c30c2f4f3e
Before, when selecting candidates to rematerialize, we would only consider SGPR candidates when there was an excess of SGPR registers. Failing to eliminate the excess would result in spills to VGPRs. This is normally not an issue, unless spilling to VGPRs results in excess VGPRs. This patch does 2 things: * It relaxes the GCNRPTarget success criteria: now we accept regions where we spill SGPRs to VGPRs, as long as this does not end up in excess VGPRs. * It changes isSaveBeneficial to consider the excess VGPRs (which includes the SGPRs that would be spilled to VGPR). With these changes, the compiler rematerializes VGPRs when the excess SGPRs would result in VGPR excess. This has some unaddressed flaws: we should attempt to rematerialize SGPRs first in order to eliminate the SGPR excess that results in VGPR excess. Related to SWDEV-549940
610 lines
38 KiB
LLVM
610 lines
38 KiB
LLVM
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1201 < %s | FileCheck %s
|
|
; CHECK: Occupancy: 16
|
|
|
|
%struct.zot = type { ptr }
|
|
%struct.bar = type { i32, i32, i32, i32, i8, i32, i32, i32, i32, i32, i32, i32, i32, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, i32, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr, ptr }
|
|
%struct.snork = type { i32, i32, float, float, i32, i32, i32 }
|
|
%struct.barney = type { ptr }
|
|
%struct.zot.0 = type { %struct.ham }
|
|
%struct.ham = type { float, float, float, float }
|
|
%struct.zot.1 = type { %struct.wobble }
|
|
%struct.wobble = type { i32, i32, i32, i32 }
|
|
|
|
@global = external local_unnamed_addr addrspace(4) constant %struct.zot
|
|
@global.1 = external local_unnamed_addr addrspace(4) constant ptr
|
|
@global.2 = external local_unnamed_addr addrspace(4) constant ptr
|
|
@global.3 = external local_unnamed_addr addrspace(4) constant ptr
|
|
@global.4 = external local_unnamed_addr addrspace(4) constant ptr
|
|
|
|
declare void @llvm.trap() #0
|
|
|
|
declare void @llvm.lifetime.end.p5(ptr addrspace(5) captures(none)) #1
|
|
|
|
define amdgpu_kernel void @eggs(ptr addrspace(4) noundef readonly byref(%struct.bar) align 8 captures(none) %arg) local_unnamed_addr #2 {
|
|
bb:
|
|
%alloca = alloca [128 x float], align 16, addrspace(5)
|
|
%load = load i32, ptr addrspace(4) %arg, align 8, !amdgpu.noclobber !0
|
|
%getelementptr = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 8
|
|
%load1 = load i32, ptr addrspace(4) %getelementptr, align 8, !amdgpu.noclobber !0
|
|
%getelementptr2 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 12
|
|
%load3 = load i32, ptr addrspace(4) %getelementptr2, align 4, !amdgpu.noclobber !0
|
|
%getelementptr4 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 24
|
|
%load5 = load i32, ptr addrspace(4) %getelementptr4, align 8, !amdgpu.noclobber !0
|
|
%getelementptr6 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 36
|
|
%load7 = load i32, ptr addrspace(4) %getelementptr6, align 4, !amdgpu.noclobber !0
|
|
%load8 = load i32, ptr addrspace(4) null, align 4294967296
|
|
%getelementptr9 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 56
|
|
%load10 = load ptr, ptr addrspace(4) %getelementptr9, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast = addrspacecast ptr %load10 to ptr addrspace(1)
|
|
%getelementptr11 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 72
|
|
%load12 = load ptr, ptr addrspace(4) %getelementptr11, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast13 = addrspacecast ptr %load12 to ptr addrspace(1)
|
|
%getelementptr14 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 80
|
|
%load15 = load ptr, ptr addrspace(4) %getelementptr14, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast16 = addrspacecast ptr %load15 to ptr addrspace(1)
|
|
%getelementptr17 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 88
|
|
%load18 = load ptr, ptr addrspace(4) %getelementptr17, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast19 = addrspacecast ptr %load18 to ptr addrspace(1)
|
|
%getelementptr20 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 96
|
|
%load21 = load ptr, ptr addrspace(4) %getelementptr20, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast22 = addrspacecast ptr %load21 to ptr addrspace(1)
|
|
%getelementptr23 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 112
|
|
%load24 = load ptr, ptr addrspace(4) %getelementptr23, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast25 = addrspacecast ptr %load24 to ptr addrspace(1)
|
|
%getelementptr26 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 120
|
|
%load27 = load ptr, ptr addrspace(4) %getelementptr26, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast28 = addrspacecast ptr %load27 to ptr addrspace(1)
|
|
%getelementptr29 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 136
|
|
%load30 = load i32, ptr addrspace(4) %getelementptr29, align 8, !amdgpu.noclobber !0
|
|
%getelementptr31 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 144
|
|
%load32 = load ptr, ptr addrspace(4) %getelementptr31, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast33 = addrspacecast ptr %load32 to ptr addrspace(1)
|
|
%getelementptr34 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 152
|
|
%load35 = load ptr, ptr addrspace(4) %getelementptr34, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast36 = addrspacecast ptr %load35 to ptr addrspace(1)
|
|
%getelementptr37 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 176
|
|
%load38 = load ptr, ptr addrspace(4) %getelementptr37, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast39 = addrspacecast ptr %load38 to ptr addrspace(1)
|
|
%getelementptr40 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 184
|
|
%load41 = load ptr, ptr addrspace(4) %getelementptr40, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast42 = addrspacecast ptr %load41 to ptr addrspace(1)
|
|
%getelementptr43 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 192
|
|
%load44 = load ptr, ptr addrspace(4) %getelementptr43, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast45 = addrspacecast ptr %load44 to ptr addrspace(1)
|
|
%getelementptr46 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 200
|
|
%load47 = load ptr, ptr addrspace(4) %getelementptr46, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast48 = addrspacecast ptr %load47 to ptr addrspace(1)
|
|
%getelementptr49 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 208
|
|
%load50 = load ptr, ptr addrspace(4) %getelementptr49, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast51 = addrspacecast ptr %load50 to ptr addrspace(1)
|
|
%getelementptr52 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 216
|
|
%load53 = load ptr, ptr addrspace(4) %getelementptr52, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast54 = addrspacecast ptr %load53 to ptr addrspace(1)
|
|
%getelementptr55 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 224
|
|
%load56 = load ptr, ptr addrspace(4) %getelementptr55, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast57 = addrspacecast ptr %load56 to ptr addrspace(1)
|
|
%getelementptr58 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 232
|
|
%load59 = load ptr, ptr addrspace(4) %getelementptr58, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast60 = addrspacecast ptr %load59 to ptr addrspace(1)
|
|
%getelementptr61 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 280
|
|
%load62 = load ptr, ptr addrspace(4) %getelementptr61, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast63 = addrspacecast ptr %load62 to ptr addrspace(1)
|
|
%getelementptr64 = getelementptr inbounds nuw i8, ptr addrspace(4) %arg, i64 296
|
|
%load65 = load ptr, ptr addrspace(4) %getelementptr64, align 8, !amdgpu.noclobber !0
|
|
%addrspacecast66 = addrspacecast ptr %load65 to ptr addrspace(1)
|
|
%call = tail call noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x()
|
|
%and = and i32 %call, 31
|
|
%icmp = icmp eq i32 %and, 0
|
|
%lshr = lshr i32 %call, 5
|
|
%getelementptr67 = getelementptr inbounds nuw %struct.snork, ptr addrspace(3) null, i32 %call
|
|
%addrspacecast68 = addrspacecast ptr addrspace(3) %getelementptr67 to ptr
|
|
%getelementptr69 = getelementptr inbounds nuw i8, ptr addrspace(3) null, i32 %lshr
|
|
%addrspacecast70 = addrspacecast ptr addrspace(3) %getelementptr69 to ptr
|
|
%getelementptr71 = getelementptr inbounds nuw i32, ptr addrspace(3) null, i32 %lshr
|
|
%addrspacecast72 = addrspacecast ptr addrspace(3) %getelementptr71 to ptr
|
|
%load73 = load ptr, ptr addrspace(4) @global, align 8
|
|
%addrspacecast74 = addrspacecast ptr %load73 to ptr addrspace(4)
|
|
%load75 = load ptr, ptr addrspace(4) @global.2, align 8
|
|
%addrspacecast76 = addrspacecast ptr %load75 to ptr addrspace(1)
|
|
%icmp77 = icmp ne i32 %load5, -1
|
|
%add = add i32 %load8, -1
|
|
%uitofp = uitofp i32 %add to float
|
|
%getelementptr78 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 20
|
|
%addrspacecast79 = addrspacecast ptr addrspace(3) %getelementptr78 to ptr
|
|
%getelementptr80 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 4
|
|
%addrspacecast81 = addrspacecast ptr addrspace(3) %getelementptr80 to ptr
|
|
%getelementptr82 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 16
|
|
%addrspacecast83 = addrspacecast ptr addrspace(3) %getelementptr82 to ptr
|
|
%getelementptr84 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 12
|
|
%addrspacecast85 = addrspacecast ptr addrspace(3) %getelementptr84 to ptr
|
|
%getelementptr86 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 8
|
|
%addrspacecast87 = addrspacecast ptr addrspace(3) %getelementptr86 to ptr
|
|
%getelementptr88 = getelementptr inbounds nuw i8, ptr addrspace(3) %getelementptr67, i32 24
|
|
%addrspacecast89 = addrspacecast ptr addrspace(3) %getelementptr88 to ptr
|
|
%load90 = load ptr, ptr addrspace(4) null, align 4294967296
|
|
%addrspacecast91 = addrspacecast ptr %load90 to ptr addrspace(4)
|
|
%load92 = load ptr, ptr addrspace(4) @global.4, align 8
|
|
%addrspacecast93 = addrspacecast ptr %load92 to ptr addrspace(1)
|
|
%load94 = load ptr, ptr addrspace(4) @global.3, align 8
|
|
%addrspacecast95 = addrspacecast ptr %load94 to ptr addrspace(1)
|
|
%load96 = load ptr, ptr addrspace(4) @global.1, align 8
|
|
%addrspacecast97 = addrspacecast ptr %load96 to ptr addrspace(1)
|
|
%icmp98 = icmp eq ptr addrspace(1) %addrspacecast63, addrspacecast (ptr null to ptr addrspace(1))
|
|
%sext = sext i32 %load to i64
|
|
%icmp99 = icmp ne i32 %add, 0
|
|
%zext = zext i1 %icmp99 to i32
|
|
%add100 = add i32 %load7, %zext
|
|
%getelementptr101 = getelementptr inbounds nuw i8, ptr addrspace(1) %addrspacecast33, i64 4294967295
|
|
%getelementptr102 = getelementptr inbounds nuw i8, ptr addrspace(1) %addrspacecast63, i64 8
|
|
br label %bb103
|
|
|
|
bb103: ; preds = %bb364, %bb
|
|
%phi = phi i32 [ -1, %bb ], [ %phi143, %bb364 ]
|
|
%phi104 = phi nsz float [ 0.0, %bb ], [ %phi144, %bb364 ]
|
|
%phi105 = phi i32 [ -1, %bb ], [ %phi365, %bb364 ]
|
|
%call106 = tail call i32 @llvm.amdgcn.ballot.i32(i1 true)
|
|
%icmp107 = icmp slt i32 %phi105, 0
|
|
%call108 = tail call i32 asm sideeffect "", "=v,0"(i32 range(i32 0, 2) 0) #7
|
|
%icmp109 = icmp ne i32 %call108, 0
|
|
%call110 = tail call i32 @llvm.amdgcn.ballot.i32(i1 %icmp109)
|
|
%icmp111 = icmp eq i32 %call110, 0
|
|
br i1 %icmp111, label %bb113, label %bb112
|
|
|
|
bb112: ; preds = %bb103
|
|
tail call void @llvm.trap()
|
|
unreachable
|
|
|
|
bb113: ; preds = %bb103
|
|
%call114 = tail call i32 @llvm.amdgcn.ballot.i32(i1 %icmp107)
|
|
%and115 = and i32 %call114, %call106
|
|
%call116 = tail call noundef range(i32 0, 33) i32 @llvm.ctpop.i32(i32 %and115)
|
|
%icmp117 = icmp samesign ugt i32 %call116, 3
|
|
br i1 %icmp117, label %bb118, label %bb141
|
|
|
|
bb118: ; preds = %bb113
|
|
%icmp119 = icmp eq i32 %and115, -1
|
|
br i1 %icmp119, label %bb120, label %bb122
|
|
|
|
bb120: ; preds = %bb118
|
|
%load121 = load volatile i8, ptr %addrspacecast70, align 1, !noalias.addrspace !1
|
|
%trunc = trunc nuw i8 %load121 to i1
|
|
br i1 %trunc, label %bb398, label %bb122
|
|
|
|
bb122: ; preds = %bb120, %bb118
|
|
br i1 %icmp, label %bb123, label %bb127
|
|
|
|
bb123: ; preds = %bb122
|
|
%atomicrmw = atomicrmw add ptr addrspace(1) %addrspacecast, i32 %call116 syncscope("agent") monotonic, align 4
|
|
%load124 = load volatile i32, ptr %addrspacecast72, align 4, !noalias.addrspace !1
|
|
%icmp125 = icmp ult i32 %load124, %load1
|
|
br i1 %icmp125, label %bb127, label %bb126
|
|
|
|
bb126: ; preds = %bb123
|
|
store volatile i8 1, ptr %addrspacecast70, align 1, !noalias.addrspace !1
|
|
br label %bb127
|
|
|
|
bb127: ; preds = %bb126, %bb123, %bb122
|
|
br i1 %icmp107, label %bb128, label %bb141
|
|
|
|
bb128: ; preds = %bb127
|
|
%load129 = load i32, ptr addrspace(1) %addrspacecast16, align 4
|
|
%urem = urem i32 %load129, %load3
|
|
%load130 = load i32, ptr addrspace(1) %addrspacecast39, align 4
|
|
%urem131 = urem i32 %load130, %load3
|
|
%zext132 = zext i32 %urem131 to i64
|
|
%getelementptr133 = getelementptr inbounds nuw i32, ptr addrspace(1) %addrspacecast42, i64 %zext132
|
|
%load134 = load i32, ptr addrspace(1) %getelementptr133, align 4
|
|
%load135 = load <4 x i32>, ptr addrspace(4) %addrspacecast74, align 16
|
|
%call136 = tail call float @llvm.amdgcn.struct.buffer.load.format.f32(<4 x i32> %load135, i32 %load134, i32 0, i32 0, i32 0)
|
|
%bitcast = bitcast float %call136 to i32
|
|
%and137 = and i32 %bitcast, 65536
|
|
%icmp138 = icmp eq i32 %and137, 0
|
|
%select = select i1 %icmp138, i32 -1, i32 %load134
|
|
%load139 = load float, ptr addrspace(1) %addrspacecast76, align 4
|
|
store i32 -1, ptr addrspace(1) null, align 4294967296
|
|
store float 0x3FF0000100000000, ptr addrspace(1) %addrspacecast45, align 4
|
|
store float 0.000000e+00, ptr addrspace(5) %alloca, align 16
|
|
%zext140 = zext i32 %urem to i64
|
|
br label %bb141
|
|
|
|
bb141: ; preds = %bb128, %bb127, %bb113
|
|
%phi142 = phi i32 [ %load30, %bb128 ], [ 0, %bb127 ], [ 0, %bb113 ]
|
|
%phi143 = phi i32 [ %select, %bb128 ], [ %phi, %bb127 ], [ %phi, %bb113 ]
|
|
%phi144 = phi nsz float [ %load139, %bb128 ], [ %phi104, %bb127 ], [ %phi104, %bb113 ]
|
|
%phi145 = phi i32 [ 1, %bb128 ], [ %phi105, %bb127 ], [ %phi105, %bb113 ]
|
|
%phi146 = phi i64 [ %sext, %bb128 ], [ 0, %bb127 ], [ 0, %bb113 ]
|
|
%phi147 = phi i64 [ %zext140, %bb128 ], [ 0, %bb127 ], [ 0, %bb113 ]
|
|
%phi148 = phi i32 [ %load129, %bb128 ], [ 0, %bb127 ], [ 0, %bb113 ]
|
|
%icmp149 = icmp sgt i32 %phi145, 0
|
|
%icmp150 = icmp ult i32 %phi142, 1073741824
|
|
%select151 = select i1 %icmp149, i1 %icmp150, i1 false
|
|
br i1 %select151, label %bb152, label %bb166
|
|
|
|
bb152: ; preds = %bb141
|
|
%and153 = and i32 %phi142, 134217727
|
|
%call154 = tail call <3 x float> @llvm.amdgcn.struct.buffer.load.format.v3f32(<4 x i32> zeroinitializer, i32 %and153, i32 0, i32 0, i32 0)
|
|
%bitcast155 = bitcast <3 x float> %call154 to <3 x i32>
|
|
%extractelement = extractelement <3 x i32> %bitcast155, i64 2
|
|
%lshr156 = lshr i32 %extractelement, 8
|
|
%zext157 = zext nneg i32 %lshr156 to i64
|
|
%getelementptr158 = getelementptr inbounds nuw i8, ptr addrspace(1) %addrspacecast33, i64 %zext157
|
|
store i8 2, ptr addrspace(1) %getelementptr158, align 1
|
|
br label %bb159
|
|
|
|
bb159: ; preds = %bb159, %bb152
|
|
%call160 = tail call <3 x float> @llvm.amdgcn.struct.buffer.load.format.v3f32(<4 x i32> zeroinitializer, i32 0, i32 0, i32 0, i32 0)
|
|
%bitcast161 = bitcast <3 x float> %call160 to <3 x i32>
|
|
%extractelement162 = extractelement <3 x i32> %bitcast161, i64 2
|
|
%lshr163 = lshr i32 %extractelement162, 8
|
|
%zext164 = zext nneg i32 %lshr163 to i64
|
|
%getelementptr165 = getelementptr inbounds nuw i8, ptr addrspace(1) %addrspacecast33, i64 %zext164
|
|
store i8 2, ptr addrspace(1) %getelementptr165, align 1
|
|
br label %bb159, !llvm.loop !2
|
|
|
|
bb166: ; preds = %bb141
|
|
%load167 = load float, ptr addrspace(1) %addrspacecast25, align 16
|
|
%load168 = load float, ptr addrspace(1) inttoptr (i64 4 to ptr addrspace(1)), align 4
|
|
%load169 = load float, ptr addrspace(1) inttoptr (i64 8 to ptr addrspace(1)), align 8
|
|
%and170 = and i32 %phi142, 536870911
|
|
%load171 = load <4 x i32>, ptr addrspace(4) null, align 4294967296
|
|
%call172 = tail call float @llvm.amdgcn.struct.buffer.load.format.f32(<4 x i32> %load171, i32 %and170, i32 0, i32 0, i32 0)
|
|
%bitcast173 = bitcast float %call172 to i32
|
|
%zext174 = zext i32 %bitcast173 to i64
|
|
%getelementptr175 = getelementptr inbounds nuw i32, ptr addrspace(1) %addrspacecast36, i64 %zext174
|
|
%load176 = load i32, ptr addrspace(1) %getelementptr175, align 4
|
|
%icmp177 = icmp ne i32 %and170, %load5
|
|
%and178 = and i1 %icmp77, %icmp177
|
|
%icmp179 = icmp eq i32 %phi143, %and170
|
|
br i1 %icmp179, label %bb180, label %bb181
|
|
|
|
bb180: ; preds = %bb166
|
|
br label %bb181
|
|
|
|
bb181: ; preds = %bb180, %bb166
|
|
%phi182 = phi i32 [ 0, %bb180 ], [ %load176, %bb166 ]
|
|
%phi183 = phi i1 [ true, %bb180 ], [ %and178, %bb166 ]
|
|
%phi184 = phi i32 [ -1, %bb180 ], [ %and170, %bb166 ]
|
|
%phi185 = phi i32 [ 0, %bb180 ], [ %phi145, %bb166 ]
|
|
br i1 %phi183, label %bb204, label %bb186
|
|
|
|
bb186: ; preds = %bb181
|
|
%fmul = fmul reassoc nnan ninf nsz arcp contract float %phi144, %uitofp
|
|
%sub = sub nuw i32 %phi184, %load7
|
|
%mul = mul i32 %sub, %load8
|
|
%add187 = add i32 %add100, %mul
|
|
%mul188 = mul i32 %add187, 3
|
|
%call189 = tail call <2 x float> @llvm.amdgcn.struct.buffer.load.format.v2f32(<4 x i32> zeroinitializer, i32 %mul188, i32 0, i32 0, i32 0)
|
|
%extractelement190 = extractelement <2 x float> %call189, i64 1
|
|
%call191 = tail call <3 x float> @llvm.amdgcn.struct.buffer.load.format.v3f32(<4 x i32> zeroinitializer, i32 0, i32 0, i32 0, i32 0)
|
|
%extractelement192 = extractelement <3 x float> %call191, i64 2
|
|
%extractelement193 = extractelement <3 x float> %call191, i64 0
|
|
%extractelement194 = extractelement <3 x float> %call191, i64 1
|
|
%fmul195 = fmul reassoc nnan ninf nsz arcp contract float %extractelement192, %fmul
|
|
%fmul196 = fmul reassoc nnan ninf nsz arcp contract float %fmul, %fmul
|
|
%fmul197 = fmul reassoc nnan ninf nsz arcp contract float %fmul196, %extractelement190
|
|
%fmul198 = fmul reassoc nnan ninf nsz arcp contract float %fmul197, %extractelement193
|
|
%fmul199 = fmul reassoc nnan ninf nsz arcp contract float %fmul198, %fmul195
|
|
%fneg = fneg reassoc nnan ninf nsz arcp contract float %fmul
|
|
%fmul200 = fmul reassoc nnan ninf nsz arcp contract float %extractelement194, %fneg
|
|
%load201 = load float, ptr addrspace(1) %addrspacecast28, align 16
|
|
%fmul202 = fmul reassoc nnan ninf nsz arcp contract float %fmul200, %load201
|
|
%fmul203 = fmul reassoc nnan ninf nsz arcp contract float %fmul202, %fmul195
|
|
%fdiv = fdiv reassoc nnan ninf nsz arcp contract float %fmul203, %fmul199
|
|
br label %bb204
|
|
|
|
bb204: ; preds = %bb186, %bb181
|
|
%phi205 = phi float [ %load169, %bb181 ], [ 0.000000e+00, %bb186 ]
|
|
%phi206 = phi float [ %load168, %bb181 ], [ 0.000000e+00, %bb186 ]
|
|
%phi207 = phi float [ 0.000000e+00, %bb181 ], [ %fdiv, %bb186 ]
|
|
%phi208 = phi float [ %load167, %bb181 ], [ 0.000000e+00, %bb186 ]
|
|
%lshr209 = lshr i32 %phi182, 27
|
|
%and210 = and i32 %lshr209, 7
|
|
%and211 = and i32 %phi182, 134217727
|
|
%zext212 = zext nneg i32 %and210 to i64
|
|
%getelementptr213 = getelementptr inbounds nuw %struct.barney, ptr addrspace(4) null, i64 %zext212
|
|
%load214 = load i64, ptr addrspace(4) %getelementptr213, align 8
|
|
%inttoptr = inttoptr i64 %load214 to ptr
|
|
%addrspacecast215 = addrspacecast ptr %inttoptr to ptr addrspace(1)
|
|
%shl = shl nuw nsw i32 %and211, 4
|
|
%zext216 = zext nneg i32 %shl to i64
|
|
%getelementptr217 = getelementptr inbounds nuw i8, ptr addrspace(1) %addrspacecast215, i64 %zext216
|
|
%getelementptr218 = getelementptr inbounds nuw i8, ptr addrspace(1) %getelementptr217, i64 4
|
|
%load219 = load i16, ptr addrspace(1) %getelementptr218, align 4
|
|
%load220 = load i32, ptr addrspace(1) null, align 4294967296
|
|
%lshr221 = lshr i32 %load220, 12
|
|
%getelementptr222 = getelementptr inbounds nuw i8, ptr addrspace(1) %getelementptr217, i64 12
|
|
%load223 = load i32, ptr addrspace(1) %getelementptr222, align 4
|
|
%lshr224 = lshr i32 %load223, 25
|
|
%and225 = and i32 %lshr224, 31
|
|
%add226 = add nsw i32 %and225, -1
|
|
%uitofp227 = uitofp i32 %add226 to float
|
|
%call228 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.minnum.f32(float nofpclass(nan inf) %phi144, float 0x3FEFFFEB00000000)
|
|
%fmul229 = fmul reassoc nnan ninf nsz arcp contract float %call228, %uitofp227
|
|
%fptoui = fptoui float %fmul229 to i32
|
|
%and230 = and i32 %load220, 65535
|
|
%and231 = and i16 %load219, 32767
|
|
%zext232 = zext nneg i16 %and231 to i32
|
|
%add233 = add nuw nsw i32 %and230, %zext232
|
|
%and234 = and i32 %load223, 1073741824
|
|
%icmp235 = icmp eq i32 %and234, 0
|
|
br i1 %icmp235, label %bb277, label %bb236
|
|
|
|
bb236: ; preds = %bb204
|
|
%icmp237 = icmp eq i16 %and231, 0
|
|
br i1 %icmp237, label %bb351, label %bb238
|
|
|
|
bb238: ; preds = %bb236
|
|
%getelementptr239 = getelementptr inbounds nuw i8, ptr addrspace(1) %getelementptr217, i64 16
|
|
%sub240 = sub nsw i32 %and211, %lshr221
|
|
%add241 = add i32 %sub240, %fptoui
|
|
%addrspacecast242 = addrspacecast ptr %inttoptr to ptr addrspace(4)
|
|
%load243 = load <4 x i32>, ptr addrspace(4) %addrspacecast242, align 16
|
|
%fmul244 = fmul reassoc nnan ninf nsz arcp contract float %phi205, %phi205
|
|
%call245 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %phi206, float nofpclass(nan inf) %phi206, float nofpclass(nan inf) %fmul244)
|
|
%call246 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %phi207, float nofpclass(nan inf) %phi207, float nofpclass(nan inf) %call245)
|
|
%call247 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.sqrt.f32(float nofpclass(nan inf) %call246)
|
|
%getelementptr248 = getelementptr inbounds %struct.zot.0, ptr addrspace(1) %addrspacecast51, i64 %phi146
|
|
%call249 = tail call float @llvm.amdgcn.rsq.f32(float 0.000000e+00)
|
|
%load250 = load i32, ptr addrspace(1) %getelementptr239, align 4
|
|
%mul251 = mul i32 %load250, %and225
|
|
%add252 = add i32 %add241, %mul251
|
|
%call253 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load243, i32 %add252, i32 0, i32 0, i32 0)
|
|
%extractelement254 = extractelement <4 x float> %call253, i64 3
|
|
%fmul255 = fmul reassoc nnan ninf nsz arcp contract float %extractelement254, %extractelement254
|
|
%call256 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.sqrt.f32(float nofpclass(nan inf) %fmul255)
|
|
store volatile i32 %phi184, ptr %addrspacecast79, align 4, !noalias.addrspace !1
|
|
store volatile i32 %phi182, ptr %addrspacecast68, align 4, !noalias.addrspace !1
|
|
store volatile i32 %and230, ptr %addrspacecast81, align 4, !noalias.addrspace !1
|
|
%fmul257 = fmul reassoc nnan ninf nsz arcp contract float %phi207, %call256
|
|
%fdiv258 = fdiv reassoc nnan ninf nsz arcp contract float %fmul257, %call247
|
|
%fmul259 = fmul reassoc nnan ninf nsz arcp contract float %fdiv258, %fdiv258
|
|
%fcmp = fcmp reassoc nnan ninf nsz arcp contract ogt float %fmul259, 0.000000e+00
|
|
%select260 = select reassoc nnan ninf nsz arcp contract i1 %fcmp, float %call249, float 0.000000e+00
|
|
%fmul261 = fmul reassoc nnan ninf nsz arcp contract float %select260, %fdiv258
|
|
store float %fmul261, ptr addrspace(1) %getelementptr248, align 16
|
|
%fdiv262 = fdiv reassoc nnan ninf nsz arcp contract float 1.000000e+00, %call247
|
|
br label %bb263
|
|
|
|
bb263: ; preds = %bb263, %bb238
|
|
%load264 = load i32, ptr addrspace(1) null, align 4294967296
|
|
%mul265 = mul i32 %load264, %and225
|
|
%add266 = add i32 %add241, %mul265
|
|
%call267 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load243, i32 %add266, i32 0, i32 0, i32 0)
|
|
%extractelement268 = extractelement <4 x float> %call267, i64 3
|
|
%fmul269 = fmul reassoc nnan ninf nsz arcp contract float %extractelement268, %extractelement268
|
|
%call270 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.sqrt.f32(float nofpclass(nan inf) %fmul269)
|
|
store volatile i32 %phi184, ptr %addrspacecast79, align 4, !noalias.addrspace !1
|
|
store volatile i32 %phi182, ptr %addrspacecast68, align 4, !noalias.addrspace !1
|
|
store volatile i32 0, ptr %addrspacecast81, align 4, !noalias.addrspace !1
|
|
%fmul271 = fmul reassoc nnan ninf nsz arcp contract float %phi207, %call270
|
|
%fmul272 = fmul reassoc nnan ninf nsz arcp contract float %fmul271, %fdiv262
|
|
%fmul273 = fmul reassoc nnan ninf nsz arcp contract float %fmul272, %fmul272
|
|
%fcmp274 = fcmp reassoc nnan ninf nsz arcp contract ogt float %fmul273, 0.000000e+00
|
|
%select275 = select reassoc nnan ninf nsz arcp contract i1 %fcmp274, float %call249, float 0.000000e+00
|
|
%fmul276 = fmul reassoc nnan ninf nsz arcp contract float %select275, %fmul272
|
|
store float %fmul276, ptr addrspace(1) %getelementptr248, align 16
|
|
br label %bb263, !llvm.loop !4
|
|
|
|
bb277: ; preds = %bb204
|
|
%icmp278 = icmp slt i32 %load223, 0
|
|
br i1 %icmp278, label %bb279, label %bb348
|
|
|
|
bb279: ; preds = %bb277
|
|
%fmul280 = fmul reassoc nnan ninf nsz arcp contract float %phi207, %phi207
|
|
%call281 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.sqrt.f32(float nofpclass(nan inf) %fmul280)
|
|
%fdiv282 = fdiv reassoc nnan ninf nsz arcp contract float 1.000000e+00, %call281
|
|
%icmp283 = icmp eq i16 %and231, 0
|
|
br i1 %icmp283, label %bb351, label %bb284
|
|
|
|
bb284: ; preds = %bb279
|
|
%sub285 = sub nsw i32 %and211, %lshr221
|
|
%addrspacecast286 = addrspacecast ptr %inttoptr to ptr addrspace(4)
|
|
%fmul287 = fmul reassoc nnan ninf nsz arcp contract float %fdiv282, %phi207
|
|
%fmul288 = fmul reassoc nnan ninf nsz arcp contract float %fdiv282, %phi206
|
|
%fmul289 = fmul reassoc nnan ninf nsz arcp contract float %fdiv282, %phi205
|
|
%getelementptr290 = getelementptr inbounds %struct.zot.0, ptr addrspace(1) %addrspacecast51, i64 %phi146
|
|
%add291 = add i32 %sub285, %fptoui
|
|
%load292 = load <4 x i32>, ptr addrspace(4) %addrspacecast286, align 16
|
|
%add293 = add i32 %add291, 1
|
|
%add294 = add i32 %add291, %and225
|
|
%add295 = add i32 %add294, 1
|
|
br label %bb296
|
|
|
|
bb296: ; preds = %bb341, %bb284
|
|
%phi297 = phi i32 [ %and230, %bb284 ], [ %add346, %bb341 ]
|
|
%call298 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load292, i32 %add291, i32 0, i32 0, i32 0)
|
|
%extractelement299 = extractelement <4 x float> %call298, i64 0
|
|
%extractelement300 = extractelement <4 x float> %call298, i64 3
|
|
%call301 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load292, i32 %add293, i32 0, i32 0, i32 0)
|
|
%extractelement302 = extractelement <4 x float> %call301, i64 3
|
|
%fsub = fsub reassoc nnan ninf nsz arcp contract float %extractelement302, %extractelement300
|
|
%call303 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %fsub, float nofpclass(nan inf) %fmul229, float nofpclass(nan inf) %extractelement300)
|
|
%call304 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load292, i32 %add294, i32 0, i32 0, i32 0)
|
|
%extractelement305 = extractelement <4 x float> %call304, i64 3
|
|
%call306 = tail call <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32> %load292, i32 %add295, i32 0, i32 0, i32 0)
|
|
%extractelement307 = extractelement <4 x float> %call306, i64 3
|
|
%fsub308 = fsub reassoc nnan ninf nsz arcp contract float %extractelement307, %extractelement305
|
|
%call309 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %fsub308, float nofpclass(nan inf) %fmul229, float nofpclass(nan inf) %extractelement305)
|
|
%fneg310 = fneg reassoc nnan ninf nsz arcp contract float %extractelement299
|
|
%fmul311 = fmul reassoc nnan ninf nsz arcp contract float %extractelement299, %extractelement299
|
|
%call312 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.sqrt.f32(float nofpclass(nan inf) %fmul311)
|
|
%fdiv313 = fdiv reassoc nnan ninf nsz arcp contract float 1.000000e+00, %call312
|
|
%fmul314 = fmul reassoc nnan ninf nsz arcp contract float %fdiv313, %fneg310
|
|
%fmul315 = fmul reassoc nnan ninf nsz arcp contract float %fmul314, %fmul287
|
|
%fmul316 = fmul reassoc nnan ninf nsz arcp contract float %fmul315, %fmul315
|
|
%fsub317 = fsub reassoc nnan ninf nsz arcp contract float 1.000000e+00, %fmul316
|
|
%fcmp318 = fcmp reassoc nnan ninf nsz arcp contract oeq float %fsub317, 0.000000e+00
|
|
br i1 %fcmp318, label %bb341, label %bb319
|
|
|
|
bb319: ; preds = %bb296
|
|
%extractelement320 = extractelement <4 x float> %call301, i64 2
|
|
%extractelement321 = extractelement <4 x float> %call298, i64 2
|
|
%fsub322 = fsub reassoc nnan ninf nsz arcp contract float %extractelement320, %extractelement321
|
|
%call323 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %fsub322, float nofpclass(nan inf) %fmul229, float nofpclass(nan inf) %extractelement321)
|
|
%extractelement324 = extractelement <4 x float> %call301, i64 1
|
|
%extractelement325 = extractelement <4 x float> %call298, i64 1
|
|
%fsub326 = fsub reassoc nnan ninf nsz arcp contract float %extractelement324, %extractelement325
|
|
%call327 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %fsub326, float nofpclass(nan inf) %fmul229, float nofpclass(nan inf) %extractelement325)
|
|
%fsub328 = fsub reassoc nnan ninf nsz arcp contract float %extractelement299, %phi208
|
|
%fmul329 = fmul reassoc nnan ninf nsz arcp contract float %fmul314, %fsub328
|
|
%fmul330 = fmul reassoc nnan ninf nsz arcp contract float %fmul289, %call323
|
|
%call331 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %call327, float nofpclass(nan inf) %fmul288, float nofpclass(nan inf) %fmul330)
|
|
%call332 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.fma.f32(float nofpclass(nan inf) %fsub328, float nofpclass(nan inf) %fmul287, float nofpclass(nan inf) %call331)
|
|
%fmul333 = fmul reassoc nnan ninf nsz arcp contract float %fmul329, %fmul315
|
|
%fsub334 = fsub reassoc nnan ninf nsz arcp contract float %call332, %fmul333
|
|
%fmul335 = fmul reassoc nnan ninf nsz arcp contract float %fsub334, %fdiv282
|
|
%fdiv336 = fdiv reassoc nnan ninf nsz arcp contract float %fmul335, %fsub317
|
|
%call337 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.minnum.f32(float nofpclass(nan inf) %call312, float nofpclass(nan inf) 0.000000e+00)
|
|
%call338 = tail call reassoc nnan ninf nsz arcp contract noundef float @llvm.maxnum.f32(float nofpclass(nan inf) %call337, float 0.000000e+00)
|
|
%fmul339 = fmul reassoc nnan ninf nsz arcp contract float %call338, %fdiv313
|
|
%call340 = tail call reassoc nnan ninf nsz arcp contract float @llvm.fabs.f32(float nofpclass(nan inf) %fdiv336)
|
|
br label %bb341
|
|
|
|
bb341: ; preds = %bb319, %bb296
|
|
%phi342 = phi nsz float [ %fmul339, %bb319 ], [ 0.000000e+00, %bb296 ]
|
|
%phi343 = phi float [ %call340, %bb319 ], [ 0.000000e+00, %bb296 ]
|
|
%fsub344 = fsub reassoc nnan ninf nsz arcp contract float %call309, %call303
|
|
%fmul345 = fmul reassoc nnan ninf nsz arcp contract float %phi342, %fsub344
|
|
%fadd = fadd reassoc nnan ninf nsz arcp contract float %fmul345, %call303
|
|
store volatile i32 %load220, ptr %addrspacecast83, align 4, !noalias.addrspace !1
|
|
store float %fadd, ptr addrspace(1) %getelementptr290, align 16
|
|
store float 1.000000e+00, ptr addrspace(1) null, align 4294967296
|
|
%add346 = add nuw nsw i32 %phi297, 1
|
|
%icmp347 = icmp samesign ult i32 %add346, %add233
|
|
br i1 %icmp347, label %bb296, label %bb351
|
|
|
|
bb348: ; preds = %bb277
|
|
%icmp349 = icmp eq i16 %and231, 0
|
|
br i1 %icmp349, label %bb351, label %bb350
|
|
|
|
bb350: ; preds = %bb398, %bb348
|
|
ret void
|
|
|
|
bb351: ; preds = %bb348, %bb341, %bb279, %bb236
|
|
%phi352 = phi float [ 0.000000e+00, %bb279 ], [ 0.000000e+00, %bb348 ], [ 0.000000e+00, %bb236 ], [ %phi343, %bb341 ]
|
|
br label %bb353
|
|
|
|
bb353: ; preds = %bb353, %bb351
|
|
%phi354 = phi i32 [ %phi185, %bb351 ], [ %add355, %bb353 ]
|
|
%add355 = add nsw i32 %phi354, -1
|
|
%getelementptr356 = getelementptr inbounds float, ptr addrspace(5) %alloca, i32 %add355
|
|
%load357 = load float, ptr addrspace(5) %getelementptr356, align 4
|
|
%icmp358 = icmp sgt i32 %phi354, 1
|
|
%fcmp359 = fcmp reassoc nnan ninf nsz arcp contract ogt float %load357, %phi352
|
|
%select360 = select i1 %icmp358, i1 %fcmp359, i1 false
|
|
br i1 %select360, label %bb353, label %bb361
|
|
|
|
bb361: ; preds = %bb353
|
|
%icmp362 = icmp eq i32 %add355, 0
|
|
store i8 1, ptr addrspace(1) %getelementptr101, align 1
|
|
%atomicrmw363 = atomicrmw add ptr addrspace(1) %addrspacecast13, i32 1 syncscope("agent") monotonic, align 4
|
|
store i32 %phi148, ptr addrspace(1) %addrspacecast19, align 4
|
|
store volatile i32 536870911, ptr %addrspacecast89, align 4, !noalias.addrspace !1
|
|
br i1 %icmp362, label %bb366, label %bb364
|
|
|
|
bb364: ; preds = %bb397, %bb389, %bb386, %bb385, %bb361
|
|
%phi365 = phi i32 [ %add355, %bb361 ], [ -1, %bb385 ], [ -1, %bb386 ], [ -1, %bb389 ], [ -1, %bb397 ]
|
|
br label %bb103
|
|
|
|
bb366: ; preds = %bb361
|
|
%load367 = load volatile i32, ptr %addrspacecast79, align 4, !noalias.addrspace !1
|
|
%load368 = load volatile i32, ptr %addrspacecast68, align 4, !noalias.addrspace !1
|
|
%load369 = load volatile i32, ptr %addrspacecast81, align 4, !noalias.addrspace !1
|
|
%load370 = load volatile float, ptr %addrspacecast87, align 4, !noalias.addrspace !1
|
|
%load371 = load volatile float, ptr %addrspacecast85, align 4, !noalias.addrspace !1
|
|
%call372 = tail call float @llvm.amdgcn.struct.buffer.load.format.f32(<4 x i32> %load171, i32 %load367, i32 0, i32 0, i32 0)
|
|
%bitcast373 = bitcast float %call372 to i32
|
|
%load374 = load <4 x i32>, ptr addrspace(4) %addrspacecast91, align 16
|
|
%call375 = tail call float @llvm.amdgcn.struct.buffer.load.format.f32(<4 x i32> %load374, i32 %bitcast373, i32 0, i32 0, i32 0)
|
|
%getelementptr376 = getelementptr inbounds nuw %struct.zot.1, ptr addrspace(1) %addrspacecast93, i64 %phi147
|
|
%load377 = load i32, ptr addrspace(1) %addrspacecast22, align 4
|
|
%and378 = and i32 %load377, -285212672
|
|
%or = or disjoint i32 %and378, 268435456
|
|
store i32 0, ptr addrspace(1) %addrspacecast95, align 4
|
|
store i32 %or, ptr addrspace(1) %addrspacecast97, align 4
|
|
%getelementptr379 = getelementptr inbounds nuw i8, ptr addrspace(1) %getelementptr376, i64 8
|
|
store float %call375, ptr addrspace(1) %getelementptr379, align 8
|
|
%load380 = load i32, ptr addrspace(1) null, align 4294967296
|
|
%load381 = load i16, ptr addrspace(1) inttoptr (i64 6 to ptr addrspace(1)), align 2
|
|
%zext382 = zext i16 %load381 to i32
|
|
%and383 = and i32 %load380, 1073741824
|
|
%icmp384 = icmp eq i32 %and383, 0
|
|
br i1 %icmp384, label %bb387, label %bb385
|
|
|
|
bb385: ; preds = %bb366
|
|
store i32 0, ptr addrspace(1) %addrspacecast60, align 4
|
|
br i1 %icmp98, label %bb364, label %bb386
|
|
|
|
bb386: ; preds = %bb385
|
|
store float 0x47EFFFFFE0000000, ptr addrspace(1) %getelementptr102, align 8
|
|
br label %bb364
|
|
|
|
bb387: ; preds = %bb366
|
|
%icmp388 = icmp slt i32 %load380, 0
|
|
br i1 %icmp388, label %bb389, label %bb397
|
|
|
|
bb389: ; preds = %bb387
|
|
%sub390 = sub i32 %load369, %zext382
|
|
%zext391 = zext i32 %sub390 to i64
|
|
%getelementptr392 = getelementptr inbounds nuw %struct.zot.1, ptr addrspace(1) null, i64 %zext391
|
|
%getelementptr393 = getelementptr inbounds nuw i8, ptr addrspace(1) %getelementptr392, i64 4
|
|
%load394 = load i32, ptr addrspace(1) %getelementptr393, align 4
|
|
store i32 0, ptr addrspace(1) %addrspacecast66, align 4
|
|
%and395 = and i32 %load394, 1073741823
|
|
%or396 = or disjoint i32 %and395, -2147483648
|
|
store i32 %or396, ptr addrspace(1) %addrspacecast48, align 4
|
|
br label %bb364
|
|
|
|
bb397: ; preds = %bb387
|
|
store float 0.000000e+00, ptr addrspace(1) inttoptr (i64 12 to ptr addrspace(1)), align 4
|
|
store float 0.000000e+00, ptr addrspace(1) %addrspacecast54, align 16
|
|
store float 0.000000e+00, ptr addrspace(1) %addrspacecast57, align 4
|
|
br label %bb364
|
|
|
|
bb398: ; preds = %bb120
|
|
call void @llvm.lifetime.end.p5(ptr addrspace(5) %alloca) #8
|
|
br label %bb350
|
|
}
|
|
|
|
declare float @llvm.minnum.f32(float, float) #3
|
|
|
|
declare float @llvm.maxnum.f32(float, float) #3
|
|
|
|
declare float @llvm.fma.f32(float, float, float) #3
|
|
|
|
declare float @llvm.fabs.f32(float) #3
|
|
|
|
declare float @llvm.sqrt.f32(float) #3
|
|
|
|
declare i32 @llvm.ctpop.i32(i32) #3
|
|
|
|
declare float @llvm.amdgcn.rsq.f32(float) #4
|
|
|
|
declare noundef range(i32 0, 1024) i32 @llvm.amdgcn.workitem.id.x() #4
|
|
|
|
declare <4 x float> @llvm.amdgcn.struct.buffer.load.format.v4f32(<4 x i32>, i32, i32, i32, i32 immarg) #5
|
|
|
|
declare i32 @llvm.amdgcn.ballot.i32(i1) #6
|
|
|
|
declare float @llvm.amdgcn.struct.buffer.load.format.f32(<4 x i32>, i32, i32, i32, i32 immarg) #5
|
|
|
|
declare <2 x float> @llvm.amdgcn.struct.buffer.load.format.v2f32(<4 x i32>, i32, i32, i32, i32 immarg) #5
|
|
|
|
declare <3 x float> @llvm.amdgcn.struct.buffer.load.format.v3f32(<4 x i32>, i32, i32, i32, i32 immarg) #5
|
|
|
|
attributes #0 = { cold noreturn nounwind memory(inaccessiblemem: write) }
|
|
attributes #1 = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) }
|
|
attributes #2 = { convergent norecurse nounwind "amdgpu-flat-work-group-size"="1,1024" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="gfx1201" "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-fmin-fmax-global-f32,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+fp8-conversion-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+wavefrontsize32" "uniform-work-group-size"="true" }
|
|
attributes #3 = { nocallback nocreateundeforpoison nofree nosync nounwind speculatable willreturn memory(none) }
|
|
attributes #4 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
|
|
attributes #5 = { nocallback nofree nosync nounwind willreturn memory(read) }
|
|
attributes #6 = { convergent nocallback nofree nounwind willreturn memory(none) }
|
|
attributes #7 = { convergent nounwind }
|
|
attributes #8 = { nounwind }
|
|
|
|
!0 = !{}
|
|
!1 = !{i32 1, i32 3, i32 4, i32 10}
|
|
!2 = distinct !{!2, !3}
|
|
!3 = !{!"llvm.loop.peeled.count", i32 1}
|
|
!4 = distinct !{!4, !3}
|