From d344c6bcf9490ca6d2460a7fe9eaefd2c961761f Mon Sep 17 00:00:00 2001 From: Matt Arsenault Date: Fri, 13 Jun 2014 04:00:30 +0000 Subject: [PATCH] R600/SI: Fix selection error on i64 rotl / rotr. Evergreen is still broken due to missing shl_parts. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210885 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/R600/AMDGPUISelLowering.cpp | 10 ++-- lib/Target/R600/R600ISelLowering.h | 3 -- test/CodeGen/R600/rotl.i64.ll | 34 +++++++++++++ test/CodeGen/R600/rotl.ll | 54 +++++++++++++++++++++ test/CodeGen/R600/rotr.ll | 67 ++++++++++++++++---------- 5 files changed, 136 insertions(+), 32 deletions(-) create mode 100644 test/CodeGen/R600/rotl.i64.ll create mode 100644 test/CodeGen/R600/rotl.ll diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp index 6309a6d312c..9b066d206b5 100644 --- a/lib/Target/R600/AMDGPUISelLowering.cpp +++ b/lib/Target/R600/AMDGPUISelLowering.cpp @@ -126,8 +126,10 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) : setOperationAction(ISD::FROUND, MVT::f32, Legal); setOperationAction(ISD::FTRUNC, MVT::f32, Legal); - // The hardware supports ROTR, but not ROTL + // The hardware supports 32-bit ROTR, but not ROTL. setOperationAction(ISD::ROTL, MVT::i32, Expand); + setOperationAction(ISD::ROTL, MVT::i64, Expand); + setOperationAction(ISD::ROTR, MVT::i64, Expand); // Lower floating point store/load to integer store/load to reduce the number // of patterns in tablegen. @@ -264,11 +266,13 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) : setOperationAction(ISD::MUL, VT, Expand); setOperationAction(ISD::OR, VT, Expand); setOperationAction(ISD::SHL, VT, Expand); - setOperationAction(ISD::SINT_TO_FP, VT, Expand); - setOperationAction(ISD::SRL, VT, Expand); setOperationAction(ISD::SRA, VT, Expand); + setOperationAction(ISD::SRL, VT, Expand); + setOperationAction(ISD::ROTL, VT, Expand); + setOperationAction(ISD::ROTR, VT, Expand); setOperationAction(ISD::SUB, VT, Expand); setOperationAction(ISD::UDIV, VT, Expand); + setOperationAction(ISD::SINT_TO_FP, VT, Expand); setOperationAction(ISD::UINT_TO_FP, VT, Expand); setOperationAction(ISD::UREM, VT, Expand); setOperationAction(ISD::SELECT, VT, Expand); diff --git a/lib/Target/R600/R600ISelLowering.h b/lib/Target/R600/R600ISelLowering.h index a8a464f338f..301509fc8cf 100644 --- a/lib/Target/R600/R600ISelLowering.h +++ b/lib/Target/R600/R600ISelLowering.h @@ -52,9 +52,6 @@ private: MachineRegisterInfo & MRI, unsigned dword_offset) const; SDValue OptimizeSwizzle(SDValue BuildVector, SDValue Swz[], SelectionDAG &DAG) const; - /// \brief Lower ROTL opcode to BITALIGN - SDValue LowerROTL(SDValue Op, SelectionDAG &DAG) const; - SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const; SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const; SDValue LowerFPTOUINT(SDValue Op, SelectionDAG &DAG) const; diff --git a/test/CodeGen/R600/rotl.i64.ll b/test/CodeGen/R600/rotl.i64.ll new file mode 100644 index 00000000000..bda0b6694a8 --- /dev/null +++ b/test/CodeGen/R600/rotl.i64.ll @@ -0,0 +1,34 @@ +; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s + +; FUNC-LABEL: @s_rotl_i64: +; SI: S_LSHL_B64 +; SI: S_SUB_I32 +; SI: S_LSHR_B64 +; SI: S_OR_B64 +define void @s_rotl_i64(i64 addrspace(1)* %in, i64 %x, i64 %y) { +entry: + %0 = shl i64 %x, %y + %1 = sub i64 64, %y + %2 = lshr i64 %x, %1 + %3 = or i64 %0, %2 + store i64 %3, i64 addrspace(1)* %in + ret void +} + +; FUNC-LABEL: @v_rotl_i64: +; SI: V_LSHL_B64 +; SI: V_SUB_I32 +; SI: V_LSHR_B64 +; SI: V_OR_B32 +; SI: V_OR_B32 +define void @v_rotl_i64(i64 addrspace(1)* %in, i64 addrspace(1)* %xptr, i64 addrspace(1)* %yptr) { +entry: + %x = load i64 addrspace(1)* %xptr, align 8 + %y = load i64 addrspace(1)* %yptr, align 8 + %tmp0 = shl i64 %x, %y + %tmp1 = sub i64 64, %y + %tmp2 = lshr i64 %x, %tmp1 + %tmp3 = or i64 %tmp0, %tmp2 + store i64 %tmp3, i64 addrspace(1)* %in, align 8 + ret void +} diff --git a/test/CodeGen/R600/rotl.ll b/test/CodeGen/R600/rotl.ll new file mode 100644 index 00000000000..83f657fd4cc --- /dev/null +++ b/test/CodeGen/R600/rotl.ll @@ -0,0 +1,54 @@ +; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck --check-prefix=R600 -check-prefix=FUNC %s +; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s + +; FUNC-LABEL: @rotl_i32: +; R600: SUB_INT {{\** T[0-9]+\.[XYZW]}}, literal.x +; R600-NEXT: 32 +; R600: BIT_ALIGN_INT {{T[0-9]+\.[XYZW]}}, KC0[2].Z, KC0[2].Z, PV.{{[XYZW]}} + +; SI: S_SUB_I32 [[SDST:s[0-9]+]], 32, {{[s][0-9]+}} +; SI: V_MOV_B32_e32 [[VDST:v[0-9]+]], [[SDST]] +; SI: V_ALIGNBIT_B32 {{v[0-9]+, [s][0-9]+, v[0-9]+}}, [[VDST]] +define void @rotl_i32(i32 addrspace(1)* %in, i32 %x, i32 %y) { +entry: + %0 = shl i32 %x, %y + %1 = sub i32 32, %y + %2 = lshr i32 %x, %1 + %3 = or i32 %0, %2 + store i32 %3, i32 addrspace(1)* %in + ret void +} + +; FUNC-LABEL: @rotl_v2i32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +define void @rotl_v2i32(<2 x i32> addrspace(1)* %in, <2 x i32> %x, <2 x i32> %y) { +entry: + %0 = shl <2 x i32> %x, %y + %1 = sub <2 x i32> , %y + %2 = lshr <2 x i32> %x, %1 + %3 = or <2 x i32> %0, %2 + store <2 x i32> %3, <2 x i32> addrspace(1)* %in + ret void +} + +; FUNC-LABEL: @rotl_v4i32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +; SI: S_SUB_I32 +; SI: V_ALIGNBIT_B32 +define void @rotl_v4i32(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y) { +entry: + %0 = shl <4 x i32> %x, %y + %1 = sub <4 x i32> , %y + %2 = lshr <4 x i32> %x, %1 + %3 = or <4 x i32> %0, %2 + store <4 x i32> %3, <4 x i32> addrspace(1)* %in + ret void +} diff --git a/test/CodeGen/R600/rotr.ll b/test/CodeGen/R600/rotr.ll index edf7aeebea0..a5a4da48073 100644 --- a/test/CodeGen/R600/rotr.ll +++ b/test/CodeGen/R600/rotr.ll @@ -1,37 +1,52 @@ -; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=R600-CHECK %s -; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck --check-prefix=SI-CHECK %s +; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck --check-prefix=R600 -check-prefix=FUNC %s +; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s -; R600-CHECK-LABEL: @rotr: -; R600-CHECK: BIT_ALIGN_INT +; FUNC-LABEL: @rotr_i32: +; R600: BIT_ALIGN_INT -; SI-CHECK-LABEL: @rotr: -; SI-CHECK: V_ALIGNBIT_B32 -define void @rotr(i32 addrspace(1)* %in, i32 %x, i32 %y) { +; SI: V_ALIGNBIT_B32 +define void @rotr_i32(i32 addrspace(1)* %in, i32 %x, i32 %y) { entry: - %0 = sub i32 32, %y - %1 = shl i32 %x, %0 - %2 = lshr i32 %x, %y - %3 = or i32 %1, %2 - store i32 %3, i32 addrspace(1)* %in + %tmp0 = sub i32 32, %y + %tmp1 = shl i32 %x, %tmp0 + %tmp2 = lshr i32 %x, %y + %tmp3 = or i32 %tmp1, %tmp2 + store i32 %tmp3, i32 addrspace(1)* %in ret void } -; R600-CHECK-LABEL: @rotl: -; R600-CHECK: SUB_INT {{\** T[0-9]+\.[XYZW]}}, literal.x -; R600-CHECK-NEXT: 32 -; R600-CHECK: BIT_ALIGN_INT {{T[0-9]+\.[XYZW]}}, KC0[2].Z, KC0[2].Z, PV.{{[XYZW]}} +; FUNC-LABEL: @rotr_v2i32: +; R600: BIT_ALIGN_INT +; R600: BIT_ALIGN_INT +; SI: V_ALIGNBIT_B32 +; SI: V_ALIGNBIT_B32 +define void @rotr_v2i32(<2 x i32> addrspace(1)* %in, <2 x i32> %x, <2 x i32> %y) { +entry: + %tmp0 = sub <2 x i32> , %y + %tmp1 = shl <2 x i32> %x, %tmp0 + %tmp2 = lshr <2 x i32> %x, %y + %tmp3 = or <2 x i32> %tmp1, %tmp2 + store <2 x i32> %tmp3, <2 x i32> addrspace(1)* %in + ret void +} + +; FUNC-LABEL: @rotr_v4i32: +; R600: BIT_ALIGN_INT +; R600: BIT_ALIGN_INT +; R600: BIT_ALIGN_INT +; R600: BIT_ALIGN_INT -; SI-CHECK-LABEL: @rotl: -; SI-CHECK: S_SUB_I32 [[SDST:s[0-9]+]], 32, {{[s][0-9]+}} -; SI-CHECK: V_MOV_B32_e32 [[VDST:v[0-9]+]], [[SDST]] -; SI-CHECK: V_ALIGNBIT_B32 {{v[0-9]+, [s][0-9]+, v[0-9]+}}, [[VDST]] -define void @rotl(i32 addrspace(1)* %in, i32 %x, i32 %y) { +; SI: V_ALIGNBIT_B32 +; SI: V_ALIGNBIT_B32 +; SI: V_ALIGNBIT_B32 +; SI: V_ALIGNBIT_B32 +define void @rotr_v4i32(<4 x i32> addrspace(1)* %in, <4 x i32> %x, <4 x i32> %y) { entry: - %0 = shl i32 %x, %y - %1 = sub i32 32, %y - %2 = lshr i32 %x, %1 - %3 = or i32 %0, %2 - store i32 %3, i32 addrspace(1)* %in + %tmp0 = sub <4 x i32> , %y + %tmp1 = shl <4 x i32> %x, %tmp0 + %tmp2 = lshr <4 x i32> %x, %y + %tmp3 = or <4 x i32> %tmp1, %tmp2 + store <4 x i32> %tmp3, <4 x i32> addrspace(1)* %in ret void } -- 2.34.1