9375962ac9
The vast majority of the following (very common) opcodes were always called with identical arguments: - `GIM_CheckType` for the root - `GIM_CheckRegBankForClass` for the root - `GIR_Copy` between the old and new root - `GIR_ConstrainSelectedInstOperands` on the new root - `GIR_BuildMI` to create the new root I added overloaded version of each opcode specialized for the root instructions. It always saves between 1 and 2 bytes per instance depending on the number of arguments specialized into the opcode. Some of these opcodes had between 5 and 15k occurences in the AArch64 GlobalISel Match Table. Additionally, the following opcodes are almost always used in the same sequence: - `GIR_EraseFromParent 0` + `GIR_Done` - `GIR_EraseRootFromParent_Done` has been created to do both. Saves 2 bytes per occurence. - `GIR_IsSafeToFold` was *always* called for each InsnID except 0. - Changed the opcode to take the number of instructions to check after `MI[0]` The savings from these are pretty neat. For `AArch64GenGlobalISel.inc`: - `AArch64InstructionSelector.cpp.o` goes down from 772kb to 704kb (-10% code size) - Self-reported MatchTable size goes from 420380 bytes to 352426 bytes (~ -17%) A smaller match table means a faster match table because we spend less time iterating and decoding. I don't have a solid measurement methodology for GlobalISel performance so I don't have precise numbers but I saw a few % of improvements in a simple testcase.
56 lines
3.4 KiB
TableGen
56 lines
3.4 KiB
TableGen
// RUN: llvm-tblgen -gen-global-isel -I %p/../../include -I %p/Common %s -o - | FileCheck %s
|
|
|
|
include "llvm/Target/Target.td"
|
|
include "GlobalISelEmitterCommon.td"
|
|
|
|
def : GINodeEquiv<G_BUILD_VECTOR, build_vector>;
|
|
|
|
def ONE : I<(outs GPR32:$dst), (ins GPR32:$src1), []>;
|
|
def TWO : I<(outs GPR32:$dst), (ins GPR32:$src1, GPR32:$src2), []>;
|
|
|
|
// G_BUILD_VECTOR is guaranteed to have at least one operand, therefore performing a
|
|
// number of operands check is not needed to avoid per-operand checks accessing the
|
|
// MI's operand list out of bounds in this particular pattern. However, we still
|
|
// must perform the check, as a G_BUILD_VECTOR instance with two source operands
|
|
// will pass all checks done by this pattern otherwise, which will lead to a
|
|
// mis-match if this pattern tried first (and it will if it has higher complexity).
|
|
def : Pat<(build_vector GPR32:$src1),
|
|
(ONE GPR32:$src1)> {
|
|
let AddedComplexity = 1000;
|
|
}
|
|
|
|
def : Pat<(build_vector GPR32:$src1, GPR32:$src2),
|
|
(TWO GPR32:$src1, GPR32:$src2)>;
|
|
|
|
// CHECK: GIM_Try, /*On fail goto*//*Label 0*/ GIMT_Encode4([[NEXT_OPCODE_LABEL:[0-9]+]]),
|
|
// CHECK-NEXT: GIM_CheckOpcode, /*MI*/0, GIMT_Encode2(TargetOpcode::G_BUILD_VECTOR),
|
|
// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 1*/ GIMT_Encode4([[NEXT_NUM_OPERANDS_LABEL_1:[0-9]+]]), // Rule ID 0 //
|
|
// CHECK-NEXT: GIM_CheckNumOperands, /*MI*/0, /*Expected*/2,
|
|
// CHECK-NEXT: GIM_RootCheckType, /*Op*/0, /*Type*/GILLT_s32,
|
|
// CHECK-NEXT: GIM_RootCheckType, /*Op*/1, /*Type*/GILLT_s32,
|
|
// CHECK-NEXT: GIM_RootCheckRegBankForClass, /*Op*/0, /*RC*/GIMT_Encode2(MyTarget::GPR32RegClassID),
|
|
// CHECK-NEXT: GIM_RootCheckRegBankForClass, /*Op*/1, /*RC*/GIMT_Encode2(MyTarget::GPR32RegClassID),
|
|
// CHECK-NEXT: // (build_vector:{ *:[i32] } GPR32:{ *:[i32] }:$src1) => (ONE:{ *:[i32] } GPR32:{ *:[i32] }:$src1)
|
|
// CHECK-NEXT: GIR_MutateOpcode, /*InsnID*/0, /*RecycleInsnID*/0, /*Opcode*/GIMT_Encode2(MyTarget::ONE),
|
|
// CHECK-NEXT: GIR_RootConstrainSelectedInstOperands,
|
|
// CHECK-NEXT: // GIR_Coverage, 0,
|
|
// CHECK-NEXT: GIR_Done,
|
|
// CHECK-NEXT: // Label 1: @[[NEXT_NUM_OPERANDS_LABEL_1]]
|
|
// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 2*/ GIMT_Encode4([[NEXT_NUM_OPERANDS_LABEL_2:[0-9]+]]), // Rule ID 1 //
|
|
// CHECK-NEXT: GIM_CheckNumOperands, /*MI*/0, /*Expected*/3,
|
|
// CHECK-NEXT: GIM_RootCheckType, /*Op*/0, /*Type*/GILLT_s32,
|
|
// CHECK-NEXT: GIM_RootCheckType, /*Op*/1, /*Type*/GILLT_s32,
|
|
// CHECK-NEXT: GIM_RootCheckType, /*Op*/2, /*Type*/GILLT_s32,
|
|
// CHECK-NEXT: GIM_RootCheckRegBankForClass, /*Op*/0, /*RC*/GIMT_Encode2(MyTarget::GPR32RegClassID),
|
|
// CHECK-NEXT: GIM_RootCheckRegBankForClass, /*Op*/1, /*RC*/GIMT_Encode2(MyTarget::GPR32RegClassID),
|
|
// CHECK-NEXT: GIM_RootCheckRegBankForClass, /*Op*/2, /*RC*/GIMT_Encode2(MyTarget::GPR32RegClassID),
|
|
// CHECK-NEXT: // (build_vector:{ *:[i32] } GPR32:{ *:[i32] }:$src1, GPR32:{ *:[i32] }:$src2) => (TWO:{ *:[i32] } GPR32:{ *:[i32] }:$src1, GPR32:{ *:[i32] }:$src2)
|
|
// CHECK-NEXT: GIR_MutateOpcode, /*InsnID*/0, /*RecycleInsnID*/0, /*Opcode*/GIMT_Encode2(MyTarget::TWO),
|
|
// CHECK-NEXT: GIR_RootConstrainSelectedInstOperands,
|
|
// CHECK-NEXT: // GIR_Coverage, 1,
|
|
// CHECK-NEXT: GIR_Done,
|
|
// CHECK-NEXT: // Label 2: @[[NEXT_NUM_OPERANDS_LABEL_2]]
|
|
// CHECK-NEXT: GIM_Reject,
|
|
// CHECK-NEXT: // Label 0: @[[NEXT_OPCODE_LABEL]]
|
|
// CHECK-NEXT: GIM_Reject,
|