:_(p0), 3, 2
%val5:_(s1) = G_IS_FPCLASS %ptr:_(p0), 3
; CHECK: *** Bad machine code: Source must be a scalar or vector of scalars ***
%var6:_(s1) = G_IS_FPCLASS %vector:_(<4 x s32>), 1, 2
%var6:_(s1) = G_IS_FPCLASS %vector:_(<4 x s32>), 1
; CHECK: *** Bad machine code: operand types must be all-vector or all-scalar ***
%var7:_(<2 x s1>) = G_IS_FPCLASS %vector:_(<4 x s32>), 1, 2
%var7:_(<2 x s1>) = G_IS_FPCLASS %vector:_(<4 x s32>), 1
; CHECK: *** Bad machine code: operand types must preserve number of vector elements ***
...