diff --git a/lib/Target/Mips/MipsISelDAGToDAG.cpp b/lib/Target/Mips/MipsISelDAGToDAG.cpp index 83763a64ab6..0e1173f1c61 100644 --- a/lib/Target/Mips/MipsISelDAGToDAG.cpp +++ b/lib/Target/Mips/MipsISelDAGToDAG.cpp @@ -108,8 +108,26 @@ bool MipsDAGToDAGISel::selectIntAddrLSL2MM(SDValue Addr, SDValue &Base, return false; } -bool MipsDAGToDAGISel::selectIntAddrMSA(SDValue Addr, SDValue &Base, - SDValue &Offset) const { +bool MipsDAGToDAGISel::selectIntAddrSImm10(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + llvm_unreachable("Unimplemented function."); + return false; +} + +bool MipsDAGToDAGISel::selectIntAddrSImm10Lsl1(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + llvm_unreachable("Unimplemented function."); + return false; +} + +bool MipsDAGToDAGISel::selectIntAddrSImm10Lsl2(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + llvm_unreachable("Unimplemented function."); + return false; +} + +bool MipsDAGToDAGISel::selectIntAddrSImm10Lsl3(SDValue Addr, SDValue &Base, + SDValue &Offset) const { llvm_unreachable("Unimplemented function."); return false; } diff --git a/lib/Target/Mips/MipsISelDAGToDAG.h b/lib/Target/Mips/MipsISelDAGToDAG.h index 289832a8064..b111397ba58 100644 --- a/lib/Target/Mips/MipsISelDAGToDAG.h +++ b/lib/Target/Mips/MipsISelDAGToDAG.h @@ -78,8 +78,17 @@ private: SDValue &Offset) const; /// Match addr+simm10 and addr - virtual bool selectIntAddrMSA(SDValue Addr, SDValue &Base, - SDValue &Offset) const; + virtual bool selectIntAddrSImm10(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectIntAddrSImm10Lsl1(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectIntAddrSImm10Lsl2(SDValue Addr, SDValue &Base, + SDValue &Offset) const; + + virtual bool selectIntAddrSImm10Lsl3(SDValue Addr, SDValue &Base, + SDValue &Offset) const; virtual bool selectAddr16(SDValue Addr, SDValue &Base, SDValue &Offset); virtual bool selectAddr16SP(SDValue Addr, SDValue &Base, SDValue &Offset); diff --git a/lib/Target/Mips/MipsInstrInfo.td b/lib/Target/Mips/MipsInstrInfo.td index 1f3a1538a1a..43e4818e264 100644 --- a/lib/Target/Mips/MipsInstrInfo.td +++ b/lib/Target/Mips/MipsInstrInfo.td @@ -1098,7 +1098,13 @@ def addrRegImm : def addrDefault : ComplexPattern; -def addrimm10 : ComplexPattern; +def addrimm10 : ComplexPattern; +def addrimm10lsl1 : ComplexPattern; +def addrimm10lsl2 : ComplexPattern; +def addrimm10lsl3 : ComplexPattern; //===----------------------------------------------------------------------===// // Instructions specific format diff --git a/lib/Target/Mips/MipsMSAInstrInfo.td b/lib/Target/Mips/MipsMSAInstrInfo.td index deb4345e266..f21a2746959 100644 --- a/lib/Target/Mips/MipsMSAInstrInfo.td +++ b/lib/Target/Mips/MipsMSAInstrInfo.td @@ -2308,9 +2308,12 @@ class LD_DESC_BASE; -class LD_H_DESC : LD_DESC_BASE<"ld.h", load, v8i16, MSA128HOpnd, mem_simm10_lsl1>; -class LD_W_DESC : LD_DESC_BASE<"ld.w", load, v4i32, MSA128WOpnd, mem_simm10_lsl2>; -class LD_D_DESC : LD_DESC_BASE<"ld.d", load, v2i64, MSA128DOpnd, mem_simm10_lsl3>; +class LD_H_DESC : LD_DESC_BASE<"ld.h", load, v8i16, MSA128HOpnd, + mem_simm10_lsl1, addrimm10lsl1>; +class LD_W_DESC : LD_DESC_BASE<"ld.w", load, v4i32, MSA128WOpnd, + mem_simm10_lsl2, addrimm10lsl2>; +class LD_D_DESC : LD_DESC_BASE<"ld.d", load, v2i64, MSA128DOpnd, + mem_simm10_lsl3, addrimm10lsl3>; class LDI_B_DESC : MSA_I10_LDI_DESC_BASE<"ldi.b", MSA128BOpnd>; class LDI_H_DESC : MSA_I10_LDI_DESC_BASE<"ldi.h", MSA128HOpnd>; @@ -2641,9 +2644,12 @@ class ST_DESC_BASE; -class ST_H_DESC : ST_DESC_BASE<"st.h", store, v8i16, MSA128HOpnd, mem_simm10_lsl1>; -class ST_W_DESC : ST_DESC_BASE<"st.w", store, v4i32, MSA128WOpnd, mem_simm10_lsl2>; -class ST_D_DESC : ST_DESC_BASE<"st.d", store, v2i64, MSA128DOpnd, mem_simm10_lsl3>; +class ST_H_DESC : ST_DESC_BASE<"st.h", store, v8i16, MSA128HOpnd, + mem_simm10_lsl1, addrimm10lsl1>; +class ST_W_DESC : ST_DESC_BASE<"st.w", store, v4i32, MSA128WOpnd, + mem_simm10_lsl2, addrimm10lsl2>; +class ST_D_DESC : ST_DESC_BASE<"st.d", store, v2i64, MSA128DOpnd, + mem_simm10_lsl3, addrimm10lsl3>; class SUBS_S_B_DESC : MSA_3R_DESC_BASE<"subs_s.b", int_mips_subs_s_b, MSA128BOpnd>; @@ -3523,16 +3529,16 @@ class MSAPat pred = [HasMSA]> : def : MSAPat<(extractelt (v4i32 MSA128W:$ws), immZExt4:$idx), (COPY_S_W MSA128W:$ws, immZExt4:$idx)>; -def : MSAPat<(v8f16 (load addrimm10:$addr)), (LD_H addrimm10:$addr)>; -def : MSAPat<(v4f32 (load addrimm10:$addr)), (LD_W addrimm10:$addr)>; -def : MSAPat<(v2f64 (load addrimm10:$addr)), (LD_D addrimm10:$addr)>; +def : MSAPat<(v8f16 (load addrimm10lsl1:$addr)), (LD_H addrimm10lsl1:$addr)>; +def : MSAPat<(v4f32 (load addrimm10lsl2:$addr)), (LD_W addrimm10lsl2:$addr)>; +def : MSAPat<(v2f64 (load addrimm10lsl3:$addr)), (LD_D addrimm10lsl3:$addr)>; -def ST_FH : MSAPat<(store (v8f16 MSA128H:$ws), addrimm10:$addr), - (ST_H MSA128H:$ws, addrimm10:$addr)>; -def ST_FW : MSAPat<(store (v4f32 MSA128W:$ws), addrimm10:$addr), - (ST_W MSA128W:$ws, addrimm10:$addr)>; -def ST_FD : MSAPat<(store (v2f64 MSA128D:$ws), addrimm10:$addr), - (ST_D MSA128D:$ws, addrimm10:$addr)>; +def ST_FH : MSAPat<(store (v8f16 MSA128H:$ws), addrimm10lsl1:$addr), + (ST_H MSA128H:$ws, addrimm10lsl1:$addr)>; +def ST_FW : MSAPat<(store (v4f32 MSA128W:$ws), addrimm10lsl2:$addr), + (ST_W MSA128W:$ws, addrimm10lsl2:$addr)>; +def ST_FD : MSAPat<(store (v2f64 MSA128D:$ws), addrimm10lsl3:$addr), + (ST_D MSA128D:$ws, addrimm10lsl3:$addr)>; class MSA_FABS_PSEUDO_DESC_BASEisBaseWithConstantOffset(Addr)) { ConstantSDNode *CN = dyn_cast(Addr.getOperand(1)); - if (isIntN(OffsetBits, CN->getSExtValue())) { + if (isIntN(OffsetBits + ShiftAmount, CN->getSExtValue())) { EVT ValTy = Addr.getValueType(); // If the first operand is a FI, get the TargetFI Node - if (FrameIndexSDNode *FIN = dyn_cast - (Addr.getOperand(0))) + if (FrameIndexSDNode *FIN = + dyn_cast(Addr.getOperand(0))) Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), ValTy); - else + else { Base = Addr.getOperand(0); + // If base is a FI, additional offset calculation is done in + // eliminateFrameIndex, otherwise we need to check the alignment + if (OffsetToAlignment(CN->getZExtValue(), 1 << ShiftAmount) != 0) + return false; + } Offset = CurDAG->getTargetConstant(CN->getZExtValue(), SDLoc(Addr), ValTy); @@ -392,17 +397,6 @@ bool MipsSEDAGToDAGISel::selectAddrRegImm9(SDValue Addr, SDValue &Base, return false; } -bool MipsSEDAGToDAGISel::selectAddrRegImm10(SDValue Addr, SDValue &Base, - SDValue &Offset) const { - if (selectAddrFrameIndex(Addr, Base, Offset)) - return true; - - if (selectAddrFrameIndexOffset(Addr, Base, Offset, 10)) - return true; - - return false; -} - /// Used on microMIPS LWC2, LDC2, SWC2 and SDC2 instructions (11-bit offset) bool MipsSEDAGToDAGISel::selectAddrRegImm11(SDValue Addr, SDValue &Base, SDValue &Offset) const { @@ -478,15 +472,49 @@ bool MipsSEDAGToDAGISel::selectIntAddrLSL2MM(SDValue Addr, SDValue &Base, return selectAddrDefault(Addr, Base, Offset); } -bool MipsSEDAGToDAGISel::selectIntAddrMSA(SDValue Addr, SDValue &Base, - SDValue &Offset) const { - if (selectAddrRegImm10(Addr, Base, Offset)) +bool MipsSEDAGToDAGISel::selectIntAddrSImm10(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + + if (selectAddrFrameIndex(Addr, Base, Offset)) return true; - if (selectAddrDefault(Addr, Base, Offset)) + if (selectAddrFrameIndexOffset(Addr, Base, Offset, 10)) return true; - return false; + return selectAddrDefault(Addr, Base, Offset); +} + +bool MipsSEDAGToDAGISel::selectIntAddrSImm10Lsl1(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + if (selectAddrFrameIndex(Addr, Base, Offset)) + return true; + + if (selectAddrFrameIndexOffset(Addr, Base, Offset, 10, 1)) + return true; + + return selectAddrDefault(Addr, Base, Offset); +} + +bool MipsSEDAGToDAGISel::selectIntAddrSImm10Lsl2(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + if (selectAddrFrameIndex(Addr, Base, Offset)) + return true; + + if (selectAddrFrameIndexOffset(Addr, Base, Offset, 10, 2)) + return true; + + return selectAddrDefault(Addr, Base, Offset); +} + +bool MipsSEDAGToDAGISel::selectIntAddrSImm10Lsl3(SDValue Addr, SDValue &Base, + SDValue &Offset) const { + if (selectAddrFrameIndex(Addr, Base, Offset)) + return true; + + if (selectAddrFrameIndexOffset(Addr, Base, Offset, 10, 3)) + return true; + + return selectAddrDefault(Addr, Base, Offset); } // Select constant vector splats. diff --git a/lib/Target/Mips/MipsSEISelDAGToDAG.h b/lib/Target/Mips/MipsSEISelDAGToDAG.h index 0f08b72a334..2a8e5877e84 100644 --- a/lib/Target/Mips/MipsSEISelDAGToDAG.h +++ b/lib/Target/Mips/MipsSEISelDAGToDAG.h @@ -44,7 +44,8 @@ private: bool selectAddrFrameIndex(SDValue Addr, SDValue &Base, SDValue &Offset) const; bool selectAddrFrameIndexOffset(SDValue Addr, SDValue &Base, SDValue &Offset, - unsigned OffsetBits) const; + unsigned OffsetBits, + unsigned ShiftAmount) const; bool selectAddrRegImm(SDValue Addr, SDValue &Base, SDValue &Offset) const override; @@ -58,9 +59,6 @@ private: bool selectAddrRegImm9(SDValue Addr, SDValue &Base, SDValue &Offset) const; - bool selectAddrRegImm10(SDValue Addr, SDValue &Base, - SDValue &Offset) const; - bool selectAddrRegImm11(SDValue Addr, SDValue &Base, SDValue &Offset) const; @@ -82,8 +80,17 @@ private: bool selectIntAddrLSL2MM(SDValue Addr, SDValue &Base, SDValue &Offset) const override; - bool selectIntAddrMSA(SDValue Addr, SDValue &Base, - SDValue &Offset) const override; + bool selectIntAddrSImm10(SDValue Addr, SDValue &Base, + SDValue &Offset) const override; + + bool selectIntAddrSImm10Lsl1(SDValue Addr, SDValue &Base, + SDValue &Offset) const override; + + bool selectIntAddrSImm10Lsl2(SDValue Addr, SDValue &Base, + SDValue &Offset) const override; + + bool selectIntAddrSImm10Lsl3(SDValue Addr, SDValue &Base, + SDValue &Offset) const override; /// \brief Select constant vector splats. bool selectVSplat(SDNode *N, APInt &Imm, diff --git a/test/CodeGen/Mips/msa/i5_ld_st.ll b/test/CodeGen/Mips/msa/i5_ld_st.ll index 991bb8436b3..c644d242a00 100644 --- a/test/CodeGen/Mips/msa/i5_ld_st.ll +++ b/test/CodeGen/Mips/msa/i5_ld_st.ll @@ -22,6 +22,59 @@ declare <16 x i8> @llvm.mips.ld.b(i8*, i32) nounwind ; CHECK: st.b ; CHECK: .size llvm_mips_ld_b_test ; + +define void @llvm_mips_ld_b_unaligned_test() nounwind { +entry: + %0 = bitcast <16 x i8>* @llvm_mips_ld_b_ARG to i8* + %1 = tail call <16 x i8> @llvm.mips.ld.b(i8* %0, i32 9) + store <16 x i8> %1, <16 x i8>* @llvm_mips_ld_b_RES + ret void +} + +; CHECK: llvm_mips_ld_b_unaligned_test: +; CHECK: ld.b [[R1:\$w[0-9]+]], 9( +; CHECK: st.b +; CHECK: .size llvm_mips_ld_b_unaligned_test +; + +define void @llvm_mips_ld_b_valid_range_tests() nounwind { +entry: + %0 = bitcast <16 x i8>* @llvm_mips_ld_b_ARG to i8* + %1 = tail call <16 x i8> @llvm.mips.ld.b(i8* %0, i32 -512) + store <16 x i8> %1, <16 x i8>* @llvm_mips_ld_b_RES + %2 = tail call <16 x i8> @llvm.mips.ld.b(i8* %0, i32 511) + store <16 x i8> %2, <16 x i8>* @llvm_mips_ld_b_RES + ret void +} + +; CHECK: llvm_mips_ld_b_valid_range_tests: +; CHECK: ld.b [[R1:\$w[0-9]+]], -512( +; CHECK: st.b +; CHECK: ld.b [[R1:\$w[0-9]+]], 511( +; CHECK: st.b +; CHECK: .size llvm_mips_ld_b_valid_range_tests +; + +define void @llvm_mips_ld_b_invalid_range_tests() nounwind { +entry: + %0 = bitcast <16 x i8>* @llvm_mips_ld_b_ARG to i8* + %1 = tail call <16 x i8> @llvm.mips.ld.b(i8* %0, i32 -513) + store <16 x i8> %1, <16 x i8>* @llvm_mips_ld_b_RES + %2 = tail call <16 x i8> @llvm.mips.ld.b(i8* %0, i32 512) + store <16 x i8> %2, <16 x i8>* @llvm_mips_ld_b_RES + ret void +} + +; CHECK: llvm_mips_ld_b_invalid_range_tests: +; CHECK: addiu $3, $2, -513 +; CHECK: ld.b [[R1:\$w[0-9]+]], 0( +; CHECK: st.b +; CHECK: addiu $2, $2, 512 +; CHECK: ld.b [[R1:\$w[0-9]+]], 0( +; CHECK: st.b +; CHECK: .size llvm_mips_ld_b_invalid_range_tests +; + @llvm_mips_ld_h_ARG = global <8 x i16> , align 16 @llvm_mips_ld_h_RES = global <8 x i16> , align 16 @@ -40,6 +93,60 @@ declare <8 x i16> @llvm.mips.ld.h(i8*, i32) nounwind ; CHECK: st.h ; CHECK: .size llvm_mips_ld_h_test ; + +define void @llvm_mips_ld_h_unaligned_test() nounwind { +entry: + %0 = bitcast <8 x i16>* @llvm_mips_ld_h_ARG to i8* + %1 = tail call <8 x i16> @llvm.mips.ld.h(i8* %0, i32 9) + store <8 x i16> %1, <8 x i16>* @llvm_mips_ld_h_RES + ret void +} + +; CHECK: llvm_mips_ld_h_unaligned_test: +; CHECK: addiu $2, $2, 9 +; CHECK: ld.h [[R1:\$w[0-9]+]], 0($2) +; CHECK: st.h +; CHECK: .size llvm_mips_ld_h_unaligned_test +; + +define void @llvm_mips_ld_h_valid_range_tests() nounwind { +entry: + %0 = bitcast <8 x i16>* @llvm_mips_ld_h_ARG to i8* + %1 = tail call <8 x i16> @llvm.mips.ld.h(i8* %0, i32 -1024) + store <8 x i16> %1, <8 x i16>* @llvm_mips_ld_h_RES + %2 = tail call <8 x i16> @llvm.mips.ld.h(i8* %0, i32 1022) + store <8 x i16> %2, <8 x i16>* @llvm_mips_ld_h_RES + ret void +} + +; CHECK: llvm_mips_ld_h_valid_range_tests: +; CHECK: ld.h [[R1:\$w[0-9]+]], -1024( +; CHECK: st.h +; CHECK: ld.h [[R1:\$w[0-9]+]], 1022( +; CHECK: st.h +; CHECK: .size llvm_mips_ld_h_valid_range_tests +; + +define void @llvm_mips_ld_h_invalid_range_tests() nounwind { +entry: + %0 = bitcast <8 x i16>* @llvm_mips_ld_h_ARG to i8* + %1 = tail call <8 x i16> @llvm.mips.ld.h(i8* %0, i32 -1026) + store <8 x i16> %1, <8 x i16>* @llvm_mips_ld_h_RES + %2 = tail call <8 x i16> @llvm.mips.ld.h(i8* %0, i32 1024) + store <8 x i16> %2, <8 x i16>* @llvm_mips_ld_h_RES + ret void +} + +; CHECK: llvm_mips_ld_h_invalid_range_tests: +; CHECK: addiu $3, $2, -1026 +; CHECK: ld.h [[R1:\$w[0-9]+]], 0( +; CHECK: st.h +; CHECK: addiu $2, $2, 1024 +; CHECK: ld.h [[R1:\$w[0-9]+]], 0( +; CHECK: st.h +; CHECK: .size llvm_mips_ld_h_invalid_range_tests +; + @llvm_mips_ld_w_ARG = global <4 x i32> , align 16 @llvm_mips_ld_w_RES = global <4 x i32> , align 16 @@ -61,6 +168,59 @@ declare <4 x i32> @llvm.mips.ld.w(i8*, i32) nounwind @llvm_mips_ld_d_ARG = global <2 x i64> , align 16 @llvm_mips_ld_d_RES = global <2 x i64> , align 16 +define void @llvm_mips_ld_w_unaligned_test() nounwind { +entry: + %0 = bitcast <4 x i32>* @llvm_mips_ld_w_ARG to i8* + %1 = tail call <4 x i32> @llvm.mips.ld.w(i8* %0, i32 9) + store <4 x i32> %1, <4 x i32>* @llvm_mips_ld_w_RES + ret void +} + +; CHECK: llvm_mips_ld_w_unaligned_test: +; CHECK: addiu $2, $2, 9 +; CHECK: ld.w [[R1:\$w[0-9]+]], 0($2) +; CHECK: st.w +; CHECK: .size llvm_mips_ld_w_unaligned_test +; + +define void @llvm_mips_ld_w_valid_range_tests() nounwind { +entry: + %0 = bitcast <4 x i32>* @llvm_mips_ld_w_ARG to i8* + %1 = tail call <4 x i32> @llvm.mips.ld.w(i8* %0, i32 -2048) + store <4 x i32> %1, <4 x i32>* @llvm_mips_ld_w_RES + %2 = tail call <4 x i32> @llvm.mips.ld.w(i8* %0, i32 2044) + store <4 x i32> %2, <4 x i32>* @llvm_mips_ld_w_RES + ret void +} + +; CHECK: llvm_mips_ld_w_valid_range_tests: +; CHECK: ld.w [[R1:\$w[0-9]+]], -2048( +; CHECK: st.w +; CHECK: ld.w [[R1:\$w[0-9]+]], 2044( +; CHECK: st.w +; CHECK: .size llvm_mips_ld_w_valid_range_tests +; + +define void @llvm_mips_ld_w_invalid_range_tests() nounwind { +entry: + %0 = bitcast <4 x i32>* @llvm_mips_ld_w_ARG to i8* + %1 = tail call <4 x i32> @llvm.mips.ld.w(i8* %0, i32 -2052) + store <4 x i32> %1, <4 x i32>* @llvm_mips_ld_w_RES + %2 = tail call <4 x i32> @llvm.mips.ld.w(i8* %0, i32 2048) + store <4 x i32> %2, <4 x i32>* @llvm_mips_ld_w_RES + ret void +} + +; CHECK: llvm_mips_ld_w_invalid_range_tests: +; CHECK: addiu $3, $2, -2052 +; CHECK: ld.w [[R1:\$w[0-9]+]], 0( +; CHECK: st.w +; CHECK: addiu $2, $2, 2048 +; CHECK: ld.w [[R1:\$w[0-9]+]], 0( +; CHECK: st.w +; CHECK: .size llvm_mips_ld_w_invalid_range_tests +; + define void @llvm_mips_ld_d_test() nounwind { entry: %0 = bitcast <2 x i64>* @llvm_mips_ld_d_ARG to i8* @@ -76,6 +236,62 @@ declare <2 x i64> @llvm.mips.ld.d(i8*, i32) nounwind ; CHECK: st.d ; CHECK: .size llvm_mips_ld_d_test ; + +define void @llvm_mips_ld_d_unaligned_test() nounwind { +entry: + %0 = bitcast <2 x i64>* @llvm_mips_ld_d_ARG to i8* + %1 = tail call <2 x i64> @llvm.mips.ld.d(i8* %0, i32 9) + store <2 x i64> %1, <2 x i64>* @llvm_mips_ld_d_RES + ret void +} + +; CHECK: llvm_mips_ld_d_unaligned_test: +; CHECK: addiu $2, $2, 9 +; CHECK: ld.d [[R1:\$w[0-9]+]], 0($2) +; CHECK: st.d +; CHECK: .size llvm_mips_ld_d_unaligned_test +; + +define void @llvm_mips_ld_d_valid_range_tests() nounwind { +entry: + %0 = bitcast <2 x i64>* @llvm_mips_ld_d_ARG to i8* + %1 = tail call <2 x i64> @llvm.mips.ld.d(i8* %0, i32 -4096) + store <2 x i64> %1, <2 x i64>* @llvm_mips_ld_d_RES + %2 = tail call <2 x i64> @llvm.mips.ld.d(i8* %0, i32 4088) + store <2 x i64> %2, <2 x i64>* @llvm_mips_ld_d_RES + ret void +} + +; CHECK: llvm_mips_ld_d_valid_range_tests: +; CHECK: ld.d [[R1:\$w[0-9]+]], -4096( +; CHECK: st.d +; CHECK: ld.d [[R1:\$w[0-9]+]], 4088( +; CHECK: st.d +; CHECK: .size llvm_mips_ld_d_valid_range_tests +; + +define void @llvm_mips_ld_d_invalid_range_tests() nounwind { +entry: + %0 = bitcast <2 x i64>* @llvm_mips_ld_d_ARG to i8* + %1 = tail call <2 x i64> @llvm.mips.ld.d(i8* %0, i32 -4104) + store <2 x i64> %1, <2 x i64>* @llvm_mips_ld_d_RES + %2 = tail call <2 x i64> @llvm.mips.ld.d(i8* %0, i32 4096) + store <2 x i64> %2, <2 x i64>* @llvm_mips_ld_d_RES + ret void +} + +; CHECK: llvm_mips_ld_d_invalid_range_tests: +; CHECK: addiu $3, $2, -4104 +; CHECK: ld.d [[R1:\$w[0-9]+]], 0( +; CHECK: st.d +; CHECK: addiu $2, $2, 4096 +; CHECK: ld.d [[R1:\$w[0-9]+]], 0( +; CHECK: st.d +; CHECK: .size llvm_mips_ld_d_invalid_range_tests +; + + + @llvm_mips_st_b_ARG = global <16 x i8> , align 16 @llvm_mips_st_b_RES = global <16 x i8> , align 16 @@ -94,6 +310,55 @@ declare void @llvm.mips.st.b(<16 x i8>, i8*, i32) nounwind ; CHECK: st.b [[R1:\$w[0-9]+]], 16( ; CHECK: .size llvm_mips_st_b_test ; + +define void @llvm_mips_st_b_unaligned_test() nounwind { +entry: + %0 = load <16 x i8>, <16 x i8>* @llvm_mips_st_b_ARG + %1 = bitcast <16 x i8>* @llvm_mips_st_b_RES to i8* + tail call void @llvm.mips.st.b(<16 x i8> %0, i8* %1, i32 9) + ret void +} + +; CHECK: llvm_mips_st_b_unaligned_test: +; CHECK: ld.b +; CHECK: st.b [[R1:\$w[0-9]+]], 9( +; CHECK: .size llvm_mips_st_b_unaligned_test +; + +define void @llvm_mips_st_b_valid_range_tests() nounwind { +entry: + %0 = load <16 x i8>, <16 x i8>* @llvm_mips_st_b_ARG + %1 = bitcast <16 x i8>* @llvm_mips_st_b_RES to i8* + tail call void @llvm.mips.st.b(<16 x i8> %0, i8* %1, i32 -512) + tail call void @llvm.mips.st.b(<16 x i8> %0, i8* %1, i32 511) + ret void +} + +; CHECK: llvm_mips_st_b_valid_range_tests: +; CHECK: ld.b +; CHECK: st.b [[R1:\$w[0-9]+]], -512( +; CHECK: st.b [[R1:\$w[0-9]+]], 511( +; CHECK: .size llvm_mips_st_b_valid_range_tests +; + +define void @llvm_mips_st_b_invalid_range_tests() nounwind { +entry: + %0 = load <16 x i8>, <16 x i8>* @llvm_mips_st_b_ARG + %1 = bitcast <16 x i8>* @llvm_mips_st_b_RES to i8* + tail call void @llvm.mips.st.b(<16 x i8> %0, i8* %1, i32 -513) + tail call void @llvm.mips.st.b(<16 x i8> %0, i8* %1, i32 512) + ret void +} + +; CHECK: llvm_mips_st_b_invalid_range_tests: +; CHECK: addiu $2, $1, -513 +; CHECK: ld.b +; CHECK: st.b [[R1:\$w[0-9]+]], 0( +; CHECK: addiu $1, $1, 512 +; CHECK: st.b [[R1:\$w[0-9]+]], 0( +; CHECK: .size llvm_mips_st_b_invalid_range_tests +; + @llvm_mips_st_h_ARG = global <8 x i16> , align 16 @llvm_mips_st_h_RES = global <8 x i16> , align 16 @@ -112,6 +377,56 @@ declare void @llvm.mips.st.h(<8 x i16>, i8*, i32) nounwind ; CHECK: st.h [[R1:\$w[0-9]+]], 16( ; CHECK: .size llvm_mips_st_h_test ; + +define void @llvm_mips_st_h_unaligned_test() nounwind { +entry: + %0 = load <8 x i16>, <8 x i16>* @llvm_mips_st_h_ARG + %1 = bitcast <8 x i16>* @llvm_mips_st_h_RES to i8* + tail call void @llvm.mips.st.h(<8 x i16> %0, i8* %1, i32 9) + ret void +} + +; CHECK: llvm_mips_st_h_unaligned_test: +; CHECK: addiu $1, $1, 9 +; CHECK: ld.h +; CHECK: st.h [[R1:\$w[0-9]+]], 0($1) +; CHECK: .size llvm_mips_st_h_unaligned_test +; + +define void @llvm_mips_st_h_valid_range_tests() nounwind { +entry: + %0 = load <8 x i16>, <8 x i16>* @llvm_mips_st_h_ARG + %1 = bitcast <8 x i16>* @llvm_mips_st_h_RES to i8* + tail call void @llvm.mips.st.h(<8 x i16> %0, i8* %1, i32 -1024) + tail call void @llvm.mips.st.h(<8 x i16> %0, i8* %1, i32 1022) + ret void +} + +; CHECK: llvm_mips_st_h_valid_range_tests: +; CHECK: ld.h +; CHECK: st.h [[R1:\$w[0-9]+]], -1024( +; CHECK: st.h [[R1:\$w[0-9]+]], 1022( +; CHECK: .size llvm_mips_st_h_valid_range_tests +; + +define void @llvm_mips_st_h_invalid_range_tests() nounwind { +entry: + %0 = load <8 x i16>, <8 x i16>* @llvm_mips_st_h_ARG + %1 = bitcast <8 x i16>* @llvm_mips_st_h_RES to i8* + tail call void @llvm.mips.st.h(<8 x i16> %0, i8* %1, i32 -1026) + tail call void @llvm.mips.st.h(<8 x i16> %0, i8* %1, i32 1024) + ret void +} + +; CHECK: llvm_mips_st_h_invalid_range_tests: +; CHECK: addiu $2, $1, -1026 +; CHECK: ld.h +; CHECK: st.h [[R1:\$w[0-9]+]], 0( +; CHECK: addiu $1, $1, 1024 +; CHECK: st.h [[R1:\$w[0-9]+]], 0( +; CHECK: .size llvm_mips_st_h_invalid_range_tests +; + @llvm_mips_st_w_ARG = global <4 x i32> , align 16 @llvm_mips_st_w_RES = global <4 x i32> , align 16 @@ -130,6 +445,56 @@ declare void @llvm.mips.st.w(<4 x i32>, i8*, i32) nounwind ; CHECK: st.w [[R1:\$w[0-9]+]], 16( ; CHECK: .size llvm_mips_st_w_test ; + +define void @llvm_mips_st_w_unaligned_test() nounwind { +entry: + %0 = load <4 x i32>, <4 x i32>* @llvm_mips_st_w_ARG + %1 = bitcast <4 x i32>* @llvm_mips_st_w_RES to i8* + tail call void @llvm.mips.st.w(<4 x i32> %0, i8* %1, i32 9) + ret void +} + +; CHECK: llvm_mips_st_w_unaligned_test: +; CHECK: addiu $1, $1, 9 +; CHECK: ld.w +; CHECK: st.w [[R1:\$w[0-9]+]], 0($1) +; CHECK: .size llvm_mips_st_w_unaligned_test +; + +define void @llvm_mips_st_w_valid_range_tests() nounwind { +entry: + %0 = load <4 x i32>, <4 x i32>* @llvm_mips_st_w_ARG + %1 = bitcast <4 x i32>* @llvm_mips_st_w_RES to i8* + tail call void @llvm.mips.st.w(<4 x i32> %0, i8* %1, i32 -2048) + tail call void @llvm.mips.st.w(<4 x i32> %0, i8* %1, i32 2044) + ret void +} + +; CHECK: llvm_mips_st_w_valid_range_tests: +; CHECK: ld.w +; CHECK: st.w [[R1:\$w[0-9]+]], -2048( +; CHECK: st.w [[R1:\$w[0-9]+]], 2044( +; CHECK: .size llvm_mips_st_w_valid_range_tests +; + +define void @llvm_mips_st_w_invalid_range_tests() nounwind { +entry: + %0 = load <4 x i32>, <4 x i32>* @llvm_mips_st_w_ARG + %1 = bitcast <4 x i32>* @llvm_mips_st_w_RES to i8* + tail call void @llvm.mips.st.w(<4 x i32> %0, i8* %1, i32 -2052) + tail call void @llvm.mips.st.w(<4 x i32> %0, i8* %1, i32 2048) + ret void +} + +; CHECK: llvm_mips_st_w_invalid_range_tests: +; CHECK: addiu $2, $1, -2052 +; CHECK: ld.w +; CHECK: st.w [[R1:\$w[0-9]+]], 0( +; CHECK: addiu $1, $1, 2048 +; CHECK: st.w [[R1:\$w[0-9]+]], 0( +; CHECK: .size llvm_mips_st_w_invalid_range_tests +; + @llvm_mips_st_d_ARG = global <2 x i64> , align 16 @llvm_mips_st_d_RES = global <2 x i64> , align 16 @@ -148,3 +513,52 @@ declare void @llvm.mips.st.d(<2 x i64>, i8*, i32) nounwind ; CHECK: st.d [[R1:\$w[0-9]+]], 16( ; CHECK: .size llvm_mips_st_d_test ; + +define void @llvm_mips_st_d_unaligned_test() nounwind { +entry: + %0 = load <2 x i64>, <2 x i64>* @llvm_mips_st_d_ARG + %1 = bitcast <2 x i64>* @llvm_mips_st_d_RES to i8* + tail call void @llvm.mips.st.d(<2 x i64> %0, i8* %1, i32 9) + ret void +} + +; CHECK: llvm_mips_st_d_unaligned_test: +; CHECK: addiu $1, $1, 9 +; CHECK: ld.d +; CHECK: st.d [[R1:\$w[0-9]+]], 0($1) +; CHECK: .size llvm_mips_st_d_unaligned_test +; + +define void @llvm_mips_st_d_valid_range_tests() nounwind { +entry: + %0 = load <2 x i64>, <2 x i64>* @llvm_mips_st_d_ARG + %1 = bitcast <2 x i64>* @llvm_mips_st_d_RES to i8* + tail call void @llvm.mips.st.d(<2 x i64> %0, i8* %1, i32 -4096) + tail call void @llvm.mips.st.d(<2 x i64> %0, i8* %1, i32 4088) + ret void +} + +; CHECK: llvm_mips_st_d_valid_range_tests: +; CHECK: ld.d +; CHECK: st.d [[R1:\$w[0-9]+]], -4096( +; CHECK: st.d [[R1:\$w[0-9]+]], 4088( +; CHECK: .size llvm_mips_st_d_valid_range_tests +; + +define void @llvm_mips_st_d_invalid_range_tests() nounwind { +entry: + %0 = load <2 x i64>, <2 x i64>* @llvm_mips_st_d_ARG + %1 = bitcast <2 x i64>* @llvm_mips_st_d_RES to i8* + tail call void @llvm.mips.st.d(<2 x i64> %0, i8* %1, i32 -4104) + tail call void @llvm.mips.st.d(<2 x i64> %0, i8* %1, i32 4096) + ret void +} + +; CHECK: llvm_mips_st_d_invalid_range_tests: +; CHECK: addiu $2, $1, -4104 +; CHECK: ld.d +; CHECK: st.d [[R1:\$w[0-9]+]], 0( +; CHECK: addiu $1, $1, 4096 +; CHECK: st.d [[R1:\$w[0-9]+]], 0( +; CHECK: .size llvm_mips_st_d_invalid_range_tests +;