AMDGPU/R600: Add get_global_offset_{x,y,z} intrinsic

[jvesely]

Summary

split the tests for r600 and amdgcn
global offset is located after kernel args and workdim

[jvesely]

grop amdgcn parts. it should use kergnarg segment.

[jvesely]

remove offset legacy tests from amdgcn

[jvesely]

ping

[jvesely]

ping

[• tstellarAMD]

Do we still need this if we have the implicitarg.ptr intrinsic?

[jvesely]

In D20298#461485, @tstellarAMD wrote:

Do we still need this if we have the implicitarg.ptr intrinsic?

This one is for R600. We can switch R600 to use implicitarg.ptr too, but it's a bit more tricky since the loads can only handle constant pointers.

We might consider exposing it as int pointer, since args use AS 8, resulting in some ugly casts in libclc.

[jvesely]

we can follow gcn and don't need these

include/llvm/IR/IntrinsicsAMDGPU.td

 defm int_r600_read_local_size : AMDGPUReadPreloadRegisterIntrinsic_xyz <
                                        "__builtin_r600_read_local_size">;
 defm int_r600_read_ngroups : AMDGPUReadPreloadRegisterIntrinsic_xyz <
                                        "__builtin_r600_read_ngroups">;
 defm int_r600_read_tgid : AMDGPUReadPreloadRegisterIntrinsic_xyz <
                                        "__builtin_r600_read_tgid">;
 defm int_r600_read_tidig : AMDGPUReadPreloadRegisterIntrinsic_xyz <
                                        "__builtin_r600_read_tidig">;
+defm int_r600_read_global_offset : AMDGPUReadPreloadRegisterIntrinsic_xyz <
+                                       "__builtin_r600_read_global_offset">;
 
 def int_r600_rat_store_typed :
   // 1st parameter: Data
   // 2nd parameter: Index
   // 3rd parameter: Constant RAT ID
   Intrinsic<[], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_i32_ty], []>,
   GCCBuiltin<"__builtin_r600_rat_store_typed">;
 

lib/Target/AMDGPU/AMDGPUISelLowering.h

public:
   ///
   /// \returns a RegisterSDNode representing Reg.
   virtual SDValue CreateLiveInRegister(SelectionDAG &DAG,
                                        const TargetRegisterClass *RC,
                                        unsigned Reg, EVT VT) const;
 
   enum ImplicitParameter {
     GRID_DIM,
-    GRID_OFFSET
+    GRID_OFFSET_X,
+    GRID_OFFSET_Y,
+    GRID_OFFSET_Z,
   };
 
   /// \brief Helper function that returns the byte offset of the given
   /// type of implicit parameter.
   uint32_t getImplicitParameterOffset(const AMDGPUMachineFunction *MFI,
                                       const ImplicitParameter Param) const;
 };
 

lib/Target/AMDGPU/AMDGPUISelLowering.cpp

 }
 
 uint32_t AMDGPUTargetLowering::getImplicitParameterOffset(
     const AMDGPUMachineFunction *MFI, const ImplicitParameter Param) const {
   uint64_t ArgOffset = MFI->ABIArgOffset;
   switch (Param) {
   case GRID_DIM:
     return ArgOffset;
-  case GRID_OFFSET:
+  case GRID_OFFSET_X:
     return ArgOffset + 4;
+  case GRID_OFFSET_Y:
+    return ArgOffset + 8;
+  case GRID_OFFSET_Z:
+    return ArgOffset + 12;
   }
   llvm_unreachable("unexpected implicit parameter type");
 }
 
 #define NODE_NAME_CASE(node) case AMDGPUISD::node: return #node;
 
 const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   switch ((AMDGPUISD::NodeType)Opcode) {

lib/Target/AMDGPU/R600ISelLowering.cpp

case ISD::INTRINSIC_WO_CHAIN: {
     case Intrinsic::r600_read_local_size_z:
       return LowerImplicitParameter(DAG, VT, DL, 8);
 
     case Intrinsic::r600_read_workdim:
     case AMDGPUIntrinsic::AMDGPU_read_workdim: { // Legacy name.
       uint32_t ByteOffset = getImplicitParameterOffset(MFI, GRID_DIM);
       return LowerImplicitParameter(DAG, VT, DL, ByteOffset / 4);
     }
+    case Intrinsic::r600_read_global_offset_x: {
+      uint32_t ByteOffset = getImplicitParameterOffset(MFI, GRID_OFFSET_X);
+      return LowerImplicitParameter(DAG, VT, DL, ByteOffset / 4);
+    }
+    case Intrinsic::r600_read_global_offset_y: {
+      uint32_t ByteOffset = getImplicitParameterOffset(MFI, GRID_OFFSET_Y);
+      return LowerImplicitParameter(DAG, VT, DL, ByteOffset / 4);
+    }
+    case Intrinsic::r600_read_global_offset_z: {
+      uint32_t ByteOffset = getImplicitParameterOffset(MFI, GRID_OFFSET_Z);
+      return LowerImplicitParameter(DAG, VT, DL, ByteOffset / 4);
+    }
 
     case Intrinsic::r600_read_tgid_x:
       return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
                                   AMDGPU::T1_X, VT);
     case Intrinsic::r600_read_tgid_y:
       return CreateLiveInRegister(DAG, &AMDGPU::R600_TReg32RegClass,
                                   AMDGPU::T1_Y, VT);
     case Intrinsic::r600_read_tgid_z:

test/CodeGen/AMDGPU/amdgcn.work-item-intrinsics.ll

+; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=SI-NOHSA -check-prefix=GCN-NOHSA -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI  -check-prefix=VI-NOHSA -check-prefix=GCN -check-prefix=GCN-NOHSA -check-prefix=FUNC %s
+
+
+; FUNC-LABEL: {{^}}workdim:
+
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0xb
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x2c
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+define void @workdim (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.read.workdim() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; The workgroup.id values are stored in sgprs offset by the number of user
+; sgprs.
+
+; FUNC-LABEL: {{^}}workgroup_id_x:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s2{{$}}
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
+define void @workgroup_id_x(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workgroup.id.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}workgroup_id_y:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+define void @workgroup_id_y(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workgroup.id.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}workgroup_id_z:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3{{$}}
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
+define void @workgroup_id_z(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workgroup.id.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 132{{$}}
+
+; FUNC-LABEL: {{^}}workitem_id_x:
+; GCN-NOHSA: buffer_store_dword v0
+define void @workitem_id_x(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workitem.id.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 2180{{$}}
+
+; FUNC-LABEL: {{^}}workitem_id_y:
+
+; GCN-NOHSA: buffer_store_dword v1
+define void @workitem_id_y(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workitem.id.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 4228{{$}}
+
+; FUNC-LABEL: {{^}}workitem_id_z:
+; GCN-NOHSA: buffer_store_dword v2
+define void @workitem_id_z(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.amdgcn.workitem.id.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+declare i32 @llvm.amdgcn.ngroups.x() #0
+declare i32 @llvm.amdgcn.ngroups.y() #0
+declare i32 @llvm.amdgcn.ngroups.z() #0
+
+declare i32 @llvm.amdgcn.global.size.x() #0
+declare i32 @llvm.amdgcn.global.size.y() #0
+declare i32 @llvm.amdgcn.global.size.z() #0
+
+declare i32 @llvm.amdgcn.local.size.x() #0
+declare i32 @llvm.amdgcn.local.size.y() #0
+declare i32 @llvm.amdgcn.local.size.z() #0
+
+declare i32 @llvm.amdgcn.workgroup.id.x() #0
+declare i32 @llvm.amdgcn.workgroup.id.y() #0
+declare i32 @llvm.amdgcn.workgroup.id.z() #0
+
+declare i32 @llvm.amdgcn.workitem.id.x() #0
+declare i32 @llvm.amdgcn.workitem.id.y() #0
+declare i32 @llvm.amdgcn.workitem.id.z() #0
+
+declare i32 @llvm.amdgcn.read.workdim() #0
+
+
+;; Legacy (r600/AMDGPU) intrinsics
+
+; FUNC-LABEL: {{^}}workdim_legacy:
+
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0xb
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x2c
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+define void @workdim_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.AMDGPU.read.workdim() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_x_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x0
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x0
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @ngroups_x_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_y_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x1
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x4
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @ngroups_y_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_z_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x2
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x8
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @ngroups_z_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_x_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x3
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0xc
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @global_size_x_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_y_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x4
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x10
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @global_size_y_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_z_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x5
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x14
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @global_size_z_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_x_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x6
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x18
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @local_size_x_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_y_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x7
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x1c
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @local_size_y_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_z_legacy:
+; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x8
+; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x20
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+define void @local_size_z_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; The tgid values are stored in sgprs offset by the number of user
+; sgprs.
+
+; FUNC-LABEL: {{^}}tgid_x_legacy:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s2{{$}}
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
+define void @tgid_x_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tgid_y_legacy:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+define void @tgid_y_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tgid_z_legacy:
+; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3{{$}}
+; GCN-NOHSA: buffer_store_dword [[VVAL]]
+
+; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
+; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
+; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 1
+; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
+define void @tgid_z_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 132{{$}}
+
+; FUNC-LABEL: {{^}}tidig_x_legacy:
+; GCN-NOHSA: buffer_store_dword v0
+define void @tidig_x_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 2180{{$}}
+
+; FUNC-LABEL: {{^}}tidig_y_legacy:
+
+; GCN-NOHSA: buffer_store_dword v1
+define void @tidig_y_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-NOHSA: .section .AMDGPU.config
+; GCN-NOHSA: .long 47180
+; GCN-NOHSA-NEXT: .long 4228{{$}}
+
+; FUNC-LABEL: {{^}}tidig_z_legacy:
+; GCN-NOHSA: buffer_store_dword v2
+define void @tidig_z_legacy(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+declare i32 @llvm.r600.read.ngroups.x() #0
+declare i32 @llvm.r600.read.ngroups.y() #0
+declare i32 @llvm.r600.read.ngroups.z() #0
+
+declare i32 @llvm.r600.read.global.size.x() #0
+declare i32 @llvm.r600.read.global.size.y() #0
+declare i32 @llvm.r600.read.global.size.z() #0
+
+declare i32 @llvm.r600.read.local.size.x() #0
+declare i32 @llvm.r600.read.local.size.y() #0
+declare i32 @llvm.r600.read.local.size.z() #0
+
+declare i32 @llvm.r600.read.tgid.x() #0
+declare i32 @llvm.r600.read.tgid.y() #0
+declare i32 @llvm.r600.read.tgid.z() #0
+
+declare i32 @llvm.r600.read.tidig.x() #0
+declare i32 @llvm.r600.read.tidig.y() #0
+declare i32 @llvm.r600.read.tidig.z() #0
+
+declare i32 @llvm.AMDGPU.read.workdim() #0
+
+attributes #0 = { readnone }

test/CodeGen/AMDGPU/r600.work-item-intrinsics.ll

+; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
+
+; FUNC-LABEL: {{^}}workdim:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[2].Z
+define void @workdim (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.workdim() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}workdim_legacy:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[2].Z
+define void @workdim_legacy (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.AMDGPU.read.workdim() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[0].X
+define void @ngroups_x (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[0].Y
+define void @ngroups_y (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}ngroups_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[0].Z
+define void @ngroups_z (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.ngroups.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_offset_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[2].W
+define void @global_offset_x (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.offset.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_offset_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[3].X
+define void @global_offset_y (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.offset.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_offset_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[3].Y
+define void @global_offset_z (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.offset.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[0].W
+define void @global_size_x (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[1].X
+define void @global_size_y (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}global_size_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[1].Y
+define void @global_size_z (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.global.size.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[1].Z
+define void @local_size_x (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[1].W
+define void @local_size_y (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}local_size_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
+; EG: MOV {{\*? *}}[[VAL]], KC0[2].X
+define void @local_size_z (i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.local.size.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; The tgid values are stored in registers
+
+; FUNC-LABEL: {{^}}tgid_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW T1.X
+define void @tgid_x(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tgid_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW T1.Y
+define void @tgid_y(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tgid_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW T1.Z
+define void @tgid_z(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tgid.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tidig_x:
+; EG: MEM_RAT_CACHELESS STORE_RAW T0.X
+define void @tidig_x(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.x() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tidig_y:
+; EG: MEM_RAT_CACHELESS STORE_RAW T0.Y
+define void @tidig_y(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.y() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+; FUNC-LABEL: {{^}}tidig_z:
+; EG: MEM_RAT_CACHELESS STORE_RAW T0.Z
+define void @tidig_z(i32 addrspace(1)* %out) {
+entry:
+  %0 = call i32 @llvm.r600.read.tidig.z() #0
+  store i32 %0, i32 addrspace(1)* %out
+  ret void
+}
+
+declare i32 @llvm.r600.read.ngroups.x() #0
+declare i32 @llvm.r600.read.ngroups.y() #0
+declare i32 @llvm.r600.read.ngroups.z() #0
+
+declare i32 @llvm.r600.read.global.offset.x() #0
+declare i32 @llvm.r600.read.global.offset.y() #0
+declare i32 @llvm.r600.read.global.offset.z() #0
+
+declare i32 @llvm.r600.read.global.size.x() #0
+declare i32 @llvm.r600.read.global.size.y() #0
+declare i32 @llvm.r600.read.global.size.z() #0
+
+declare i32 @llvm.r600.read.local.size.x() #0
+declare i32 @llvm.r600.read.local.size.y() #0
+declare i32 @llvm.r600.read.local.size.z() #0
+
+declare i32 @llvm.r600.read.tgid.x() #0
+declare i32 @llvm.r600.read.tgid.y() #0
+declare i32 @llvm.r600.read.tgid.z() #0
+
+declare i32 @llvm.r600.read.tidig.x() #0
+declare i32 @llvm.r600.read.tidig.y() #0
+declare i32 @llvm.r600.read.tidig.z() #0
+
+declare i32 @llvm.r600.read.workdim() #0
+declare i32 @llvm.AMDGPU.read.workdim() #0
+
+attributes #0 = { readnone }

test/CodeGen/AMDGPU/work-item-intrinsics.ll

-; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=SI-NOHSA -check-prefix=GCN-NOHSA -check-prefix=FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI  -check-prefix=VI-NOHSA -check-prefix=GCN -check-prefix=GCN-NOHSA -check-prefix=FUNC %s
-; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
-
-
-; FUNC-LABEL: {{^}}ngroups_x:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[0].X
-
-; GCN-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-define void @ngroups_x (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.ngroups.x() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}ngroups_y:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[0].Y
-
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x1
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x4
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-define void @ngroups_y (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.ngroups.y() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}ngroups_z:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[0].Z
-
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x2
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x8
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-define void @ngroups_z (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.ngroups.z() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}global_size_x:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[0].W
-
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x3
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0xc
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-define void @global_size_x (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.x() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}global_size_y:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[1].X
-
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x4
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x10
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-define void @global_size_y (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.y() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}global_size_z:
-; EG: MEM_RAT_CACHELESS STORE_RAW [[VAL:T[0-9]+\.X]]
-; EG: MOV {{\*? *}}[[VAL]], KC0[1].Y
-
-; SI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x5
-; VI-NOHSA: s_load_dword [[VAL:s[0-9]+]], s[0:1], 0x14
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], [[VAL]]
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-define void @global_size_z (i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.global.size.z() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; The tgid values are stored in sgprs offset by the number of user
-; sgprs.
-
-; FUNC-LABEL: {{^}}tgid_x:
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s2{{$}}
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
-; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
-; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
-; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 0
-; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
-define void @tgid_x(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tgid.x() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}tgid_y:
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
-define void @tgid_y(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tgid.y() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; FUNC-LABEL: {{^}}tgid_z:
-; GCN-NOHSA: v_mov_b32_e32 [[VVAL:v[0-9]+]], s3{{$}}
-; GCN-NOHSA: buffer_store_dword [[VVAL]]
-
-; GCN-NOHSA: COMPUTE_PGM_RSRC2:USER_SGPR: 2
-; GCN: COMPUTE_PGM_RSRC2:TGID_X_EN: 1
-; GCN: COMPUTE_PGM_RSRC2:TGID_Y_EN: 0
-; GCN: COMPUTE_PGM_RSRC2:TGID_Z_EN: 1
-; GCN: COMPUTE_PGM_RSRC2:TIDIG_COMP_CNT: 0
-define void @tgid_z(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tgid.z() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; GCN-NOHSA: .section .AMDGPU.config
-; GCN-NOHSA: .long 47180
-; GCN-NOHSA-NEXT: .long 132{{$}}
-
-; FUNC-LABEL: {{^}}tidig_x:
-; GCN-NOHSA: buffer_store_dword v0
-define void @tidig_x(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tidig.x() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; GCN-NOHSA: .section .AMDGPU.config
-; GCN-NOHSA: .long 47180
-; GCN-NOHSA-NEXT: .long 2180{{$}}
-
-; FUNC-LABEL: {{^}}tidig_y:
-
-; GCN-NOHSA: buffer_store_dword v1
-define void @tidig_y(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tidig.y() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-; GCN-NOHSA: .section .AMDGPU.config
-; GCN-NOHSA: .long 47180
-; GCN-NOHSA-NEXT: .long 4228{{$}}
-
-; FUNC-LABEL: {{^}}tidig_z:
-; GCN-NOHSA: buffer_store_dword v2
-define void @tidig_z(i32 addrspace(1)* %out) {
-entry:
-  %0 = call i32 @llvm.r600.read.tidig.z() #0
-  store i32 %0, i32 addrspace(1)* %out
-  ret void
-}
-
-declare i32 @llvm.r600.read.ngroups.x() #0
-declare i32 @llvm.r600.read.ngroups.y() #0
-declare i32 @llvm.r600.read.ngroups.z() #0
-
-declare i32 @llvm.r600.read.global.size.x() #0
-declare i32 @llvm.r600.read.global.size.y() #0
-declare i32 @llvm.r600.read.global.size.z() #0
-
-declare i32 @llvm.r600.read.tgid.x() #0
-declare i32 @llvm.r600.read.tgid.y() #0
-declare i32 @llvm.r600.read.tgid.z() #0
-
-declare i32 @llvm.r600.read.tidig.x() #0
-declare i32 @llvm.r600.read.tidig.y() #0
-declare i32 @llvm.r600.read.tidig.z() #0
-
-declare i32 @llvm.AMDGPU.read.workdim() #0
-
-attributes #0 = { readnone }