178 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 42 3; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -verify-each -atomic-expand %s | FileCheck -check-prefix=GFX90A %s4; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 -verify-each -atomic-expand %s | FileCheck -check-prefix=GFX1100 %s5 6; Contains a variety of tests with different types of atomic expansions to check that MMRAs are7; preserved.8 9define i16 @test_atomicrmw_xchg_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {10; GFX90A-LABEL: define i16 @test_atomicrmw_xchg_i16_global_agent(11; GFX90A-SAME: ptr addrspace(1) [[PTR:%.*]], i16 [[VALUE:%.*]]) #[[ATTR0:[0-9]+]] {12; GFX90A-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR]], i64 -4)13; GFX90A-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i6414; GFX90A-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 315; GFX90A-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 316; GFX90A-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i3217; GFX90A-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]18; GFX90A-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -119; GFX90A-NEXT: [[TMP3:%.*]] = zext i16 [[VALUE]] to i3220; GFX90A-NEXT: [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]21; GFX90A-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4, !mmra [[META0:![0-9]+]]22; GFX90A-NEXT: br label [[ATOMICRMW_START:%.*]]23; GFX90A: atomicrmw.start:24; GFX90A-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]25; GFX90A-NEXT: [[TMP5:%.*]] = and i32 [[LOADED]], [[INV_MASK]]26; GFX90A-NEXT: [[TMP6:%.*]] = or i32 [[TMP5]], [[VALOPERAND_SHIFTED]]27; GFX90A-NEXT: [[TMP7:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] syncscope("agent") seq_cst seq_cst, align 4, !mmra [[META0]]28; GFX90A-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP7]], 129; GFX90A-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP7]], 030; GFX90A-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]31; GFX90A: atomicrmw.end:32; GFX90A-NEXT: [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]33; GFX90A-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i1634; GFX90A-NEXT: ret i16 [[EXTRACTED]]35;36; GFX1100-LABEL: define i16 @test_atomicrmw_xchg_i16_global_agent(37; GFX1100-SAME: ptr addrspace(1) [[PTR:%.*]], i16 [[VALUE:%.*]]) #[[ATTR0:[0-9]+]] {38; GFX1100-NEXT: [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR]], i64 -4)39; GFX1100-NEXT: [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i6440; GFX1100-NEXT: [[PTRLSB:%.*]] = and i64 [[TMP1]], 341; GFX1100-NEXT: [[TMP2:%.*]] = shl i64 [[PTRLSB]], 342; GFX1100-NEXT: [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i3243; GFX1100-NEXT: [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]44; GFX1100-NEXT: [[INV_MASK:%.*]] = xor i32 [[MASK]], -145; GFX1100-NEXT: [[TMP3:%.*]] = zext i16 [[VALUE]] to i3246; GFX1100-NEXT: [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]47; GFX1100-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4, !mmra [[META0:![0-9]+]]48; GFX1100-NEXT: br label [[ATOMICRMW_START:%.*]]49; GFX1100: atomicrmw.start:50; GFX1100-NEXT: [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]51; GFX1100-NEXT: [[TMP5:%.*]] = and i32 [[LOADED]], [[INV_MASK]]52; GFX1100-NEXT: [[TMP6:%.*]] = or i32 [[TMP5]], [[VALOPERAND_SHIFTED]]53; GFX1100-NEXT: [[TMP7:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] syncscope("agent") seq_cst seq_cst, align 4, !mmra [[META0]]54; GFX1100-NEXT: [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP7]], 155; GFX1100-NEXT: [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP7]], 056; GFX1100-NEXT: br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]57; GFX1100: atomicrmw.end:58; GFX1100-NEXT: [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]59; GFX1100-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i1660; GFX1100-NEXT: ret i16 [[EXTRACTED]]61;62 %res = atomicrmw xchg ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !mmra !263 ret i16 %res64}65 66define i16 @test_cmpxchg_i16_global_agent_align4(ptr addrspace(1) %out, i16 %in, i16 %old) {67; GFX90A-LABEL: define i16 @test_cmpxchg_i16_global_agent_align4(68; GFX90A-SAME: ptr addrspace(1) [[OUT:%.*]], i16 [[IN:%.*]], i16 [[OLD:%.*]]) #[[ATTR0]] {69; GFX90A-NEXT: [[GEP:%.*]] = getelementptr i16, ptr addrspace(1) [[OUT]], i64 470; GFX90A-NEXT: [[TMP1:%.*]] = zext i16 [[IN]] to i3271; GFX90A-NEXT: [[TMP2:%.*]] = zext i16 [[OLD]] to i3272; GFX90A-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(1) [[GEP]], align 4, !mmra [[META0]]73; GFX90A-NEXT: [[TMP4:%.*]] = and i32 [[TMP3]], -6553674; GFX90A-NEXT: br label [[PARTWORD_CMPXCHG_LOOP:%.*]]75; GFX90A: partword.cmpxchg.loop:76; GFX90A-NEXT: [[TMP5:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[TMP11:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]77; GFX90A-NEXT: [[TMP6:%.*]] = or i32 [[TMP5]], [[TMP1]]78; GFX90A-NEXT: [[TMP7:%.*]] = or i32 [[TMP5]], [[TMP2]]79; GFX90A-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[GEP]], i32 [[TMP7]], i32 [[TMP6]] seq_cst seq_cst, align 4, !mmra [[META0]]80; GFX90A-NEXT: [[TMP9:%.*]] = extractvalue { i32, i1 } [[TMP8]], 081; GFX90A-NEXT: [[TMP10:%.*]] = extractvalue { i32, i1 } [[TMP8]], 182; GFX90A-NEXT: br i1 [[TMP10]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]83; GFX90A: partword.cmpxchg.failure:84; GFX90A-NEXT: [[TMP11]] = and i32 [[TMP9]], -6553685; GFX90A-NEXT: [[TMP12:%.*]] = icmp ne i32 [[TMP5]], [[TMP11]]86; GFX90A-NEXT: br i1 [[TMP12]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]87; GFX90A: partword.cmpxchg.end:88; GFX90A-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[TMP9]] to i1689; GFX90A-NEXT: [[TMP13:%.*]] = insertvalue { i16, i1 } poison, i16 [[EXTRACTED]], 090; GFX90A-NEXT: [[TMP14:%.*]] = insertvalue { i16, i1 } [[TMP13]], i1 [[TMP10]], 191; GFX90A-NEXT: [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP14]], 092; GFX90A-NEXT: ret i16 [[EXTRACT]]93;94; GFX1100-LABEL: define i16 @test_cmpxchg_i16_global_agent_align4(95; GFX1100-SAME: ptr addrspace(1) [[OUT:%.*]], i16 [[IN:%.*]], i16 [[OLD:%.*]]) #[[ATTR0]] {96; GFX1100-NEXT: [[GEP:%.*]] = getelementptr i16, ptr addrspace(1) [[OUT]], i64 497; GFX1100-NEXT: [[TMP1:%.*]] = zext i16 [[IN]] to i3298; GFX1100-NEXT: [[TMP2:%.*]] = zext i16 [[OLD]] to i3299; GFX1100-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(1) [[GEP]], align 4, !mmra [[META0]]100; GFX1100-NEXT: [[TMP4:%.*]] = and i32 [[TMP3]], -65536101; GFX1100-NEXT: br label [[PARTWORD_CMPXCHG_LOOP:%.*]]102; GFX1100: partword.cmpxchg.loop:103; GFX1100-NEXT: [[TMP5:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[TMP11:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]104; GFX1100-NEXT: [[TMP6:%.*]] = or i32 [[TMP5]], [[TMP1]]105; GFX1100-NEXT: [[TMP7:%.*]] = or i32 [[TMP5]], [[TMP2]]106; GFX1100-NEXT: [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[GEP]], i32 [[TMP7]], i32 [[TMP6]] seq_cst seq_cst, align 4, !mmra [[META0]]107; GFX1100-NEXT: [[TMP9:%.*]] = extractvalue { i32, i1 } [[TMP8]], 0108; GFX1100-NEXT: [[TMP10:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1109; GFX1100-NEXT: br i1 [[TMP10]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]110; GFX1100: partword.cmpxchg.failure:111; GFX1100-NEXT: [[TMP11]] = and i32 [[TMP9]], -65536112; GFX1100-NEXT: [[TMP12:%.*]] = icmp ne i32 [[TMP5]], [[TMP11]]113; GFX1100-NEXT: br i1 [[TMP12]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]114; GFX1100: partword.cmpxchg.end:115; GFX1100-NEXT: [[EXTRACTED:%.*]] = trunc i32 [[TMP9]] to i16116; GFX1100-NEXT: [[TMP13:%.*]] = insertvalue { i16, i1 } poison, i16 [[EXTRACTED]], 0117; GFX1100-NEXT: [[TMP14:%.*]] = insertvalue { i16, i1 } [[TMP13]], i1 [[TMP10]], 1118; GFX1100-NEXT: [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP14]], 0119; GFX1100-NEXT: ret i16 [[EXTRACT]]120;121 %gep = getelementptr i16, ptr addrspace(1) %out, i64 4122 %res = cmpxchg ptr addrspace(1) %gep, i16 %old, i16 %in seq_cst seq_cst, align 4, !mmra !2123 %extract = extractvalue {i16, i1} %res, 0124 ret i16 %extract125}126 127define void @syncscope_workgroup_nortn(ptr %addr, float %val) {128; GFX90A-LABEL: define void @syncscope_workgroup_nortn(129; GFX90A-SAME: ptr [[ADDR:%.*]], float [[VAL:%.*]]) #[[ATTR0]] {130; GFX90A-NEXT: [[IS_SHARED:%.*]] = call i1 @llvm.amdgcn.is.shared(ptr [[ADDR]])131; GFX90A-NEXT: br i1 [[IS_SHARED]], label [[ATOMICRMW_SHARED:%.*]], label [[ATOMICRMW_CHECK_PRIVATE:%.*]]132; GFX90A: atomicrmw.shared:133; GFX90A-NEXT: [[TMP1:%.*]] = addrspacecast ptr [[ADDR]] to ptr addrspace(3)134; GFX90A-NEXT: [[TMP2:%.*]] = atomicrmw fadd ptr addrspace(3) [[TMP1]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !mmra [[META0]], !amdgpu.no.fine.grained.memory [[META3:![0-9]+]], !amdgpu.ignore.denormal.mode [[META3]]135; GFX90A-NEXT: br label [[ATOMICRMW_PHI:%.*]]136; GFX90A: atomicrmw.check.private:137; GFX90A-NEXT: [[IS_PRIVATE:%.*]] = call i1 @llvm.amdgcn.is.private(ptr [[ADDR]])138; GFX90A-NEXT: br i1 [[IS_PRIVATE]], label [[ATOMICRMW_PRIVATE:%.*]], label [[ATOMICRMW_GLOBAL:%.*]]139; GFX90A: atomicrmw.private:140; GFX90A-NEXT: [[TMP3:%.*]] = addrspacecast ptr [[ADDR]] to ptr addrspace(5)141; GFX90A-NEXT: [[LOADED_PRIVATE:%.*]] = load float, ptr addrspace(5) [[TMP3]], align 4142; GFX90A-NEXT: [[VAL_NEW:%.*]] = fadd float [[LOADED_PRIVATE]], [[VAL]]143; GFX90A-NEXT: store float [[VAL_NEW]], ptr addrspace(5) [[TMP3]], align 4144; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]145; GFX90A: atomicrmw.global:146; GFX90A-NEXT: [[TMP4:%.*]] = addrspacecast ptr [[ADDR]] to ptr addrspace(1)147; GFX90A-NEXT: [[RES:%.*]] = atomicrmw fadd ptr addrspace(1) [[TMP4]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !mmra [[META0]], !amdgpu.no.fine.grained.memory [[META3]], !amdgpu.ignore.denormal.mode [[META3]]148; GFX90A-NEXT: br label [[ATOMICRMW_PHI]]149; GFX90A: atomicrmw.phi:150; GFX90A-NEXT: br label [[ATOMICRMW_END:%.*]]151; GFX90A: atomicrmw.end:152; GFX90A-NEXT: ret void153;154; GFX1100-LABEL: define void @syncscope_workgroup_nortn(155; GFX1100-SAME: ptr [[ADDR:%.*]], float [[VAL:%.*]]) #[[ATTR0]] {156; GFX1100-NEXT: [[RES:%.*]] = atomicrmw fadd ptr [[ADDR]], float [[VAL]] syncscope("workgroup") seq_cst, align 4, !mmra [[META0]], !amdgpu.no.fine.grained.memory [[META3:![0-9]+]], !amdgpu.ignore.denormal.mode [[META3]]157; GFX1100-NEXT: ret void158;159 %res = atomicrmw fadd ptr %addr, float %val syncscope("workgroup") seq_cst, !mmra !2, !amdgpu.no.fine.grained.memory !3, !amdgpu.ignore.denormal.mode !3160 ret void161}162 163!0 = !{!"foo", !"bar"}164!1 = !{!"bux", !"baz"}165!2 = !{!0, !1}166!3 = !{}167;.168; GFX90A: [[META0]] = !{[[META1:![0-9]+]], [[META2:![0-9]+]]}169; GFX90A: [[META1]] = !{!"foo", !"bar"}170; GFX90A: [[META2]] = !{!"bux", !"baz"}171; GFX90A: [[META3]] = !{}172;.173; GFX1100: [[META0]] = !{[[META1:![0-9]+]], [[META2:![0-9]+]]}174; GFX1100: [[META1]] = !{!"foo", !"bar"}175; GFX1100: [[META2]] = !{!"bux", !"baz"}176; GFX1100: [[META3]] = !{}177;.178