mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2025-02-01 13:11:39 +01:00
f98161801a
Summary: Uses the named operands tablegen feature to look up the indices of offset, address, and p2align operands for all load and store instructions. This replaces brittle, incorrect logic for identifying loads and store when eliminating frame indices, which previously crashed on bulk-memory ops. It also cleans up the SetP2Alignment pass. Reviewers: aheejin, dschuff Subscribers: sbc100, jgravelle-google, hiraditya, sunfish, jfb, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D59007 llvm-svn: 355770
208 lines
7.8 KiB
LLVM
208 lines
7.8 KiB
LLVM
; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+bulk-memory | FileCheck %s --check-prefixes CHECK,BULK-MEM
|
|
; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=-bulk-memory | FileCheck %s --check-prefixes CHECK,NO-BULK-MEM
|
|
|
|
; Test that basic bulk memory codegen works correctly
|
|
|
|
target datalayout = "e-m:e-p:32:32-i64:64-n32:64-S128"
|
|
target triple = "wasm32-unknown-unknown"
|
|
|
|
declare void @llvm.memcpy.p0i8.p0i8.i8(i8*, i8*, i8, i1)
|
|
declare void @llvm.memcpy.p0i8.p0i8.i32(i8*, i8*, i32, i1)
|
|
declare void @llvm.memcpy.p0i32.p0i32.i32(i32*, i32*, i32, i1)
|
|
|
|
declare void @llvm.memmove.p0i8.p0i8.i8(i8*, i8*, i8, i1)
|
|
declare void @llvm.memmove.p0i8.p0i8.i32(i8*, i8*, i32, i1)
|
|
declare void @llvm.memmove.p0i32.p0i32.i32(i32*, i32*, i32, i1)
|
|
|
|
declare void @llvm.memset.p0i8.i8(i8*, i8, i8, i1)
|
|
declare void @llvm.memset.p0i8.i32(i8*, i8, i32, i1)
|
|
declare void @llvm.memset.p0i32.i32(i32*, i8, i32, i1)
|
|
|
|
; CHECK-LABEL: memcpy_i8:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memcpy_i8 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memcpy_i8(i8* %dest, i8* %src, i8 zeroext %len) {
|
|
call void @llvm.memcpy.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memmove_i8:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memmove_i8 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memmove_i8(i8* %dest, i8* %src, i8 zeroext %len) {
|
|
call void @llvm.memmove.p0i8.p0i8.i8(i8* %dest, i8* %src, i8 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memset_i8:
|
|
; NO-BULK-MEM-NOT: memory.fill
|
|
; BULK-MEM-NEXT: .functype memset_i8 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.fill 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memset_i8(i8* %dest, i8 %val, i8 zeroext %len) {
|
|
call void @llvm.memset.p0i8.i8(i8* %dest, i8 %val, i8 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memcpy_i32:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memcpy_i32 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memcpy_i32(i32* %dest, i32* %src, i32 %len) {
|
|
call void @llvm.memcpy.p0i32.p0i32.i32(i32* %dest, i32* %src, i32 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memmove_i32:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memmove_i32 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memmove_i32(i32* %dest, i32* %src, i32 %len) {
|
|
call void @llvm.memmove.p0i32.p0i32.i32(i32* %dest, i32* %src, i32 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memset_i32:
|
|
; NO-BULK-MEM-NOT: memory.fill
|
|
; BULK-MEM-NEXT: .functype memset_i32 (i32, i32, i32) -> ()
|
|
; BULK-MEM-NEXT: memory.fill 0, $0, $1, $2
|
|
; BULK-MEM-NEXT: return
|
|
define void @memset_i32(i32* %dest, i8 %val, i32 %len) {
|
|
call void @llvm.memset.p0i32.i32(i32* %dest, i8 %val, i32 %len, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memcpy_1:
|
|
; CHECK-NEXT: .functype memcpy_1 (i32, i32) -> ()
|
|
; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
|
|
; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
|
|
; CHECK-NEXT: return
|
|
define void @memcpy_1(i8* %dest, i8* %src) {
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memmove_1:
|
|
; CHECK-NEXT: .functype memmove_1 (i32, i32) -> ()
|
|
; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
|
|
; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
|
|
; CHECK-NEXT: return
|
|
define void @memmove_1(i8* %dest, i8* %src) {
|
|
call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memset_1:
|
|
; NO-BULK-MEM-NOT: memory.fill
|
|
; BULK-MEM-NEXT: .functype memset_1 (i32, i32) -> ()
|
|
; BULK-MEM-NEXT: i32.store8 0($0), $1
|
|
; BULK-MEM-NEXT: return
|
|
define void @memset_1(i8* %dest, i8 %val) {
|
|
call void @llvm.memset.p0i8.i32(i8* %dest, i8 %val, i32 1, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memcpy_1024:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memcpy_1024 (i32, i32) -> ()
|
|
; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memcpy_1024(i8* %dest, i8* %src) {
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1024, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memmove_1024:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memmove_1024 (i32, i32) -> ()
|
|
; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memmove_1024(i8* %dest, i8* %src) {
|
|
call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1024, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memset_1024:
|
|
; NO-BULK-MEM-NOT: memory.fill
|
|
; BULK-MEM-NEXT: .functype memset_1024 (i32, i32) -> ()
|
|
; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
|
|
; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L0]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memset_1024(i8* %dest, i8 %val) {
|
|
call void @llvm.memset.p0i8.i32(i8* %dest, i8 %val, i32 1024, i1 0)
|
|
ret void
|
|
}
|
|
|
|
; The following tests check that frame index elimination works for
|
|
; bulk memory instructions. The stack pointer is bumped by 16 instead
|
|
; of 10 because the stack pointer in WebAssembly is currently always
|
|
; 16-byte aligned, even in leaf functions, although it is not written
|
|
; back to the global in this case.
|
|
|
|
; TODO: Change TransientStackAlignment to 1 to avoid this extra
|
|
; arithmetic. This will require forcing the use of StackAlignment in
|
|
; PrologEpilogEmitter.cpp when
|
|
; WebAssemblyFrameLowering::needsSPWriteback would be true.
|
|
|
|
; CHECK-LABEL: memcpy_alloca_src:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memcpy_alloca_src (i32) -> ()
|
|
; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
|
|
; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 16
|
|
; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 6
|
|
; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 10
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L4]], $pop[[L5]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memcpy_alloca_src(i8* %dst) {
|
|
%a = alloca [10 x i8]
|
|
%p = bitcast [10 x i8]* %a to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %p, i32 10, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memcpy_alloca_dst:
|
|
; NO-BULK-MEM-NOT: memory.copy
|
|
; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i32) -> ()
|
|
; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
|
|
; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 16
|
|
; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 6
|
|
; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 10
|
|
; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L4]], $0, $pop[[L5]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memcpy_alloca_dst(i8* %src) {
|
|
%a = alloca [10 x i8]
|
|
%p = bitcast [10 x i8]* %a to i8*
|
|
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %p, i8* %src, i32 10, i1 false)
|
|
ret void
|
|
}
|
|
|
|
; CHECK-LABEL: memset_alloca:
|
|
; NO-BULK-MEM-NOT: memory.fill
|
|
; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()
|
|
; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer
|
|
; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 16
|
|
; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 6
|
|
; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]
|
|
; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 10
|
|
; BULK-MEM-NEXT: memory.fill 0, $pop[[L4]], $0, $pop[[L5]]
|
|
; BULK-MEM-NEXT: return
|
|
define void @memset_alloca(i8 %val) {
|
|
%a = alloca [10 x i8]
|
|
%p = bitcast [10 x i8]* %a to i8*
|
|
call void @llvm.memset.p0i8.i32(i8* %p, i8 %val, i32 10, i1 false)
|
|
ret void
|
|
}
|