256 lines · plain
1; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mcpu=mvp -mattr=+bulk-memory | FileCheck %s --check-prefixes CHECK,BULK-MEM2; RUN: llc < %s -asm-verbose=false -verify-machineinstrs -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mcpu=mvp -mattr=-bulk-memory | FileCheck %s --check-prefixes CHECK,NO-BULK-MEM3 4; Test that basic bulk memory codegen works correctly5 6target triple = "wasm32-unknown-unknown"7 8declare void @llvm.memcpy.p0.p0.i8(ptr, ptr, i8, i1)9declare void @llvm.memcpy.p0.p0.i32(ptr, ptr, i32, i1)10 11declare void @llvm.memmove.p0.p0.i8(ptr, ptr, i8, i1)12declare void @llvm.memmove.p0.p0.i32(ptr, ptr, i32, i1)13 14declare void @llvm.memset.p0.i8(ptr, i8, i8, i1)15declare void @llvm.memset.p0.i32(ptr, i8, i32, i1)16 17; CHECK-LABEL: memcpy_i8:18; NO-BULK-MEM-NOT: memory.copy19; BULK-MEM-NEXT: .functype memcpy_i8 (i32, i32, i32) -> ()20; BULK-MEM-NEXT: block21; BULK-MEM-NEXT: i32.eqz $push0=, $222; BULK-MEM-NEXT: br_if 0, $pop023; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $224; BULK-MEM-NEXT: .LBB{{.*}}:25; BULK-MEM-NEXT: end_block26; BULK-MEM-NEXT: return27define void @memcpy_i8(ptr %dest, ptr %src, i8 zeroext %len) {28 call void @llvm.memcpy.p0.p0.i8(ptr %dest, ptr %src, i8 %len, i1 0)29 ret void30}31 32; CHECK-LABEL: memmove_i8:33; NO-BULK-MEM-NOT: memory.copy34; BULK-MEM-NEXT: .functype memmove_i8 (i32, i32, i32) -> ()35; BULK-MEM-NEXT: block36; BULK-MEM-NEXT: i32.eqz $push0=, $237; BULK-MEM-NEXT: br_if 0, $pop038; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $239; BULK-MEM-NEXT: .LBB{{.*}}:40; BULK-MEM-NEXT: end_block41; BULK-MEM-NEXT: return42define void @memmove_i8(ptr %dest, ptr %src, i8 zeroext %len) {43 call void @llvm.memmove.p0.p0.i8(ptr %dest, ptr %src, i8 %len, i1 0)44 ret void45}46 47; CHECK-LABEL: memset_i8:48; NO-BULK-MEM-NOT: memory.fill49; BULK-MEM-NEXT: .functype memset_i8 (i32, i32, i32) -> ()50; BULK-MEM-NEXT: block51; BULK-MEM-NEXT: i32.eqz $push0=, $252; BULK-MEM-NEXT: br_if 0, $pop053; BULK-MEM-NEXT: memory.fill 0, $0, $1, $254; BULK-MEM-NEXT: .LBB{{.*}}:55; BULK-MEM-NEXT: end_block56; BULK-MEM-NEXT: return57define void @memset_i8(ptr %dest, i8 %val, i8 zeroext %len) {58 call void @llvm.memset.p0.i8(ptr %dest, i8 %val, i8 %len, i1 0)59 ret void60}61 62; CHECK-LABEL: memcpy_i32:63; NO-BULK-MEM-NOT: memory.copy64; BULK-MEM-NEXT: .functype memcpy_i32 (i32, i32, i32) -> ()65; BULK-MEM-NEXT: block66; BULK-MEM-NEXT: i32.eqz $push0=, $267; BULK-MEM-NEXT: br_if 0, $pop068; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $269; BULK-MEM-NEXT: .LBB{{.*}}:70; BULK-MEM-NEXT: end_block71; BULK-MEM-NEXT: return72define void @memcpy_i32(ptr %dest, ptr %src, i32 %len) {73 call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 %len, i1 0)74 ret void75}76 77; CHECK-LABEL: memmove_i32:78; NO-BULK-MEM-NOT: memory.copy79; BULK-MEM-NEXT: .functype memmove_i32 (i32, i32, i32) -> ()80; BULK-MEM-NEXT: block81; BULK-MEM-NEXT: i32.eqz $push0=, $282; BULK-MEM-NEXT: br_if 0, $pop083; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $284; BULK-MEM-NEXT: .LBB{{.*}}:85; BULK-MEM-NEXT: end_block86; BULK-MEM-NEXT: return87define void @memmove_i32(ptr %dest, ptr %src, i32 %len) {88 call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 %len, i1 0)89 ret void90}91 92; CHECK-LABEL: memset_i32:93; NO-BULK-MEM-NOT: memory.fill94; BULK-MEM-NEXT: .functype memset_i32 (i32, i32, i32) -> ()95; BULK-MEM-NEXT: block96; BULK-MEM-NEXT: i32.eqz $push0=, $297; BULK-MEM-NEXT: br_if 0, $pop098; BULK-MEM-NEXT: memory.fill 0, $0, $1, $299; BULK-MEM-NEXT: .LBB{{.*}}:100; BULK-MEM-NEXT: end_block101; BULK-MEM-NEXT: return102define void @memset_i32(ptr %dest, i8 %val, i32 %len) {103 call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 %len, i1 0)104 ret void105}106 107; CHECK-LABEL: memcpy_0:108; CHECK-NEXT: .functype memcpy_0 (i32, i32) -> ()109; CHECK-NEXT: return110define void @memcpy_0(ptr %dest, ptr %src) {111 call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 0, i1 0)112 ret void113}114 115; CHECK-LABEL: memmove_0:116; CHECK-NEXT: .functype memmove_0 (i32, i32) -> ()117; CHECK-NEXT: return118define void @memmove_0(ptr %dest, ptr %src) {119 call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 0, i1 0)120 ret void121}122 123; CHECK-LABEL: memset_0:124; NO-BULK-MEM-NOT: memory.fill125; BULK-MEM-NEXT: .functype memset_0 (i32, i32) -> ()126; BULK-MEM-NEXT: return127define void @memset_0(ptr %dest, i8 %val) {128 call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 0, i1 0)129 ret void130}131 132; CHECK-LABEL: memcpy_1:133; CHECK-NEXT: .functype memcpy_1 (i32, i32) -> ()134; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)135; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]136; CHECK-NEXT: return137define void @memcpy_1(ptr %dest, ptr %src) {138 call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 1, i1 0)139 ret void140}141 142; CHECK-LABEL: memmove_1:143; CHECK-NEXT: .functype memmove_1 (i32, i32) -> ()144; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)145; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]146; CHECK-NEXT: return147define void @memmove_1(ptr %dest, ptr %src) {148 call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 1, i1 0)149 ret void150}151 152; CHECK-LABEL: memset_1:153; NO-BULK-MEM-NOT: memory.fill154; BULK-MEM-NEXT: .functype memset_1 (i32, i32) -> ()155; BULK-MEM-NEXT: i32.store8 0($0), $1156; BULK-MEM-NEXT: return157define void @memset_1(ptr %dest, i8 %val) {158 call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 1, i1 0)159 ret void160}161 162; CHECK-LABEL: memcpy_1024:163; NO-BULK-MEM-NOT: memory.copy164; BULK-MEM-NEXT: .functype memcpy_1024 (i32, i32) -> ()165; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024166; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]167; BULK-MEM-NEXT: return168define void @memcpy_1024(ptr %dest, ptr %src) {169 call void @llvm.memcpy.p0.p0.i32(ptr %dest, ptr %src, i32 1024, i1 0)170 ret void171}172 173; CHECK-LABEL: memmove_1024:174; NO-BULK-MEM-NOT: memory.copy175; BULK-MEM-NEXT: .functype memmove_1024 (i32, i32) -> ()176; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024177; BULK-MEM-NEXT: memory.copy 0, 0, $0, $1, $pop[[L0]]178; BULK-MEM-NEXT: return179define void @memmove_1024(ptr %dest, ptr %src) {180 call void @llvm.memmove.p0.p0.i32(ptr %dest, ptr %src, i32 1024, i1 0)181 ret void182}183 184; CHECK-LABEL: memset_1024:185; NO-BULK-MEM-NOT: memory.fill186; BULK-MEM-NEXT: .functype memset_1024 (i32, i32) -> ()187; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024188; BULK-MEM-NEXT: memory.fill 0, $0, $1, $pop[[L0]]189; BULK-MEM-NEXT: return190define void @memset_1024(ptr %dest, i8 %val) {191 call void @llvm.memset.p0.i32(ptr %dest, i8 %val, i32 1024, i1 0)192 ret void193}194 195; The following tests check that frame index elimination works for196; bulk memory instructions. The stack pointer is bumped by 112 instead197; of 100 because the stack pointer in WebAssembly is currently always198; 16-byte aligned, even in leaf functions, although it is not written199; back to the global in this case.200 201; TODO: Change TransientStackAlignment to 1 to avoid this extra202; arithmetic. This will require forcing the use of StackAlignment in203; PrologEpilogEmitter.cpp when204; WebAssemblyFrameLowering::needsSPWriteback would be true.205 206; CHECK-LABEL: memcpy_alloca_src:207; NO-BULK-MEM-NOT: memory.copy208; BULK-MEM-NEXT: .functype memcpy_alloca_src (i32) -> ()209; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer210; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112211; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]212; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 12213; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]214; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 100215; BULK-MEM-NEXT: memory.copy 0, 0, $0, $pop[[L4]], $pop[[L5]]216; BULK-MEM-NEXT: return217define void @memcpy_alloca_src(ptr %dst) {218 %a = alloca [100 x i8]219 call void @llvm.memcpy.p0.p0.i32(ptr %dst, ptr %a, i32 100, i1 false)220 ret void221}222 223; CHECK-LABEL: memcpy_alloca_dst:224; NO-BULK-MEM-NOT: memory.copy225; BULK-MEM-NEXT: .functype memcpy_alloca_dst (i32) -> ()226; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer227; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112228; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]229; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 12230; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]231; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 100232; BULK-MEM-NEXT: memory.copy 0, 0, $pop[[L4]], $0, $pop[[L5]]233; BULK-MEM-NEXT: return234define void @memcpy_alloca_dst(ptr %src) {235 %a = alloca [100 x i8]236 call void @llvm.memcpy.p0.p0.i32(ptr %a, ptr %src, i32 100, i1 false)237 ret void238}239 240; CHECK-LABEL: memset_alloca:241; NO-BULK-MEM-NOT: memory.fill242; BULK-MEM-NEXT: .functype memset_alloca (i32) -> ()243; BULK-MEM-NEXT: global.get $push[[L0:[0-9]+]]=, __stack_pointer244; BULK-MEM-NEXT: i32.const $push[[L1:[0-9]+]]=, 112245; BULK-MEM-NEXT: i32.sub $push[[L2:[0-9]+]]=, $pop[[L0]], $pop[[L1]]246; BULK-MEM-NEXT: i32.const $push[[L3:[0-9]+]]=, 12247; BULK-MEM-NEXT: i32.add $push[[L4:[0-9]+]]=, $pop[[L2]], $pop[[L3]]248; BULK-MEM-NEXT: i32.const $push[[L5:[0-9]+]]=, 100249; BULK-MEM-NEXT: memory.fill 0, $pop[[L4]], $0, $pop[[L5]]250; BULK-MEM-NEXT: return251define void @memset_alloca(i8 %val) {252 %a = alloca [100 x i8]253 call void @llvm.memset.p0.i32(ptr %a, i8 %val, i32 100, i1 false)254 ret void255}256