mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-26 12:43:36 +01:00
8c42ad8897
Unlike normal loads these don't have an extension field, but we know from TargetLowering whether these are sign-extending or zero-extending, and so can optimise away unnecessary extensions. This was noticed on RISC-V, where sign extensions in the calling convention would result in unnecessary explicit extension instructions, but this also fixes some Mips inefficiencies. PowerPC sees churn in the tests as all the zero extensions are only for promoting 32-bit to 64-bit, but these zero extensions are still not optimised away as they should be, likely due to i32 being a legal type. This also simplifies the WebAssembly code somewhat, which currently works around the lack of target-independent combines with some ugly patterns that break once they're optimised away. Re-landed with correct handling in ComputeNumSignBits for Tmp == VTBits, where zero-extending atomics were incorrectly returning 0 rather than the (slightly confusing) required return value of 1. Re-landed again after D102819 fixed PowerPC to correctly zero-extend all of its atomics as it claimed to do, since the combination of that bug and this optimisation caused buildbot regressions. Reviewed By: RKSimon, atanasyan Differential Revision: https://reviews.llvm.org/D101342
968 lines
48 KiB
TableGen
968 lines
48 KiB
TableGen
// WebAssemblyInstrAtomics.td-WebAssembly Atomic codegen support-*- tablegen -*-
|
|
//
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
|
//
|
|
//===----------------------------------------------------------------------===//
|
|
///
|
|
/// \file
|
|
/// WebAssembly Atomic operand code-gen constructs.
|
|
///
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
let UseNamedOperandTable = 1 in
|
|
multiclass ATOMIC_I<dag oops_r, dag iops_r, dag oops_s, dag iops_s,
|
|
list<dag> pattern_r, string asmstr_r,
|
|
string asmstr_s, bits<32> atomic_op,
|
|
string is64 = "false"> {
|
|
defm "" : I<oops_r, iops_r, oops_s, iops_s, pattern_r, asmstr_r, asmstr_s,
|
|
!or(0xfe00, !and(0xff, atomic_op)), is64>,
|
|
Requires<[HasAtomics]>;
|
|
}
|
|
|
|
multiclass ATOMIC_NRI<dag oops, dag iops, list<dag> pattern, string asmstr = "",
|
|
bits<32> atomic_op = -1> {
|
|
defm "" : NRI<oops, iops, pattern, asmstr,
|
|
!or(0xfe00, !and(0xff, atomic_op))>,
|
|
Requires<[HasAtomics]>;
|
|
}
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic wait / notify
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
let hasSideEffects = 1 in {
|
|
defm MEMORY_ATOMIC_NOTIFY_A32 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset32_op:$off, I32:$addr, I32:$count),
|
|
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
|
|
"memory.atomic.notify \t$dst, ${off}(${addr})${p2align}, $count",
|
|
"memory.atomic.notify \t${off}${p2align}", 0x00, "false">;
|
|
defm MEMORY_ATOMIC_NOTIFY_A64 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset64_op:$off, I64:$addr, I32:$count),
|
|
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
|
|
"memory.atomic.notify \t$dst, ${off}(${addr})${p2align}, $count",
|
|
"memory.atomic.notify \t${off}${p2align}", 0x00, "true">;
|
|
let mayLoad = 1 in {
|
|
defm MEMORY_ATOMIC_WAIT32_A32 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset32_op:$off, I32:$addr, I32:$exp,
|
|
I64:$timeout),
|
|
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
|
|
"memory.atomic.wait32 \t$dst, ${off}(${addr})${p2align}, $exp, $timeout",
|
|
"memory.atomic.wait32 \t${off}${p2align}", 0x01, "false">;
|
|
defm MEMORY_ATOMIC_WAIT32_A64 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset64_op:$off, I64:$addr, I32:$exp,
|
|
I64:$timeout),
|
|
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
|
|
"memory.atomic.wait32 \t$dst, ${off}(${addr})${p2align}, $exp, $timeout",
|
|
"memory.atomic.wait32 \t${off}${p2align}", 0x01, "true">;
|
|
defm MEMORY_ATOMIC_WAIT64_A32 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset32_op:$off, I32:$addr, I64:$exp,
|
|
I64:$timeout),
|
|
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
|
|
"memory.atomic.wait64 \t$dst, ${off}(${addr})${p2align}, $exp, $timeout",
|
|
"memory.atomic.wait64 \t${off}${p2align}", 0x02, "false">;
|
|
defm MEMORY_ATOMIC_WAIT64_A64 :
|
|
ATOMIC_I<(outs I32:$dst),
|
|
(ins P2Align:$p2align, offset64_op:$off, I64:$addr, I64:$exp,
|
|
I64:$timeout),
|
|
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
|
|
"memory.atomic.wait64 \t$dst, ${off}(${addr})${p2align}, $exp, $timeout",
|
|
"memory.atomic.wait64 \t${off}${p2align}", 0x02, "true">;
|
|
} // mayLoad = 1
|
|
} // hasSideEffects = 1
|
|
|
|
// Select notifys with no constant offset.
|
|
def NotifyPatNoOffset_A32 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify I32:$addr, I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A32 0, 0, I32:$addr, I32:$count)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def NotifyPatNoOffset_A64 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify I64:$addr, I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A64 0, 0, I64:$addr, I32:$count)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
|
|
// Select notifys with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
multiclass NotifyPatImmOff<PatFrag operand, string inst> {
|
|
def : Pat<(i32 (int_wasm_memory_atomic_notify (operand I32:$addr, imm:$off),
|
|
I32:$count)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, I32:$addr, I32:$count)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(i32 (int_wasm_memory_atomic_notify (operand I64:$addr, imm:$off),
|
|
I32:$count)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, I64:$addr, I32:$count)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : NotifyPatImmOff<regPlusImm, "MEMORY_ATOMIC_NOTIFY">;
|
|
defm : NotifyPatImmOff<or_is_add, "MEMORY_ATOMIC_NOTIFY">;
|
|
|
|
// Select notifys with just a constant offset.
|
|
def NotifyPatOffsetOnly_A32 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify imm:$off, I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A32 0, imm:$off, (CONST_I32 0), I32:$count)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def NotifyPatOffsetOnly_A64 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify imm:$off, I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A64 0, imm:$off, (CONST_I64 0), I32:$count)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
|
|
def NotifyPatGlobalAddrOffOnly_A32 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify (WebAssemblywrapper tglobaladdr:$off),
|
|
I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A32 0, tglobaladdr:$off, (CONST_I32 0), I32:$count)
|
|
>,
|
|
Requires<[HasAddr32, HasAtomics, IsNotPIC]>;
|
|
def NotifyPatGlobalAddrOffOnly_A64 :
|
|
Pat<(i32 (int_wasm_memory_atomic_notify (WebAssemblywrapper tglobaladdr:$off),
|
|
I32:$count)),
|
|
(MEMORY_ATOMIC_NOTIFY_A64 0, tglobaladdr:$off, (CONST_I64 0), I32:$count)
|
|
>,
|
|
Requires<[HasAddr64, HasAtomics, IsNotPIC]>;
|
|
|
|
// Select waits with no constant offset.
|
|
multiclass WaitPatNoOffset<ValueType ty, Intrinsic kind,
|
|
string inst> {
|
|
def : Pat<(i32 (kind I32:$addr, ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A32) 0, 0, I32:$addr, ty:$exp, I64:$timeout)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(i32 (kind I64:$addr, ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A64) 0, 0, I64:$addr, ty:$exp, I64:$timeout)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : WaitPatNoOffset<i32, int_wasm_memory_atomic_wait32,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatNoOffset<i64, int_wasm_memory_atomic_wait64,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
defm : WaitPatNoOffset<i32, int_wasm_memory_atomic_wait32,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatNoOffset<i64, int_wasm_memory_atomic_wait64,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
|
|
// Select waits with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
multiclass WaitPatImmOff<ValueType ty, Intrinsic kind, PatFrag operand,
|
|
string inst> {
|
|
def : Pat<(i32 (kind (operand I32:$addr, imm:$off), ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, I32:$addr, ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(i32 (kind (operand I64:$addr, imm:$off), ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, I64:$addr, ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : WaitPatImmOff<i32, int_wasm_memory_atomic_wait32, regPlusImm,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatImmOff<i32, int_wasm_memory_atomic_wait32, or_is_add,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatImmOff<i64, int_wasm_memory_atomic_wait64, regPlusImm,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
defm : WaitPatImmOff<i64, int_wasm_memory_atomic_wait64, or_is_add,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
|
|
// Select waits with just a constant offset.
|
|
multiclass WaitPatOffsetOnly<ValueType ty, Intrinsic kind, string inst> {
|
|
def : Pat<(i32 (kind imm:$off, ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, (CONST_I32 0), ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(i32 (kind imm:$off, ty:$exp, I64:$timeout)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, (CONST_I64 0), ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : WaitPatOffsetOnly<i32, int_wasm_memory_atomic_wait32,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatOffsetOnly<i64, int_wasm_memory_atomic_wait64,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
|
|
multiclass WaitPatGlobalAddrOffOnly<ValueType ty, Intrinsic kind, string inst> {
|
|
def : Pat<(i32 (kind (WebAssemblywrapper tglobaladdr:$off), ty:$exp,
|
|
I64:$timeout)),
|
|
(!cast<NI>(inst#_A32) 0, tglobaladdr:$off, (CONST_I32 0), ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr32, HasAtomics, IsNotPIC]>;
|
|
def : Pat<(i32 (kind (WebAssemblywrapper tglobaladdr:$off), ty:$exp,
|
|
I64:$timeout)),
|
|
(!cast<NI>(inst#_A64) 0, tglobaladdr:$off, (CONST_I64 0), ty:$exp,
|
|
I64:$timeout)>,
|
|
Requires<[HasAddr64, HasAtomics, IsNotPIC]>;
|
|
}
|
|
defm : WaitPatGlobalAddrOffOnly<i32, int_wasm_memory_atomic_wait32,
|
|
"MEMORY_ATOMIC_WAIT32">;
|
|
defm : WaitPatGlobalAddrOffOnly<i64, int_wasm_memory_atomic_wait64,
|
|
"MEMORY_ATOMIC_WAIT64">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic fences
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// A compiler fence instruction that prevents reordering of instructions.
|
|
let Defs = [ARGUMENTS] in {
|
|
let isPseudo = 1, hasSideEffects = 1 in
|
|
defm COMPILER_FENCE : ATOMIC_NRI<(outs), (ins), [], "compiler_fence">;
|
|
let hasSideEffects = 1 in
|
|
defm ATOMIC_FENCE : ATOMIC_NRI<(outs), (ins i8imm:$flags), [], "atomic.fence",
|
|
0x03>;
|
|
} // Defs = [ARGUMENTS]
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic loads
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass AtomicLoad<WebAssemblyRegClass rc, string name, int atomic_op> {
|
|
defm "" : WebAssemblyLoad<rc, name, !or(0xfe00, !and(0xff, atomic_op)),
|
|
[HasAtomics]>;
|
|
}
|
|
|
|
defm ATOMIC_LOAD_I32 : AtomicLoad<I32, "i32.atomic.load", 0x10>;
|
|
defm ATOMIC_LOAD_I64 : AtomicLoad<I64, "i64.atomic.load", 0x11>;
|
|
|
|
// Select loads with no constant offset.
|
|
defm : LoadPatNoOffset<i32, atomic_load_32, "ATOMIC_LOAD_I32">;
|
|
defm : LoadPatNoOffset<i64, atomic_load_64, "ATOMIC_LOAD_I64">;
|
|
|
|
// Select loads with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
defm : LoadPatImmOff<i32, atomic_load_32, regPlusImm, "ATOMIC_LOAD_I32">;
|
|
defm : LoadPatImmOff<i64, atomic_load_64, regPlusImm, "ATOMIC_LOAD_I64">;
|
|
defm : LoadPatImmOff<i32, atomic_load_32, or_is_add, "ATOMIC_LOAD_I32">;
|
|
defm : LoadPatImmOff<i64, atomic_load_64, or_is_add, "ATOMIC_LOAD_I64">;
|
|
|
|
// Select loads with just a constant offset.
|
|
defm : LoadPatOffsetOnly<i32, atomic_load_32, "ATOMIC_LOAD_I32">;
|
|
defm : LoadPatOffsetOnly<i64, atomic_load_64, "ATOMIC_LOAD_I64">;
|
|
|
|
defm : LoadPatGlobalAddrOffOnly<i32, atomic_load_32, "ATOMIC_LOAD_I32">;
|
|
defm : LoadPatGlobalAddrOffOnly<i64, atomic_load_64, "ATOMIC_LOAD_I64">;
|
|
|
|
|
|
// Extending loads. Note that there are only zero-extending atomic loads, no
|
|
// sign-extending loads.
|
|
defm ATOMIC_LOAD8_U_I32 : AtomicLoad<I32, "i32.atomic.load8_u", 0x12>;
|
|
defm ATOMIC_LOAD16_U_I32 : AtomicLoad<I32, "i32.atomic.load16_u", 0x13>;
|
|
defm ATOMIC_LOAD8_U_I64 : AtomicLoad<I64, "i64.atomic.load8_u", 0x14>;
|
|
defm ATOMIC_LOAD16_U_I64 : AtomicLoad<I64, "i64.atomic.load16_u", 0x15>;
|
|
defm ATOMIC_LOAD32_U_I64 : AtomicLoad<I64, "i64.atomic.load32_u", 0x16>;
|
|
|
|
// Fragments for extending loads. These are different from regular loads because
|
|
// the SDNodes are derived from AtomicSDNode rather than LoadSDNode and
|
|
// therefore don't have the extension type field. So instead of matching that,
|
|
// we match the patterns that the type legalizer expands them to.
|
|
|
|
// Unlike regular loads, extension to i64 is handled differently than i32.
|
|
// i64 (zext (i8 (atomic_load_8))) gets legalized to
|
|
// i64 (and (i64 (anyext (i32 (atomic_load_8)))), 255)
|
|
// Extension to i32 is elided by SelectionDAG as our atomic loads are
|
|
// zero-extending.
|
|
def zext_aload_8_64 :
|
|
PatFrag<(ops node:$addr),
|
|
(i64 (zext (i32 (atomic_load_8 node:$addr))))>;
|
|
def zext_aload_16_64 :
|
|
PatFrag<(ops node:$addr),
|
|
(i64 (zext (i32 (atomic_load_16 node:$addr))))>;
|
|
def zext_aload_32_64 :
|
|
PatFrag<(ops node:$addr),
|
|
(i64 (zext (i32 (atomic_load_32 node:$addr))))>;
|
|
|
|
// We don't have single sext atomic load instructions. So for sext loads, we
|
|
// match bare subword loads (for 32-bit results) and anyext loads (for 64-bit
|
|
// results) and select a zext load; the next instruction will be sext_inreg
|
|
// which is selected by itself.
|
|
def sext_aload_8_64 :
|
|
PatFrag<(ops node:$addr), (anyext (i32 (atomic_load_8 node:$addr)))>;
|
|
def sext_aload_16_64 :
|
|
PatFrag<(ops node:$addr), (anyext (i32 (atomic_load_16 node:$addr)))>;
|
|
|
|
// Select zero-extending loads with no constant offset.
|
|
defm : LoadPatNoOffset<i64, zext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatNoOffset<i64, zext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatNoOffset<i64, zext_aload_32_64, "ATOMIC_LOAD32_U_I64">;
|
|
|
|
// Select sign-extending loads with no constant offset
|
|
defm : LoadPatNoOffset<i32, atomic_load_8, "ATOMIC_LOAD8_U_I32">;
|
|
defm : LoadPatNoOffset<i32, atomic_load_16, "ATOMIC_LOAD16_U_I32">;
|
|
defm : LoadPatNoOffset<i64, sext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatNoOffset<i64, sext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
// 32->64 sext load gets selected as i32.atomic.load, i64.extend_i32_s
|
|
|
|
// Zero-extending loads with constant offset
|
|
defm : LoadPatImmOff<i64, zext_aload_8_64, regPlusImm, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatImmOff<i64, zext_aload_16_64, regPlusImm, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatImmOff<i64, zext_aload_32_64, regPlusImm, "ATOMIC_LOAD32_U_I64">;
|
|
defm : LoadPatImmOff<i64, zext_aload_8_64, or_is_add, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatImmOff<i64, zext_aload_16_64, or_is_add, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatImmOff<i64, zext_aload_32_64, or_is_add, "ATOMIC_LOAD32_U_I64">;
|
|
|
|
// Sign-extending loads with constant offset
|
|
defm : LoadPatImmOff<i32, atomic_load_8, regPlusImm, "ATOMIC_LOAD8_U_I32">;
|
|
defm : LoadPatImmOff<i32, atomic_load_16, regPlusImm, "ATOMIC_LOAD16_U_I32">;
|
|
defm : LoadPatImmOff<i32, atomic_load_8, or_is_add, "ATOMIC_LOAD8_U_I32">;
|
|
defm : LoadPatImmOff<i32, atomic_load_16, or_is_add, "ATOMIC_LOAD16_U_I32">;
|
|
defm : LoadPatImmOff<i64, sext_aload_8_64, regPlusImm, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatImmOff<i64, sext_aload_16_64, regPlusImm, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatImmOff<i64, sext_aload_8_64, or_is_add, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatImmOff<i64, sext_aload_16_64, or_is_add, "ATOMIC_LOAD16_U_I64">;
|
|
// No 32->64 patterns, just use i32.atomic.load and i64.extend_s/i64
|
|
|
|
// Extending loads with just a constant offset
|
|
defm : LoadPatOffsetOnly<i64, zext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatOffsetOnly<i64, zext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatOffsetOnly<i64, zext_aload_32_64, "ATOMIC_LOAD32_U_I64">;
|
|
defm : LoadPatOffsetOnly<i32, atomic_load_8, "ATOMIC_LOAD8_U_I32">;
|
|
defm : LoadPatOffsetOnly<i32, atomic_load_16, "ATOMIC_LOAD16_U_I32">;
|
|
defm : LoadPatOffsetOnly<i64, sext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatOffsetOnly<i64, sext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
|
|
defm : LoadPatGlobalAddrOffOnly<i64, zext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatGlobalAddrOffOnly<i64, zext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
defm : LoadPatGlobalAddrOffOnly<i64, zext_aload_32_64, "ATOMIC_LOAD32_U_I64">;
|
|
defm : LoadPatGlobalAddrOffOnly<i32, atomic_load_8, "ATOMIC_LOAD8_U_I32">;
|
|
defm : LoadPatGlobalAddrOffOnly<i32, atomic_load_16, "ATOMIC_LOAD16_U_I32">;
|
|
defm : LoadPatGlobalAddrOffOnly<i64, sext_aload_8_64, "ATOMIC_LOAD8_U_I64">;
|
|
defm : LoadPatGlobalAddrOffOnly<i64, sext_aload_16_64, "ATOMIC_LOAD16_U_I64">;
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic stores
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass AtomicStore<WebAssemblyRegClass rc, string name, int atomic_op> {
|
|
defm "" : WebAssemblyStore<rc, name, !or(0xfe00, !and(0xff, atomic_op)),
|
|
[HasAtomics]>;
|
|
}
|
|
|
|
defm ATOMIC_STORE_I32 : AtomicStore<I32, "i32.atomic.store", 0x17>;
|
|
defm ATOMIC_STORE_I64 : AtomicStore<I64, "i64.atomic.store", 0x18>;
|
|
|
|
// We need an 'atomic' version of store patterns because store and atomic_store
|
|
// nodes have different operand orders:
|
|
// store: (store $val, $ptr)
|
|
// atomic_store: (store $ptr, $val)
|
|
|
|
|
|
// Select stores with no constant offset.
|
|
multiclass AStorePatNoOffset<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(kind I32:$addr, ty:$val),
|
|
(!cast<NI>(inst#_A32) 0, 0, I32:$addr, ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(kind I64:$addr, ty:$val),
|
|
(!cast<NI>(inst#_A64) 0, 0, I64:$addr, ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : AStorePatNoOffset<i32, atomic_store_32, "ATOMIC_STORE_I32">;
|
|
defm : AStorePatNoOffset<i64, atomic_store_64, "ATOMIC_STORE_I64">;
|
|
|
|
// Select stores with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
multiclass AStorePatImmOff<ValueType ty, PatFrag kind, PatFrag operand,
|
|
string inst> {
|
|
def : Pat<(kind (operand I32:$addr, imm:$off), ty:$val),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, I32:$addr, ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(kind (operand I64:$addr, imm:$off), ty:$val),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, I64:$addr, ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : AStorePatImmOff<i32, atomic_store_32, regPlusImm, "ATOMIC_STORE_I32">;
|
|
defm : AStorePatImmOff<i64, atomic_store_64, regPlusImm, "ATOMIC_STORE_I64">;
|
|
|
|
// Select stores with just a constant offset.
|
|
multiclass AStorePatOffsetOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(kind imm:$off, ty:$val),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, (CONST_I32 0), ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(kind imm:$off, ty:$val),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, (CONST_I64 0), ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
defm : AStorePatOffsetOnly<i32, atomic_store_32, "ATOMIC_STORE_I32">;
|
|
defm : AStorePatOffsetOnly<i64, atomic_store_64, "ATOMIC_STORE_I64">;
|
|
|
|
multiclass AStorePatGlobalAddrOffOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(kind (WebAssemblywrapper tglobaladdr:$off), ty:$val),
|
|
(!cast<NI>(inst#_A32) 0, tglobaladdr:$off, (CONST_I32 0), ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics, IsNotPIC]>;
|
|
def : Pat<(kind (WebAssemblywrapper tglobaladdr:$off), ty:$val),
|
|
(!cast<NI>(inst#_A64) 0, tglobaladdr:$off, (CONST_I64 0), ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics, IsNotPIC]>;
|
|
}
|
|
defm : AStorePatGlobalAddrOffOnly<i32, atomic_store_32, "ATOMIC_STORE_I32">;
|
|
defm : AStorePatGlobalAddrOffOnly<i64, atomic_store_64, "ATOMIC_STORE_I64">;
|
|
|
|
|
|
// Truncating stores.
|
|
defm ATOMIC_STORE8_I32 : AtomicStore<I32, "i32.atomic.store8", 0x19>;
|
|
defm ATOMIC_STORE16_I32 : AtomicStore<I32, "i32.atomic.store16", 0x1a>;
|
|
defm ATOMIC_STORE8_I64 : AtomicStore<I64, "i64.atomic.store8", 0x1b>;
|
|
defm ATOMIC_STORE16_I64 : AtomicStore<I64, "i64.atomic.store16", 0x1c>;
|
|
defm ATOMIC_STORE32_I64 : AtomicStore<I64, "i64.atomic.store32", 0x1d>;
|
|
|
|
// Fragments for truncating stores.
|
|
|
|
// We don't have single truncating atomic store instructions. For 32-bit
|
|
// instructions, we just need to match bare atomic stores. On the other hand,
|
|
// truncating stores from i64 values are once truncated to i32 first.
|
|
class trunc_astore_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$val),
|
|
(kind node:$addr, (i32 (trunc (i64 node:$val))))>;
|
|
def trunc_astore_8_64 : trunc_astore_64<atomic_store_8>;
|
|
def trunc_astore_16_64 : trunc_astore_64<atomic_store_16>;
|
|
def trunc_astore_32_64 : trunc_astore_64<atomic_store_32>;
|
|
|
|
|
|
// Truncating stores with no constant offset
|
|
defm : AStorePatNoOffset<i32, atomic_store_8, "ATOMIC_STORE8_I32">;
|
|
defm : AStorePatNoOffset<i32, atomic_store_16, "ATOMIC_STORE16_I32">;
|
|
defm : AStorePatNoOffset<i64, trunc_astore_8_64, "ATOMIC_STORE8_I64">;
|
|
defm : AStorePatNoOffset<i64, trunc_astore_16_64, "ATOMIC_STORE16_I64">;
|
|
defm : AStorePatNoOffset<i64, trunc_astore_32_64, "ATOMIC_STORE32_I64">;
|
|
|
|
// Truncating stores with a constant offset
|
|
defm : AStorePatImmOff<i32, atomic_store_8, regPlusImm, "ATOMIC_STORE8_I32">;
|
|
defm : AStorePatImmOff<i32, atomic_store_16, regPlusImm, "ATOMIC_STORE16_I32">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_8_64, regPlusImm, "ATOMIC_STORE8_I64">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_16_64, regPlusImm,
|
|
"ATOMIC_STORE16_I64">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_32_64, regPlusImm,
|
|
"ATOMIC_STORE32_I64">;
|
|
defm : AStorePatImmOff<i32, atomic_store_8, or_is_add, "ATOMIC_STORE8_I32">;
|
|
defm : AStorePatImmOff<i32, atomic_store_16, or_is_add, "ATOMIC_STORE16_I32">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_8_64, or_is_add, "ATOMIC_STORE8_I64">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_16_64, or_is_add,
|
|
"ATOMIC_STORE16_I64">;
|
|
defm : AStorePatImmOff<i64, trunc_astore_32_64, or_is_add,
|
|
"ATOMIC_STORE32_I64">;
|
|
|
|
// Truncating stores with just a constant offset
|
|
defm : AStorePatOffsetOnly<i32, atomic_store_8, "ATOMIC_STORE8_I32">;
|
|
defm : AStorePatOffsetOnly<i32, atomic_store_16, "ATOMIC_STORE16_I32">;
|
|
defm : AStorePatOffsetOnly<i64, trunc_astore_8_64, "ATOMIC_STORE8_I64">;
|
|
defm : AStorePatOffsetOnly<i64, trunc_astore_16_64, "ATOMIC_STORE16_I64">;
|
|
defm : AStorePatOffsetOnly<i64, trunc_astore_32_64, "ATOMIC_STORE32_I64">;
|
|
|
|
defm : AStorePatGlobalAddrOffOnly<i32, atomic_store_8, "ATOMIC_STORE8_I32">;
|
|
defm : AStorePatGlobalAddrOffOnly<i32, atomic_store_16, "ATOMIC_STORE16_I32">;
|
|
defm : AStorePatGlobalAddrOffOnly<i64, trunc_astore_8_64, "ATOMIC_STORE8_I64">;
|
|
defm : AStorePatGlobalAddrOffOnly<i64, trunc_astore_16_64, "ATOMIC_STORE16_I64">;
|
|
defm : AStorePatGlobalAddrOffOnly<i64, trunc_astore_32_64, "ATOMIC_STORE32_I64">;
|
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic binary read-modify-writes
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
multiclass WebAssemblyBinRMW<WebAssemblyRegClass rc, string name,
|
|
int atomic_op> {
|
|
defm "_A32" :
|
|
ATOMIC_I<(outs rc:$dst),
|
|
(ins P2Align:$p2align, offset32_op:$off, I32:$addr, rc:$val),
|
|
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
|
|
!strconcat(name, "\t$dst, ${off}(${addr})${p2align}, $val"),
|
|
!strconcat(name, "\t${off}${p2align}"), atomic_op, "false">;
|
|
defm "_A64" :
|
|
ATOMIC_I<(outs rc:$dst),
|
|
(ins P2Align:$p2align, offset64_op:$off, I64:$addr, rc:$val),
|
|
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
|
|
!strconcat(name, "\t$dst, ${off}(${addr})${p2align}, $val"),
|
|
!strconcat(name, "\t${off}${p2align}"), atomic_op, "true">;
|
|
}
|
|
|
|
defm ATOMIC_RMW_ADD_I32 : WebAssemblyBinRMW<I32, "i32.atomic.rmw.add", 0x1e>;
|
|
defm ATOMIC_RMW_ADD_I64 : WebAssemblyBinRMW<I64, "i64.atomic.rmw.add", 0x1f>;
|
|
defm ATOMIC_RMW8_U_ADD_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.add_u", 0x20>;
|
|
defm ATOMIC_RMW16_U_ADD_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.add_u", 0x21>;
|
|
defm ATOMIC_RMW8_U_ADD_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.add_u", 0x22>;
|
|
defm ATOMIC_RMW16_U_ADD_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.add_u", 0x23>;
|
|
defm ATOMIC_RMW32_U_ADD_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.add_u", 0x24>;
|
|
|
|
defm ATOMIC_RMW_SUB_I32 : WebAssemblyBinRMW<I32, "i32.atomic.rmw.sub", 0x25>;
|
|
defm ATOMIC_RMW_SUB_I64 : WebAssemblyBinRMW<I64, "i64.atomic.rmw.sub", 0x26>;
|
|
defm ATOMIC_RMW8_U_SUB_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.sub_u", 0x27>;
|
|
defm ATOMIC_RMW16_U_SUB_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.sub_u", 0x28>;
|
|
defm ATOMIC_RMW8_U_SUB_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.sub_u", 0x29>;
|
|
defm ATOMIC_RMW16_U_SUB_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.sub_u", 0x2a>;
|
|
defm ATOMIC_RMW32_U_SUB_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.sub_u", 0x2b>;
|
|
|
|
defm ATOMIC_RMW_AND_I32 : WebAssemblyBinRMW<I32, "i32.atomic.rmw.and", 0x2c>;
|
|
defm ATOMIC_RMW_AND_I64 : WebAssemblyBinRMW<I64, "i64.atomic.rmw.and", 0x2d>;
|
|
defm ATOMIC_RMW8_U_AND_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.and_u", 0x2e>;
|
|
defm ATOMIC_RMW16_U_AND_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.and_u", 0x2f>;
|
|
defm ATOMIC_RMW8_U_AND_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.and_u", 0x30>;
|
|
defm ATOMIC_RMW16_U_AND_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.and_u", 0x31>;
|
|
defm ATOMIC_RMW32_U_AND_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.and_u", 0x32>;
|
|
|
|
defm ATOMIC_RMW_OR_I32 : WebAssemblyBinRMW<I32, "i32.atomic.rmw.or", 0x33>;
|
|
defm ATOMIC_RMW_OR_I64 : WebAssemblyBinRMW<I64, "i64.atomic.rmw.or", 0x34>;
|
|
defm ATOMIC_RMW8_U_OR_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.or_u", 0x35>;
|
|
defm ATOMIC_RMW16_U_OR_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.or_u", 0x36>;
|
|
defm ATOMIC_RMW8_U_OR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.or_u", 0x37>;
|
|
defm ATOMIC_RMW16_U_OR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.or_u", 0x38>;
|
|
defm ATOMIC_RMW32_U_OR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.or_u", 0x39>;
|
|
|
|
defm ATOMIC_RMW_XOR_I32 : WebAssemblyBinRMW<I32, "i32.atomic.rmw.xor", 0x3a>;
|
|
defm ATOMIC_RMW_XOR_I64 : WebAssemblyBinRMW<I64, "i64.atomic.rmw.xor", 0x3b>;
|
|
defm ATOMIC_RMW8_U_XOR_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.xor_u", 0x3c>;
|
|
defm ATOMIC_RMW16_U_XOR_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.xor_u", 0x3d>;
|
|
defm ATOMIC_RMW8_U_XOR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.xor_u", 0x3e>;
|
|
defm ATOMIC_RMW16_U_XOR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.xor_u", 0x3f>;
|
|
defm ATOMIC_RMW32_U_XOR_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.xor_u", 0x40>;
|
|
|
|
defm ATOMIC_RMW_XCHG_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw.xchg", 0x41>;
|
|
defm ATOMIC_RMW_XCHG_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw.xchg", 0x42>;
|
|
defm ATOMIC_RMW8_U_XCHG_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw8.xchg_u", 0x43>;
|
|
defm ATOMIC_RMW16_U_XCHG_I32 :
|
|
WebAssemblyBinRMW<I32, "i32.atomic.rmw16.xchg_u", 0x44>;
|
|
defm ATOMIC_RMW8_U_XCHG_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw8.xchg_u", 0x45>;
|
|
defm ATOMIC_RMW16_U_XCHG_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw16.xchg_u", 0x46>;
|
|
defm ATOMIC_RMW32_U_XCHG_I64 :
|
|
WebAssemblyBinRMW<I64, "i64.atomic.rmw32.xchg_u", 0x47>;
|
|
|
|
// Select binary RMWs with no constant offset.
|
|
multiclass BinRMWPatNoOffset<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind I32:$addr, ty:$val)),
|
|
(!cast<NI>(inst#_A32) 0, 0, I32:$addr, ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(ty (kind I64:$addr, ty:$val)),
|
|
(!cast<NI>(inst#_A64) 0, 0, I64:$addr, ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
|
|
// Select binary RMWs with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
multiclass BinRMWPatImmOff<ValueType ty, PatFrag kind, PatFrag operand,
|
|
string inst> {
|
|
def : Pat<(ty (kind (operand I32:$addr, imm:$off), ty:$val)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, I32:$addr, ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(ty (kind (operand I64:$addr, imm:$off), ty:$val)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, I64:$addr, ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
|
|
// Select binary RMWs with just a constant offset.
|
|
multiclass BinRMWPatOffsetOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind imm:$off, ty:$val)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, (CONST_I32 0), ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(ty (kind imm:$off, ty:$val)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, (CONST_I64 0), ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
|
|
multiclass BinRMWPatGlobalAddrOffOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind (WebAssemblywrapper tglobaladdr:$off), ty:$val)),
|
|
(!cast<NI>(inst#_A32) 0, tglobaladdr:$off, (CONST_I32 0), ty:$val)>,
|
|
Requires<[HasAddr32, HasAtomics, IsNotPIC]>;
|
|
def : Pat<(ty (kind (WebAssemblywrapper tglobaladdr:$off), ty:$val)),
|
|
(!cast<NI>(inst#_A64) 0, tglobaladdr:$off, (CONST_I64 0), ty:$val)>,
|
|
Requires<[HasAddr64, HasAtomics, IsNotPIC]>;
|
|
}
|
|
|
|
// Patterns for various addressing modes.
|
|
multiclass BinRMWPattern<PatFrag rmw_32, PatFrag rmw_64, string inst_32,
|
|
string inst_64> {
|
|
defm : BinRMWPatNoOffset<i32, rmw_32, inst_32>;
|
|
defm : BinRMWPatNoOffset<i64, rmw_64, inst_64>;
|
|
|
|
defm : BinRMWPatImmOff<i32, rmw_32, regPlusImm, inst_32>;
|
|
defm : BinRMWPatImmOff<i64, rmw_64, regPlusImm, inst_64>;
|
|
defm : BinRMWPatImmOff<i32, rmw_32, or_is_add, inst_32>;
|
|
defm : BinRMWPatImmOff<i64, rmw_64, or_is_add, inst_64>;
|
|
|
|
defm : BinRMWPatOffsetOnly<i32, rmw_32, inst_32>;
|
|
defm : BinRMWPatOffsetOnly<i64, rmw_64, inst_64>;
|
|
|
|
defm : BinRMWPatGlobalAddrOffOnly<i32, rmw_32, inst_32>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, rmw_64, inst_64>;
|
|
}
|
|
|
|
defm : BinRMWPattern<atomic_load_add_32, atomic_load_add_64,
|
|
"ATOMIC_RMW_ADD_I32", "ATOMIC_RMW_ADD_I64">;
|
|
defm : BinRMWPattern<atomic_load_sub_32, atomic_load_sub_64,
|
|
"ATOMIC_RMW_SUB_I32", "ATOMIC_RMW_SUB_I64">;
|
|
defm : BinRMWPattern<atomic_load_and_32, atomic_load_and_64,
|
|
"ATOMIC_RMW_AND_I32", "ATOMIC_RMW_AND_I64">;
|
|
defm : BinRMWPattern<atomic_load_or_32, atomic_load_or_64,
|
|
"ATOMIC_RMW_OR_I32", "ATOMIC_RMW_OR_I64">;
|
|
defm : BinRMWPattern<atomic_load_xor_32, atomic_load_xor_64,
|
|
"ATOMIC_RMW_XOR_I32", "ATOMIC_RMW_XOR_I64">;
|
|
defm : BinRMWPattern<atomic_swap_32, atomic_swap_64,
|
|
"ATOMIC_RMW_XCHG_I32", "ATOMIC_RMW_XCHG_I64">;
|
|
|
|
// Truncating & zero-extending binary RMW patterns.
|
|
// These are combined patterns of truncating store patterns and zero-extending
|
|
// load patterns above.
|
|
class zext_bin_rmw_8_32<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$val), (i32 (kind node:$addr, node:$val))>;
|
|
class zext_bin_rmw_16_32<PatFrag kind> : zext_bin_rmw_8_32<kind>;
|
|
class zext_bin_rmw_8_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$val),
|
|
(zext (i32 (kind node:$addr, (i32 (trunc (i64 node:$val))))))>;
|
|
class zext_bin_rmw_16_64<PatFrag kind> : zext_bin_rmw_8_64<kind>;
|
|
class zext_bin_rmw_32_64<PatFrag kind> : zext_bin_rmw_8_64<kind>;
|
|
|
|
// Truncating & sign-extending binary RMW patterns.
|
|
// These are combined patterns of truncating store patterns and sign-extending
|
|
// load patterns above. We match subword RMWs (for 32-bit) and anyext RMWs (for
|
|
// 64-bit) and select a zext RMW; the next instruction will be sext_inreg which
|
|
// is selected by itself.
|
|
class sext_bin_rmw_8_32<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$val), (kind node:$addr, node:$val)>;
|
|
class sext_bin_rmw_16_32<PatFrag kind> : sext_bin_rmw_8_32<kind>;
|
|
class sext_bin_rmw_8_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$val),
|
|
(anyext (i32 (kind node:$addr, (i32 (trunc (i64 node:$val))))))>;
|
|
class sext_bin_rmw_16_64<PatFrag kind> : sext_bin_rmw_8_64<kind>;
|
|
// 32->64 sext RMW gets selected as i32.atomic.rmw.***, i64.extend_i32_s
|
|
|
|
// Patterns for various addressing modes for truncating-extending binary RMWs.
|
|
multiclass BinRMWTruncExtPattern<
|
|
PatFrag rmw_8, PatFrag rmw_16, PatFrag rmw_32, PatFrag rmw_64,
|
|
string inst8_32, string inst16_32, string inst8_64, string inst16_64, string inst32_64> {
|
|
// Truncating-extending binary RMWs with no constant offset
|
|
defm : BinRMWPatNoOffset<i32, zext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatNoOffset<i32, zext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatNoOffset<i64, zext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatNoOffset<i64, zext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : BinRMWPatNoOffset<i64, zext_bin_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : BinRMWPatNoOffset<i32, sext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatNoOffset<i32, sext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatNoOffset<i64, sext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatNoOffset<i64, sext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
|
|
// Truncating-extending binary RMWs with a constant offset
|
|
defm : BinRMWPatImmOff<i32, zext_bin_rmw_8_32<rmw_8>, regPlusImm, inst8_32>;
|
|
defm : BinRMWPatImmOff<i32, zext_bin_rmw_16_32<rmw_16>, regPlusImm,
|
|
inst16_32>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_8_64<rmw_8>, regPlusImm, inst8_64>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_16_64<rmw_16>, regPlusImm,
|
|
inst16_64>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_32_64<rmw_32>, regPlusImm,
|
|
inst32_64>;
|
|
defm : BinRMWPatImmOff<i32, zext_bin_rmw_8_32<rmw_8>, or_is_add, inst8_32>;
|
|
defm : BinRMWPatImmOff<i32, zext_bin_rmw_16_32<rmw_16>, or_is_add, inst16_32>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_8_64<rmw_8>, or_is_add, inst8_64>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_16_64<rmw_16>, or_is_add, inst16_64>;
|
|
defm : BinRMWPatImmOff<i64, zext_bin_rmw_32_64<rmw_32>, or_is_add, inst32_64>;
|
|
|
|
defm : BinRMWPatImmOff<i32, sext_bin_rmw_8_32<rmw_8>, regPlusImm, inst8_32>;
|
|
defm : BinRMWPatImmOff<i32, sext_bin_rmw_16_32<rmw_16>, regPlusImm,
|
|
inst16_32>;
|
|
defm : BinRMWPatImmOff<i64, sext_bin_rmw_8_64<rmw_8>, regPlusImm, inst8_64>;
|
|
defm : BinRMWPatImmOff<i64, sext_bin_rmw_16_64<rmw_16>, regPlusImm,
|
|
inst16_64>;
|
|
defm : BinRMWPatImmOff<i32, sext_bin_rmw_8_32<rmw_8>, or_is_add, inst8_32>;
|
|
defm : BinRMWPatImmOff<i32, sext_bin_rmw_16_32<rmw_16>, or_is_add, inst16_32>;
|
|
defm : BinRMWPatImmOff<i64, sext_bin_rmw_8_64<rmw_8>, or_is_add, inst8_64>;
|
|
defm : BinRMWPatImmOff<i64, sext_bin_rmw_16_64<rmw_16>, or_is_add, inst16_64>;
|
|
|
|
// Truncating-extending binary RMWs with just a constant offset
|
|
defm : BinRMWPatOffsetOnly<i32, zext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatOffsetOnly<i32, zext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatOffsetOnly<i64, zext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatOffsetOnly<i64, zext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : BinRMWPatOffsetOnly<i64, zext_bin_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : BinRMWPatOffsetOnly<i32, sext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatOffsetOnly<i32, sext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatOffsetOnly<i64, sext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatOffsetOnly<i64, sext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
|
|
defm : BinRMWPatGlobalAddrOffOnly<i32, zext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i32, zext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, zext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, zext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, zext_bin_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : BinRMWPatGlobalAddrOffOnly<i32, sext_bin_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i32, sext_bin_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, sext_bin_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : BinRMWPatGlobalAddrOffOnly<i64, sext_bin_rmw_16_64<rmw_16>, inst16_64>;
|
|
}
|
|
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_load_add_8, atomic_load_add_16, atomic_load_add_32, atomic_load_add_64,
|
|
"ATOMIC_RMW8_U_ADD_I32", "ATOMIC_RMW16_U_ADD_I32",
|
|
"ATOMIC_RMW8_U_ADD_I64", "ATOMIC_RMW16_U_ADD_I64", "ATOMIC_RMW32_U_ADD_I64">;
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_load_sub_8, atomic_load_sub_16, atomic_load_sub_32, atomic_load_sub_64,
|
|
"ATOMIC_RMW8_U_SUB_I32", "ATOMIC_RMW16_U_SUB_I32",
|
|
"ATOMIC_RMW8_U_SUB_I64", "ATOMIC_RMW16_U_SUB_I64", "ATOMIC_RMW32_U_SUB_I64">;
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_load_and_8, atomic_load_and_16, atomic_load_and_32, atomic_load_and_64,
|
|
"ATOMIC_RMW8_U_AND_I32", "ATOMIC_RMW16_U_AND_I32",
|
|
"ATOMIC_RMW8_U_AND_I64", "ATOMIC_RMW16_U_AND_I64", "ATOMIC_RMW32_U_AND_I64">;
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_load_or_8, atomic_load_or_16, atomic_load_or_32, atomic_load_or_64,
|
|
"ATOMIC_RMW8_U_OR_I32", "ATOMIC_RMW16_U_OR_I32",
|
|
"ATOMIC_RMW8_U_OR_I64", "ATOMIC_RMW16_U_OR_I64", "ATOMIC_RMW32_U_OR_I64">;
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_load_xor_8, atomic_load_xor_16, atomic_load_xor_32, atomic_load_xor_64,
|
|
"ATOMIC_RMW8_U_XOR_I32", "ATOMIC_RMW16_U_XOR_I32",
|
|
"ATOMIC_RMW8_U_XOR_I64", "ATOMIC_RMW16_U_XOR_I64", "ATOMIC_RMW32_U_XOR_I64">;
|
|
defm : BinRMWTruncExtPattern<
|
|
atomic_swap_8, atomic_swap_16, atomic_swap_32, atomic_swap_64,
|
|
"ATOMIC_RMW8_U_XCHG_I32", "ATOMIC_RMW16_U_XCHG_I32",
|
|
"ATOMIC_RMW8_U_XCHG_I64", "ATOMIC_RMW16_U_XCHG_I64",
|
|
"ATOMIC_RMW32_U_XCHG_I64">;
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
// Atomic ternary read-modify-writes
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
// TODO LLVM IR's cmpxchg instruction returns a pair of {loaded value, success
|
|
// flag}. When we use the success flag or both values, we can't make use of i64
|
|
// truncate/extend versions of instructions for now, which is suboptimal.
|
|
// Consider adding a pass after instruction selection that optimizes this case
|
|
// if it is frequent.
|
|
|
|
multiclass WebAssemblyTerRMW<WebAssemblyRegClass rc, string name,
|
|
int atomic_op> {
|
|
defm "_A32" :
|
|
ATOMIC_I<(outs rc:$dst),
|
|
(ins P2Align:$p2align, offset32_op:$off, I32:$addr, rc:$exp,
|
|
rc:$new_),
|
|
(outs), (ins P2Align:$p2align, offset32_op:$off), [],
|
|
!strconcat(name, "\t$dst, ${off}(${addr})${p2align}, $exp, $new_"),
|
|
!strconcat(name, "\t${off}${p2align}"), atomic_op, "false">;
|
|
defm "_A64" :
|
|
ATOMIC_I<(outs rc:$dst),
|
|
(ins P2Align:$p2align, offset64_op:$off, I64:$addr, rc:$exp,
|
|
rc:$new_),
|
|
(outs), (ins P2Align:$p2align, offset64_op:$off), [],
|
|
!strconcat(name, "\t$dst, ${off}(${addr})${p2align}, $exp, $new_"),
|
|
!strconcat(name, "\t${off}${p2align}"), atomic_op, "true">;
|
|
}
|
|
|
|
defm ATOMIC_RMW_CMPXCHG_I32 :
|
|
WebAssemblyTerRMW<I32, "i32.atomic.rmw.cmpxchg", 0x48>;
|
|
defm ATOMIC_RMW_CMPXCHG_I64 :
|
|
WebAssemblyTerRMW<I64, "i64.atomic.rmw.cmpxchg", 0x49>;
|
|
defm ATOMIC_RMW8_U_CMPXCHG_I32 :
|
|
WebAssemblyTerRMW<I32, "i32.atomic.rmw8.cmpxchg_u", 0x4a>;
|
|
defm ATOMIC_RMW16_U_CMPXCHG_I32 :
|
|
WebAssemblyTerRMW<I32, "i32.atomic.rmw16.cmpxchg_u", 0x4b>;
|
|
defm ATOMIC_RMW8_U_CMPXCHG_I64 :
|
|
WebAssemblyTerRMW<I64, "i64.atomic.rmw8.cmpxchg_u", 0x4c>;
|
|
defm ATOMIC_RMW16_U_CMPXCHG_I64 :
|
|
WebAssemblyTerRMW<I64, "i64.atomic.rmw16.cmpxchg_u", 0x4d>;
|
|
defm ATOMIC_RMW32_U_CMPXCHG_I64 :
|
|
WebAssemblyTerRMW<I64, "i64.atomic.rmw32.cmpxchg_u", 0x4e>;
|
|
|
|
// Select ternary RMWs with no constant offset.
|
|
multiclass TerRMWPatNoOffset<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind I32:$addr, ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A32) 0, 0, I32:$addr, ty:$exp, ty:$new)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(ty (kind I64:$addr, ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A64) 0, 0, I64:$addr, ty:$exp, ty:$new)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
|
|
// Select ternary RMWs with a constant offset.
|
|
|
|
// Pattern with address + immediate offset
|
|
multiclass TerRMWPatImmOff<ValueType ty, PatFrag kind, PatFrag operand,
|
|
string inst> {
|
|
def : Pat<(ty (kind (operand I32:$addr, imm:$off), ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, I32:$addr, ty:$exp, ty:$new)>,
|
|
Requires<[HasAddr32, HasAtomics]>;
|
|
def : Pat<(ty (kind (operand I64:$addr, imm:$off), ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, I64:$addr, ty:$exp, ty:$new)>,
|
|
Requires<[HasAddr64, HasAtomics]>;
|
|
}
|
|
|
|
// Select ternary RMWs with just a constant offset.
|
|
multiclass TerRMWPatOffsetOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind imm:$off, ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A32) 0, imm:$off, (CONST_I32 0), ty:$exp,
|
|
ty:$new)>;
|
|
def : Pat<(ty (kind imm:$off, ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A64) 0, imm:$off, (CONST_I64 0), ty:$exp,
|
|
ty:$new)>;
|
|
}
|
|
|
|
multiclass TerRMWPatGlobalAddrOffOnly<ValueType ty, PatFrag kind, string inst> {
|
|
def : Pat<(ty (kind (WebAssemblywrapper tglobaladdr:$off), ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A32) 0, tglobaladdr:$off, (CONST_I32 0), ty:$exp,
|
|
ty:$new)>,
|
|
Requires<[HasAddr32, HasAtomics, IsNotPIC]>;
|
|
def : Pat<(ty (kind (WebAssemblywrapper tglobaladdr:$off), ty:$exp, ty:$new)),
|
|
(!cast<NI>(inst#_A64) 0, tglobaladdr:$off, (CONST_I64 0), ty:$exp,
|
|
ty:$new)>,
|
|
Requires<[HasAddr64, HasAtomics, IsNotPIC]>;
|
|
}
|
|
|
|
// Patterns for various addressing modes.
|
|
multiclass TerRMWPattern<PatFrag rmw_32, PatFrag rmw_64, string inst_32,
|
|
string inst_64> {
|
|
defm : TerRMWPatNoOffset<i32, rmw_32, inst_32>;
|
|
defm : TerRMWPatNoOffset<i64, rmw_64, inst_64>;
|
|
|
|
defm : TerRMWPatImmOff<i32, rmw_32, regPlusImm, inst_32>;
|
|
defm : TerRMWPatImmOff<i64, rmw_64, regPlusImm, inst_64>;
|
|
defm : TerRMWPatImmOff<i32, rmw_32, or_is_add, inst_32>;
|
|
defm : TerRMWPatImmOff<i64, rmw_64, or_is_add, inst_64>;
|
|
|
|
defm : TerRMWPatOffsetOnly<i32, rmw_32, inst_32>;
|
|
defm : TerRMWPatOffsetOnly<i64, rmw_64, inst_64>;
|
|
|
|
defm : TerRMWPatGlobalAddrOffOnly<i32, rmw_32, inst_32>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, rmw_64, inst_64>;
|
|
}
|
|
|
|
defm : TerRMWPattern<atomic_cmp_swap_32, atomic_cmp_swap_64,
|
|
"ATOMIC_RMW_CMPXCHG_I32", "ATOMIC_RMW_CMPXCHG_I64">;
|
|
|
|
// Truncating & zero-extending ternary RMW patterns.
|
|
// DAG legalization & optimization before instruction selection may introduce
|
|
// additional nodes such as anyext or assertzext depending on operand types.
|
|
class zext_ter_rmw_8_32<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$exp, node:$new),
|
|
(i32 (kind node:$addr, node:$exp, node:$new))>;
|
|
class zext_ter_rmw_16_32<PatFrag kind> : zext_ter_rmw_8_32<kind>;
|
|
class zext_ter_rmw_8_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$exp, node:$new),
|
|
(zext (i32 (assertzext (i32 (kind node:$addr,
|
|
(i32 (trunc (i64 node:$exp))),
|
|
(i32 (trunc (i64 node:$new))))))))>;
|
|
class zext_ter_rmw_16_64<PatFrag kind> : zext_ter_rmw_8_64<kind>;
|
|
class zext_ter_rmw_32_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$exp, node:$new),
|
|
(zext (i32 (kind node:$addr,
|
|
(i32 (trunc (i64 node:$exp))),
|
|
(i32 (trunc (i64 node:$new))))))>;
|
|
|
|
// Truncating & sign-extending ternary RMW patterns.
|
|
// We match subword RMWs (for 32-bit) and anyext RMWs (for 64-bit) and select a
|
|
// zext RMW; the next instruction will be sext_inreg which is selected by
|
|
// itself.
|
|
class sext_ter_rmw_8_32<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$exp, node:$new),
|
|
(kind node:$addr, node:$exp, node:$new)>;
|
|
class sext_ter_rmw_16_32<PatFrag kind> : sext_ter_rmw_8_32<kind>;
|
|
class sext_ter_rmw_8_64<PatFrag kind> :
|
|
PatFrag<(ops node:$addr, node:$exp, node:$new),
|
|
(anyext (i32 (assertzext (i32
|
|
(kind node:$addr,
|
|
(i32 (trunc (i64 node:$exp))),
|
|
(i32 (trunc (i64 node:$new))))))))>;
|
|
class sext_ter_rmw_16_64<PatFrag kind> : sext_ter_rmw_8_64<kind>;
|
|
// 32->64 sext RMW gets selected as i32.atomic.rmw.***, i64.extend_i32_s
|
|
|
|
// Patterns for various addressing modes for truncating-extending ternary RMWs.
|
|
multiclass TerRMWTruncExtPattern<
|
|
PatFrag rmw_8, PatFrag rmw_16, PatFrag rmw_32, PatFrag rmw_64,
|
|
string inst8_32, string inst16_32, string inst8_64, string inst16_64,
|
|
string inst32_64> {
|
|
// Truncating-extending ternary RMWs with no constant offset
|
|
defm : TerRMWPatNoOffset<i32, zext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatNoOffset<i32, zext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatNoOffset<i64, zext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatNoOffset<i64, zext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : TerRMWPatNoOffset<i64, zext_ter_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : TerRMWPatNoOffset<i32, sext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatNoOffset<i32, sext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatNoOffset<i64, sext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatNoOffset<i64, sext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
|
|
// Truncating-extending ternary RMWs with a constant offset
|
|
defm : TerRMWPatImmOff<i32, zext_ter_rmw_8_32<rmw_8>, regPlusImm, inst8_32>;
|
|
defm : TerRMWPatImmOff<i32, zext_ter_rmw_16_32<rmw_16>, regPlusImm,
|
|
inst16_32>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_8_64<rmw_8>, regPlusImm, inst8_64>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_16_64<rmw_16>, regPlusImm,
|
|
inst16_64>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_32_64<rmw_32>, regPlusImm,
|
|
inst32_64>;
|
|
defm : TerRMWPatImmOff<i32, zext_ter_rmw_8_32<rmw_8>, or_is_add, inst8_32>;
|
|
defm : TerRMWPatImmOff<i32, zext_ter_rmw_16_32<rmw_16>, or_is_add, inst16_32>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_8_64<rmw_8>, or_is_add, inst8_64>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_16_64<rmw_16>, or_is_add, inst16_64>;
|
|
defm : TerRMWPatImmOff<i64, zext_ter_rmw_32_64<rmw_32>, or_is_add, inst32_64>;
|
|
|
|
defm : TerRMWPatImmOff<i32, sext_ter_rmw_8_32<rmw_8>, regPlusImm, inst8_32>;
|
|
defm : TerRMWPatImmOff<i32, sext_ter_rmw_16_32<rmw_16>, regPlusImm,
|
|
inst16_32>;
|
|
defm : TerRMWPatImmOff<i64, sext_ter_rmw_8_64<rmw_8>, regPlusImm, inst8_64>;
|
|
defm : TerRMWPatImmOff<i64, sext_ter_rmw_16_64<rmw_16>, regPlusImm,
|
|
inst16_64>;
|
|
defm : TerRMWPatImmOff<i32, sext_ter_rmw_8_32<rmw_8>, or_is_add, inst8_32>;
|
|
defm : TerRMWPatImmOff<i32, sext_ter_rmw_16_32<rmw_16>, or_is_add, inst16_32>;
|
|
defm : TerRMWPatImmOff<i64, sext_ter_rmw_8_64<rmw_8>, or_is_add, inst8_64>;
|
|
defm : TerRMWPatImmOff<i64, sext_ter_rmw_16_64<rmw_16>, or_is_add, inst16_64>;
|
|
|
|
// Truncating-extending ternary RMWs with just a constant offset
|
|
defm : TerRMWPatOffsetOnly<i32, zext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatOffsetOnly<i32, zext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatOffsetOnly<i64, zext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatOffsetOnly<i64, zext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : TerRMWPatOffsetOnly<i64, zext_ter_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : TerRMWPatOffsetOnly<i32, sext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatOffsetOnly<i32, sext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatOffsetOnly<i64, sext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatOffsetOnly<i64, sext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
|
|
defm : TerRMWPatGlobalAddrOffOnly<i32, zext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i32, zext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, zext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, zext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, zext_ter_rmw_32_64<rmw_32>, inst32_64>;
|
|
|
|
defm : TerRMWPatGlobalAddrOffOnly<i32, sext_ter_rmw_8_32<rmw_8>, inst8_32>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i32, sext_ter_rmw_16_32<rmw_16>, inst16_32>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, sext_ter_rmw_8_64<rmw_8>, inst8_64>;
|
|
defm : TerRMWPatGlobalAddrOffOnly<i64, sext_ter_rmw_16_64<rmw_16>, inst16_64>;
|
|
}
|
|
|
|
defm : TerRMWTruncExtPattern<
|
|
atomic_cmp_swap_8, atomic_cmp_swap_16, atomic_cmp_swap_32, atomic_cmp_swap_64,
|
|
"ATOMIC_RMW8_U_CMPXCHG_I32", "ATOMIC_RMW16_U_CMPXCHG_I32",
|
|
"ATOMIC_RMW8_U_CMPXCHG_I64", "ATOMIC_RMW16_U_CMPXCHG_I64",
|
|
"ATOMIC_RMW32_U_CMPXCHG_I64">;
|