// Calling convention for compute kernels
def CC_AMDGPU_Kernel : CallingConv<[
- CCIfType<[v4i32, v4f32], CCAssignToStack <16, 16>>,
- CCIfType<[i64, f64], CCAssignToStack < 8, 8>>,
- CCIfType<[i32, f32], CCAssignToStack < 4, 4>>,
- CCIfType<[i16], CCAssignToStack < 2, 4>>,
- CCIfType<[i8], CCAssignToStack < 1, 4>>
+ CCIfType<[v4i32, v4f32], CCAssignToStack <16, 16>>,
+ CCIfType<[i64, f64, v2f32, v2i32], CCAssignToStack < 8, 8>>,
+ CCIfType<[i32, f32], CCAssignToStack < 4, 4>>,
+ CCIfType<[i16], CCAssignToStack < 2, 4>>,
+ CCIfType<[i8], CCAssignToStack < 1, 4>>
]>;
def CC_AMDGPU : CallingConv<[
if (ST.getGeneration() > AMDGPUSubtarget::NORTHERN_ISLANDS) {
break;
}
+
+ unsigned RegClassID;
+ switch(N->getValueType(0).getVectorNumElements()) {
+ case 2: RegClassID = AMDGPU::R600_Reg64RegClassID; break;
+ case 4: RegClassID = AMDGPU::R600_Reg128RegClassID; break;
+ default: llvm_unreachable("Do not know how to lower this BUILD_VECTOR");
+ }
// BUILD_VECTOR is usually lowered into an IMPLICIT_DEF + 4 INSERT_SUBREG
// that adds a 128 bits reg copy when going through TwoAddressInstructions
// pass. We want to avoid 128 bits copies as much as possible because they
// can't be bundled by our scheduler.
SDValue RegSeqArgs[9] = {
- CurDAG->getTargetConstant(AMDGPU::R600_Reg128RegClassID, MVT::i32),
+ CurDAG->getTargetConstant(RegClassID, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32),
SDValue(), CurDAG->getTargetConstant(AMDGPU::sub2, MVT::i32),
setOperationAction(ISD::LOAD, MVT::f64, Promote);
AddPromotedToType(ISD::LOAD, MVT::f64, MVT::i64);
+ setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i32, Expand);
+ setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2f32, Expand);
+
setOperationAction(ISD::FNEG, MVT::v2f32, Expand);
setOperationAction(ISD::FNEG, MVT::v4f32, Expand);
case AMDGPU::R600_ExportBuf:
case AMDGPU::R600_ExportSwz:
case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
+ case AMDGPU::RAT_WRITE_CACHELESS_64_eg:
case AMDGPU::RAT_WRITE_CACHELESS_128_eg:
- case AMDGPU::RAT_STORE_DWORD_cm:
+ case AMDGPU::RAT_STORE_DWORD32_cm:
+ case AMDGPU::RAT_STORE_DWORD64_cm:
DEBUG(dbgs() << CfCount << ":"; MI->dump(););
CfCount++;
break;
addRegisterClass(MVT::f32, &AMDGPU::R600_Reg32RegClass);
addRegisterClass(MVT::v4i32, &AMDGPU::R600_Reg128RegClass);
addRegisterClass(MVT::i32, &AMDGPU::R600_Reg32RegClass);
+ addRegisterClass(MVT::v2f32, &AMDGPU::R600_Reg64RegClass);
+ addRegisterClass(MVT::v2i32, &AMDGPU::R600_Reg64RegClass);
+
computeRegisterProperties();
setOperationAction(ISD::FADD, MVT::v4f32, Expand);
+ setOperationAction(ISD::FADD, MVT::v2f32, Expand);
setOperationAction(ISD::FMUL, MVT::v4f32, Expand);
+ setOperationAction(ISD::FMUL, MVT::v2f32, Expand);
setOperationAction(ISD::FDIV, MVT::v4f32, Expand);
+ setOperationAction(ISD::FDIV, MVT::v2f32, Expand);
setOperationAction(ISD::FSUB, MVT::v4f32, Expand);
+ setOperationAction(ISD::FSUB, MVT::v2f32, Expand);
setOperationAction(ISD::FCOS, MVT::f32, Custom);
setOperationAction(ISD::FSIN, MVT::f32, Custom);
setOperationAction(ISD::SETCC, MVT::v4i32, Expand);
+ setOperationAction(ISD::SETCC, MVT::v2i32, Expand);
setOperationAction(ISD::BR_CC, MVT::i32, Expand);
setOperationAction(ISD::BR_CC, MVT::f32, Expand);
// Legalize loads and stores to the private address space.
setOperationAction(ISD::LOAD, MVT::i32, Custom);
- setOperationAction(ISD::LOAD, MVT::v2i32, Expand);
+ setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Custom);
setLoadExtAction(ISD::SEXTLOAD, MVT::i16, Custom);
setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Custom);
setOperationAction(ISD::STORE, MVT::i8, Custom);
setOperationAction(ISD::STORE, MVT::i32, Custom);
- setOperationAction(ISD::STORE, MVT::v2i32, Expand);
+ setOperationAction(ISD::STORE, MVT::v2i32, Custom);
setOperationAction(ISD::STORE, MVT::v4i32, Custom);
setOperationAction(ISD::LOAD, MVT::i32, Custom);
}
case AMDGPU::RAT_WRITE_CACHELESS_32_eg:
+ case AMDGPU::RAT_WRITE_CACHELESS_64_eg:
case AMDGPU::RAT_WRITE_CACHELESS_128_eg: {
unsigned EOP = (llvm::next(I)->getOpcode() == AMDGPU::RETURN) ? 1 : 0;
DAG.getConstant(4 * i + ConstantBlock * 16, MVT::i32));
Slots[i] = DAG.getNode(AMDGPUISD::CONST_ADDRESS, DL, MVT::i32, NewPtr);
}
- Result = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v4i32, Slots, 4);
+ EVT NewVT = MVT::v4i32;
+ unsigned NumElements = 4;
+ if (VT.isVector()) {
+ NewVT = VT;
+ NumElements = VT.getVectorNumElements();
+ }
+ Result = DAG.getNode(ISD::BUILD_VECTOR, DL, NewVT, Slots, NumElements);
} else {
// non constant ptr cant be folded, keeps it as a v4f32 load
Result = DAG.getNode(AMDGPUISD::CONST_ADDRESS, DL, MVT::v4i32,
MachineBasicBlock::iterator MI, DebugLoc DL,
unsigned DestReg, unsigned SrcReg,
bool KillSrc) const {
- if (AMDGPU::R600_Reg128RegClass.contains(DestReg)
- && AMDGPU::R600_Reg128RegClass.contains(SrcReg)) {
- for (unsigned I = 0; I < 4; I++) {
+ unsigned VectorComponents = 0;
+ if (AMDGPU::R600_Reg128RegClass.contains(DestReg) &&
+ AMDGPU::R600_Reg128RegClass.contains(SrcReg)) {
+ VectorComponents = 4;
+ } else if(AMDGPU::R600_Reg64RegClass.contains(DestReg) &&
+ AMDGPU::R600_Reg64RegClass.contains(SrcReg)) {
+ VectorComponents = 2;
+ }
+
+ if (VectorComponents > 0) {
+ for (unsigned I = 0; I < VectorComponents; I++) {
unsigned SubRegIndex = RI.getSubRegFromChannel(I);
buildDefaultInstruction(MBB, MI, AMDGPU::MOV,
RI.getSubReg(DestReg, SubRegIndex),
RegState::Define | RegState::Implicit);
}
} else {
-
- // We can't copy vec4 registers
- assert(!AMDGPU::R600_Reg128RegClass.contains(DestReg)
- && !AMDGPU::R600_Reg128RegClass.contains(SrcReg));
-
MachineInstr *NewMI = buildDefaultInstruction(MBB, MI, AMDGPU::MOV,
DestReg, SrcReg);
NewMI->getOperand(getOperandIdx(*NewMI, AMDGPU::OpName::src0))
[(global_store i32:$rw_gpr, i32:$index_gpr)]
>;
+// 64-bit store
+def RAT_WRITE_CACHELESS_64_eg : RAT_WRITE_CACHELESS_eg <
+ (ins R600_Reg64:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
+ 0x3, "RAT_WRITE_CACHELESS_64_eg $rw_gpr.XY, $index_gpr, $eop",
+ [(global_store v2i32:$rw_gpr, i32:$index_gpr)]
+>;
+
//128-bit store
def RAT_WRITE_CACHELESS_128_eg : RAT_WRITE_CACHELESS_eg <
(ins R600_Reg128:$rw_gpr, R600_TReg32_X:$index_gpr, InstFlag:$eop),
let Constraints = "$src_gpr.ptr = $dst_gpr";
}
+class VTX_READ_64_eg <bits<8> buffer_id, list<dag> pattern>
+ : VTX_READ_eg <"VTX_READ_64 $dst_gpr.XY, $src_gpr", buffer_id,
+ (outs R600_Reg64:$dst_gpr), pattern> {
+
+ let MEGA_FETCH_COUNT = 8;
+ let DST_SEL_X = 0;
+ let DST_SEL_Y = 1;
+ let DST_SEL_Z = 7;
+ let DST_SEL_W = 7;
+ let DATA_FORMAT = 0x1D; // COLOR_32_32
+}
+
class VTX_READ_128_eg <bits<8> buffer_id, list<dag> pattern>
: VTX_READ_eg <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
(outs R600_Reg128:$dst_gpr), pattern> {
[(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
>;
+def VTX_READ_PARAM_64_eg : VTX_READ_64_eg <0,
+ [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
def VTX_READ_PARAM_128_eg : VTX_READ_128_eg <0,
[(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
>;
[(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
>;
+// 64-bit reads
+def VTX_READ_GLOBAL_64_eg : VTX_READ_64_eg <1,
+ [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
// 128-bit reads
def VTX_READ_GLOBAL_128_eg : VTX_READ_128_eg <1,
[(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
def : Pat<(fsqrt f32:$src), (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_cm $src))>;
-def RAT_STORE_DWORD_cm : EG_CF_RAT <
- 0x57, 0x14, 0x1, (outs),
- (ins R600_TReg32_X:$rw_gpr, R600_TReg32_X:$index_gpr),
- "EXPORT_RAT_INST_STORE_DWORD $rw_gpr, $index_gpr",
- [(global_store i32:$rw_gpr, i32:$index_gpr)]
+class RAT_STORE_DWORD_cm <bits<4> mask, dag ins, list<dag> pat> : EG_CF_RAT <
+ 0x57, 0x14, mask, (outs), ins,
+ "EXPORT_RAT_INST_STORE_DWORD $rw_gpr, $index_gpr", pat
> {
let eop = 0; // This bit is not used on Cayman.
}
+def RAT_STORE_DWORD32_cm : RAT_STORE_DWORD_cm <0x1,
+ (ins R600_TReg32_X:$rw_gpr, R600_TReg32_X:$index_gpr),
+ [(global_store i32:$rw_gpr, i32:$index_gpr)]
+>;
+
+def RAT_STORE_DWORD64_cm : RAT_STORE_DWORD_cm <0x3,
+ (ins R600_Reg64:$rw_gpr, R600_TReg32_X:$index_gpr),
+ [(global_store v2i32:$rw_gpr, i32:$index_gpr)]
+>;
+
class VTX_READ_cm <string name, bits<8> buffer_id, dag outs, list<dag> pattern>
: VTX_WORD0_cm, VTX_READ<name, buffer_id, outs, pattern> {
let Constraints = "$src_gpr.ptr = $dst_gpr";
}
+class VTX_READ_64_cm <bits<8> buffer_id, list<dag> pattern>
+ : VTX_READ_cm <"VTX_READ_64 $dst_gpr, $src_gpr", buffer_id,
+ (outs R600_Reg64:$dst_gpr), pattern> {
+
+ let DST_SEL_X = 0;
+ let DST_SEL_Y = 1;
+ let DST_SEL_Z = 7;
+ let DST_SEL_W = 7;
+ let DATA_FORMAT = 0x1D; // COLOR_32_32
+}
+
class VTX_READ_128_cm <bits<8> buffer_id, list<dag> pattern>
: VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr", buffer_id,
(outs R600_Reg128:$dst_gpr), pattern> {
[(set i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
>;
+def VTX_READ_PARAM_64_cm : VTX_READ_64_cm <0,
+ [(set v2i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
+>;
+
def VTX_READ_PARAM_128_cm : VTX_READ_128_cm <0,
[(set v4i32:$dst_gpr, (load_param ADDRVTX_READ:$src_gpr))]
>;
[(set i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
>;
+// 64-bit reads
+def VTX_READ_GLOBAL_64_cm : VTX_READ_64_cm <1,
+ [(set v2i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
+>;
+
// 128-bit reads
def VTX_READ_GLOBAL_128_cm : VTX_READ_128_cm <1,
[(set v4i32:$dst_gpr, (global_load ADDRVTX_READ:$src_gpr))]
def : Vector4_Build <v4f32, f32>;
def : Vector4_Build <v4i32, i32>;
+def : Extract_Element <f32, v2f32, 0, sub0>;
+def : Extract_Element <f32, v2f32, 1, sub1>;
+
+def : Insert_Element <f32, v2f32, 0, sub0>;
+def : Insert_Element <f32, v2f32, 1, sub1>;
+
+def : Extract_Element <i32, v2i32, 0, sub0>;
+def : Extract_Element <i32, v2i32, 1, sub1>;
+
+def : Insert_Element <i32, v2i32, 0, sub0>;
+def : Insert_Element <i32, v2i32, 1, sub1>;
+
// bitconvert patterns
def : BitConvert <i32, f32, R600_Reg32>;
def : BitConvert <f32, i32, R600_Reg32>;
+def : BitConvert <v2f32, v2i32, R600_Reg64>;
+def : BitConvert <v2i32, v2f32, R600_Reg64>;
def : BitConvert <v4f32, v4i32, R600_Reg128>;
def : BitConvert <v4i32, v4f32, R600_Reg128>;
let HWEncoding = encoding;
}
+class R600Reg_64<string n, list<Register> subregs, bits<16> encoding> :
+ RegisterWithSubRegs<n, subregs> {
+ let Namespace = "AMDGPU";
+ let SubRegIndices = [sub0, sub1];
+ let HWEncoding = encoding;
+}
+
+
foreach Index = 0-127 in {
foreach Chan = [ "X", "Y", "Z", "W" ] in {
// 32-bit Temporary Registers
!cast<Register>("T"#Index#"_Z"),
!cast<Register>("T"#Index#"_W")],
Index>;
+
+ def T#Index#_XY : R600Reg_64 <"T"#Index#"",
+ [!cast<Register>("T"#Index#"_X"),
+ !cast<Register>("T"#Index#"_Y")],
+ Index>;
}
// KCACHE_BANK0
let CopyCost = -1;
}
+def R600_Reg64 : RegisterClass<"AMDGPU", [v2f32, v2i32], 64,
+ (add (sequence "T%u_XY", 0, 63))>;
+
//===----------------------------------------------------------------------===//
// Register classes for indirect addressing
//===----------------------------------------------------------------------===//
-; RUN: llc < %s -march=r600 -mcpu=tahiti | FileCheck %s
+; RUN: llc < %s -march=r600 -mcpu=tahiti | FileCheck %s --check-prefix=SI-CHECK
; SI-CHECK: @f64_kernel_arg
; SI-CHECK-DAG: S_LOAD_DWORDX2 SGPR{{[0-9]}}_SGPR{{[0-9]}}, SGPR0_SGPR1, 9
--- /dev/null
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
+; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
+
+; R600-CHECK: @build_vector2
+; R600-CHECK: MOV
+; R600-CHECK: MOV
+; R600-CHECK-NOT: MOV
+; SI-CHECK: @build_vector2
+; SI-CHECK-DAG: V_MOV_B32_e32 [[X:VGPR[0-9]]], 5
+; SI-CHECK-DAG: V_MOV_B32_e32 [[Y:VGPR[0-9]]], 6
+; SI-CHECK: BUFFER_STORE_DWORDX2 [[X]]_[[Y]]
+define void @build_vector2 (<2 x i32> addrspace(1)* %out) {
+entry:
+ store <2 x i32> <i32 5, i32 6>, <2 x i32> addrspace(1)* %out
+ ret void
+}
+
+; R600-CHECK: @build_vector4
+; R600-CHECK: MOV
+; R600-CHECK: MOV
+; R600-CHECK: MOV
+; R600-CHECK: MOV
+; R600-CHECK-NOT: MOV
+; SI-CHECK: @build_vector4
+; SI-CHECK-DAG: V_MOV_B32_e32 [[X:VGPR[0-9]]], 5
+; SI-CHECK-DAG: V_MOV_B32_e32 [[Y:VGPR[0-9]]], 6
+; SI-CHECK-DAG: V_MOV_B32_e32 [[Z:VGPR[0-9]]], 7
+; SI-CHECK-DAG: V_MOV_B32_e32 [[W:VGPR[0-9]]], 8
+; SI-CHECK: BUFFER_STORE_DWORDX4 [[X]]_[[Y]]_[[Z]]_[[W]]
+define void @build_vector4 (<4 x i32> addrspace(1)* %out) {
+entry:
+ store <4 x i32> <i32 5, i32 6, i32 7, i32 8>, <4 x i32> addrspace(1)* %out
+ ret void
+}
declare void @llvm.AMDGPU.store.output(float, i32)
+; CHECK: @fadd_v2f32
+; CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[3].X, KC0[3].Z
+; CHECK-DAG: ADD * T{{[0-9]\.[XYZW]}}, KC0[2].W, KC0[3].Y
+define void @fadd_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
+entry:
+ %0 = fadd <2 x float> %a, %b
+ store <2 x float> %0, <2 x float> addrspace(1)* %out
+ ret void
+}
+
; CHECK: @fadd_v4f32
; CHECK: ADD T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
-;CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}
-;CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+; These tests check that fdiv is expanded correctly and also test that the
+; scheduler is scheduling the RECIP_IEEE and MUL_IEEE instructions in separate
+; instruction groups.
-define void @test(<4 x float> addrspace(1)* %out, <4 x float> %a, <4 x float> %b) {
+; CHECK: @fdiv_v2f32
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Z
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[3].Y
+; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW]}}, KC0[3].X, PS
+; CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW]}}, KC0[2].W, PS
+define void @fdiv_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
entry:
- %0 = fdiv <4 x float> %a, %b
- store <4 x float> %0, <4 x float> addrspace(1)* %out
+ %0 = fdiv <2 x float> %a, %b
+ store <2 x float> %0, <2 x float> addrspace(1)* %out
+ ret void
+}
+
+; CHECK: @fdiv_v4f32
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK-DAG: RECIP_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; CHECK-DAG: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+; CHECK-DAG: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}, PS
+
+define void @fdiv_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
+ %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
+ %a = load <4 x float> addrspace(1) * %in
+ %b = load <4 x float> addrspace(1) * %b_ptr
+ %result = fdiv <4 x float> %a, %b
+ store <4 x float> %result, <4 x float> addrspace(1)* %out
ret void
}
declare void @llvm.AMDGPU.store.output(float, i32)
+; CHECK: @fmul_v2f32
+; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW]}}
+define void @fmul_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
+entry:
+ %0 = fmul <2 x float> %a, %b
+ store <2 x float> %0, <2 x float> addrspace(1)* %out
+ ret void
+}
+
; CHECK: @fmul_v4f32
; CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; CHECK: MUL_IEEE * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
+; R600-CHECK: @fp_to_sint_v2i32
+; R600-CHECK: FLT_TO_INT T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
+; R600-CHECK: FLT_TO_INT * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
+; SI-CHECK: @fp_to_sint_v2i32
+; SI-CHECK: V_CVT_I32_F32_e32
+; SI-CHECK: V_CVT_I32_F32_e32
+define void @fp_to_sint_v2i32(<2 x i32> addrspace(1)* %out, <2 x float> %in) {
+ %result = fptosi <2 x float> %in to <2 x i32>
+ store <2 x i32> %result, <2 x i32> addrspace(1)* %out
+ ret void
+}
+
; R600-CHECK: @fp_to_sint_v4i32
; R600-CHECK: FLT_TO_INT {{[* ]*}}T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
; R600-CHECK: FLT_TO_INT {{[* ]*}}T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
; SI-CHECK: V_CVT_I32_F32_e32
; SI-CHECK: V_CVT_I32_F32_e32
; SI-CHECK: V_CVT_I32_F32_e32
-
define void @fp_to_sint_v4i32(<4 x i32> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
%value = load <4 x float> addrspace(1) * %in
%result = fptosi <4 x float> %value to <4 x i32>
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+; CHECK: @fp_to_uint_v2i32
+; CHECK: FLT_TO_UINT * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
+; CHECK: FLT_TO_UINT * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
+
+define void @fp_to_uint_v2i32(<2 x i32> addrspace(1)* %out, <2 x float> %in) {
+ %result = fptoui <2 x float> %in to <2 x i32>
+ store <2 x i32> %result, <2 x i32> addrspace(1)* %out
+ ret void
+}
+
; CHECK: @fp_to_uint_v4i32
; CHECK: FLT_TO_UINT * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
; CHECK: FLT_TO_UINT * T{{[0-9]+\.[XYZW], PV\.[XYZW]}}
declare void @llvm.AMDGPU.store.output(float, i32)
-; CHECK: @fsub_v4f32
-; CHECK: ADD T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: @fsub_v2f32
+; CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[3].X, -KC0[3].Z
+; CHECK-DAG: ADD * T{{[0-9]+\.[XYZW]}}, KC0[2].W, -KC0[3].Y
+define void @fsub_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %a, <2 x float> %b) {
+entry:
+ %0 = fsub <2 x float> %a, %b
+ store <2 x float> %0, <2 x float> addrspace(1)* %out
+ ret void
+}
+; CHECK: @fsub_v4f32
+; CHECK: ADD T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
+; CHECK: ADD * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], -T[0-9]+\.[XYZW]}}
define void @fsub_v4f32(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
%b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1
%a = load <4 x float> addrspace(1) * %in
; load a v2f32 value from the global address space
; R600-CHECK: @load_v2f32
-; R600-CHECK: VTX_READ_32
-; R600-CHECK: VTX_READ_32
+; R600-CHECK: VTX_READ_64
; SI-CHECK: @load_v2f32
; SI-CHECK: BUFFER_LOAD_DWORDX2
; load a v2i32 value from the global address space.
; EG-CHECK: @load_v2i32
-; EG-CHECK-DAG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 4
-; EG-CHECK-DAG: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
+; EG-CHECK: VTX_READ_64 T{{[0-9]+}}.XY, T{{[0-9]+}}.X, 0
; SI-CHECK: @load_v2i32
; SI-CHECK: BUFFER_LOAD_DWORDX2 VGPR{{[0-9]+}}
define void @load_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG-CHECK %s
-;EG-CHECK: @test2
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: @setcc_v2i32
+; EG-CHECK-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[3].X, KC0[3].Z
+; EG-CHECK-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[2].W, KC0[3].Y
-define void @test2(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
- %b_ptr = getelementptr <2 x i32> addrspace(1)* %in, i32 1
- %a = load <2 x i32> addrspace(1) * %in
- %b = load <2 x i32> addrspace(1) * %b_ptr
+define void @setcc_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b) {
%result = icmp eq <2 x i32> %a, %b
%sext = sext <2 x i1> %result to <2 x i32>
store <2 x i32> %sext, <2 x i32> addrspace(1)* %out
ret void
}
-;EG-CHECK: @test4
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;EG-CHECK: SETE_INT {{\*? *}}T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; CHECK: @setcc_v4i32
+; EG-CHECK-DAG: SETE_INT T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG-CHECK-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG-CHECK-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+; EG-CHECK-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-define void @test4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
+define void @setcc_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
%b_ptr = getelementptr <4 x i32> addrspace(1)* %in, i32 1
%a = load <4 x i32> addrspace(1) * %in
%b = load <4 x i32> addrspace(1) * %b_ptr
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
+; R600-CHECK: @sint_to_fp_v2i32
+; R600-CHECK-DAG: INT_TO_FLT * T{{[0-9]+\.[XYZW]}}, KC0[2].W
+; R600-CHECK-DAG: INT_TO_FLT * T{{[0-9]+\.[XYZW]}}, KC0[3].X
+; SI-CHECK: @sint_to_fp_v2i32
+; SI-CHECK: V_CVT_F32_I32_e32
+; SI-CHECK: V_CVT_F32_I32_e32
+define void @sint_to_fp_v2i32(<2 x float> addrspace(1)* %out, <2 x i32> %in) {
+ %result = sitofp <2 x i32> %in to <2 x float>
+ store <2 x float> %result, <2 x float> addrspace(1)* %out
+ ret void
+}
+
; R600-CHECK: @sint_to_fp_v4i32
; R600-CHECK: INT_TO_FLT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; R600-CHECK: INT_TO_FLT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; vec2 floating-point stores
; EG-CHECK: @store_v2f32
-; EG-CHECK: RAT_WRITE_CACHELESS_32_eg
-; EG-CHECK-NEXT: RAT_WRITE_CACHELESS_32_eg
+; EG-CHECK: RAT_WRITE_CACHELESS_64_eg
; CM-CHECK: @store_v2f32
; CM-CHECK: EXPORT_RAT_INST_STORE_DWORD
-; CM-CHECK-NEXT: EXPORT_RAT_INST_STORE_DWORD
; SI-CHECK: @store_v2f32
; SI-CHECK: BUFFER_STORE_DWORDX2
; be two 32-bit stores.
; EG-CHECK: @vecload2
-; EG-CHECK: RAT_WRITE_CACHELESS_32_eg
-; EG-CHECK: RAT_WRITE_CACHELESS_32_eg
+; EG-CHECK: RAT_WRITE_CACHELESS_64_eg
; CM-CHECK: @vecload2
; CM-CHECK: EXPORT_RAT_INST_STORE_DWORD
-; CM-CHECK: EXPORT_RAT_INST_STORE_DWORD
; SI-CHECK: @vecload2
; SI-CHECK: BUFFER_STORE_DWORDX2
define void @vecload2(i32 addrspace(1)* nocapture %out, i32 addrspace(2)* nocapture %mem) #0 {
;EG-CHECK: @test2
;EG-CHECK: SUB_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;EG-CHECK: SUB_INT T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;EG-CHECK: SUB_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
;SI-CHECK: @test2
;SI-CHECK: V_SUB_I32_e32 VGPR{{[0-9]+, VGPR[0-9]+, VGPR[0-9]+}}
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=R600-CHECK
; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
+; R600-CHECK: @uint_to_fp_v2i32
+; R600-CHECK-DAG: UINT_TO_FLT * T{{[0-9]+\.[XYZW]}}, KC0[2].W
+; R600-CHECK-DAG: UINT_TO_FLT * T{{[0-9]+\.[XYZW]}}, KC0[3].X
+; SI-CHECK: @uint_to_fp_v2i32
+; SI-CHECK: V_CVT_F32_U32_e32
+; SI-CHECK: V_CVT_F32_U32_e32
+define void @uint_to_fp_v2i32(<2 x float> addrspace(1)* %out, <2 x i32> %in) {
+ %result = uitofp <2 x i32> %in to <2 x float>
+ store <2 x float> %result, <2 x float> addrspace(1)* %out
+ ret void
+}
+
; R600-CHECK: @uint_to_fp_v4i32
; R600-CHECK: UINT_TO_FLT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
; R600-CHECK: UINT_TO_FLT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
projects / oota-llvm.git / commitdiff