summaryrefslogtreecommitdiff
path: root/lib/Target/AMDGPU/CIInstructions.td
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Target/AMDGPU/CIInstructions.td')
-rw-r--r--lib/Target/AMDGPU/CIInstructions.td149
1 files changed, 149 insertions, 0 deletions
diff --git a/lib/Target/AMDGPU/CIInstructions.td b/lib/Target/AMDGPU/CIInstructions.td
new file mode 100644
index 00000000000..2f5fdbe9207
--- /dev/null
+++ b/lib/Target/AMDGPU/CIInstructions.td
@@ -0,0 +1,149 @@
+//===-- CIInstructions.td - CI Instruction Defintions ---------------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+// Instruction definitions for CI and newer.
+//===----------------------------------------------------------------------===//
+
+
+def isCIVI : Predicate <
+ "Subtarget->getGeneration() == AMDGPUSubtarget::SEA_ISLANDS || "
+ "Subtarget->getGeneration() == AMDGPUSubtarget::VOLCANIC_ISLANDS"
+>, AssemblerPredicate<"FeatureCIInsts">;
+
+def HasFlatAddressSpace : Predicate<"Subtarget->hasFlatAddressSpace()">;
+
+//===----------------------------------------------------------------------===//
+// VOP1 Instructions
+//===----------------------------------------------------------------------===//
+
+let SubtargetPredicate = isCIVI in {
+
+defm V_TRUNC_F64 : VOP1Inst <vop1<0x17>, "v_trunc_f64",
+ VOP_F64_F64, ftrunc
+>;
+defm V_CEIL_F64 : VOP1Inst <vop1<0x18>, "v_ceil_f64",
+ VOP_F64_F64, fceil
+>;
+defm V_FLOOR_F64 : VOP1Inst <vop1<0x1A>, "v_floor_f64",
+ VOP_F64_F64, ffloor
+>;
+defm V_RNDNE_F64 : VOP1Inst <vop1<0x19>, "v_rndne_f64",
+ VOP_F64_F64, frint
+>;
+defm V_LOG_LEGACY_F32 : VOP1Inst <vop1<0x45, 0x4c>, "v_log_legacy_f32",
+ VOP_F32_F32
+>;
+defm V_EXP_LEGACY_F32 : VOP1Inst <vop1<0x46, 0x4b>, "v_exp_legacy_f32",
+ VOP_F32_F32
+>;
+
+//===----------------------------------------------------------------------===//
+// Flat Instructions
+//===----------------------------------------------------------------------===//
+
+def FLAT_LOAD_UBYTE : FLAT_Load_Helper <0x8, "flat_load_ubyte", VGPR_32>;
+def FLAT_LOAD_SBYTE : FLAT_Load_Helper <0x9, "flat_load_sbyte", VGPR_32>;
+def FLAT_LOAD_USHORT : FLAT_Load_Helper <0xa, "flat_load_ushort", VGPR_32>;
+def FLAT_LOAD_SSHORT : FLAT_Load_Helper <0xb, "flat_load_sshort", VGPR_32>;
+def FLAT_LOAD_DWORD : FLAT_Load_Helper <0xc, "flat_load_dword", VGPR_32>;
+def FLAT_LOAD_DWORDX2 : FLAT_Load_Helper <0xd, "flat_load_dwordx2", VReg_64>;
+def FLAT_LOAD_DWORDX4 : FLAT_Load_Helper <0xe, "flat_load_dwordx4", VReg_128>;
+def FLAT_LOAD_DWORDX3 : FLAT_Load_Helper <0xf, "flat_load_dwordx3", VReg_96>;
+def FLAT_STORE_BYTE : FLAT_Store_Helper <0x18, "flat_store_byte", VGPR_32>;
+def FLAT_STORE_SHORT : FLAT_Store_Helper <0x1a, "flat_store_short", VGPR_32>;
+def FLAT_STORE_DWORD : FLAT_Store_Helper <0x1c, "flat_store_dword", VGPR_32>;
+def FLAT_STORE_DWORDX2 : FLAT_Store_Helper <
+ 0x1d, "flat_store_dwordx2", VReg_64
+>;
+def FLAT_STORE_DWORDX4 : FLAT_Store_Helper <
+ 0x1e, "flat_store_dwordx4", VReg_128
+>;
+def FLAT_STORE_DWORDX3 : FLAT_Store_Helper <
+ 0x1f, "flat_store_dwordx3", VReg_96
+>;
+defm FLAT_ATOMIC_SWAP : FLAT_ATOMIC <0x30, "flat_atomic_swap", VGPR_32>;
+defm FLAT_ATOMIC_CMPSWAP : FLAT_ATOMIC <
+ 0x31, "flat_atomic_cmpswap", VGPR_32, VReg_64
+>;
+defm FLAT_ATOMIC_ADD : FLAT_ATOMIC <0x32, "flat_atomic_add", VGPR_32>;
+defm FLAT_ATOMIC_SUB : FLAT_ATOMIC <0x33, "flat_atomic_sub", VGPR_32>;
+defm FLAT_ATOMIC_RSUB : FLAT_ATOMIC <0x34, "flat_atomic_rsub", VGPR_32>;
+defm FLAT_ATOMIC_SMIN : FLAT_ATOMIC <0x35, "flat_atomic_smin", VGPR_32>;
+defm FLAT_ATOMIC_UMIN : FLAT_ATOMIC <0x36, "flat_atomic_umin", VGPR_32>;
+defm FLAT_ATOMIC_SMAX : FLAT_ATOMIC <0x37, "flat_atomic_smax", VGPR_32>;
+defm FLAT_ATOMIC_UMAX : FLAT_ATOMIC <0x38, "flat_atomic_umax", VGPR_32>;
+defm FLAT_ATOMIC_AND : FLAT_ATOMIC <0x39, "flat_atomic_and", VGPR_32>;
+defm FLAT_ATOMIC_OR : FLAT_ATOMIC <0x3a, "flat_atomic_or", VGPR_32>;
+defm FLAT_ATOMIC_XOR : FLAT_ATOMIC <0x3b, "flat_atomic_xor", VGPR_32>;
+defm FLAT_ATOMIC_INC : FLAT_ATOMIC <0x3c, "flat_atomic_inc", VGPR_32>;
+defm FLAT_ATOMIC_DEC : FLAT_ATOMIC <0x3d, "flat_atomic_dec", VGPR_32>;
+defm FLAT_ATOMIC_FCMPSWAP : FLAT_ATOMIC <
+ 0x3e, "flat_atomic_fcmpswap", VGPR_32, VReg_64
+>;
+defm FLAT_ATOMIC_FMIN : FLAT_ATOMIC <0x3f, "flat_atomic_fmin", VGPR_32>;
+defm FLAT_ATOMIC_FMAX : FLAT_ATOMIC <0x40, "flat_atomic_fmax", VGPR_32>;
+defm FLAT_ATOMIC_SWAP_X2 : FLAT_ATOMIC <0x50, "flat_atomic_swap_x2", VReg_64>;
+defm FLAT_ATOMIC_CMPSWAP_X2 : FLAT_ATOMIC <
+ 0x51, "flat_atomic_cmpswap_x2", VReg_64, VReg_128
+>;
+defm FLAT_ATOMIC_ADD_X2 : FLAT_ATOMIC <0x52, "flat_atomic_add_x2", VReg_64>;
+defm FLAT_ATOMIC_SUB_X2 : FLAT_ATOMIC <0x53, "flat_atomic_sub_x2", VReg_64>;
+defm FLAT_ATOMIC_RSUB_X2 : FLAT_ATOMIC <0x54, "flat_atomic_rsub_x2", VReg_64>;
+defm FLAT_ATOMIC_SMIN_X2 : FLAT_ATOMIC <0x55, "flat_atomic_smin_x2", VReg_64>;
+defm FLAT_ATOMIC_UMIN_X2 : FLAT_ATOMIC <0x56, "flat_atomic_umin_x2", VReg_64>;
+defm FLAT_ATOMIC_SMAX_X2 : FLAT_ATOMIC <0x57, "flat_atomic_smax_x2", VReg_64>;
+defm FLAT_ATOMIC_UMAX_X2 : FLAT_ATOMIC <0x58, "flat_atomic_umax_x2", VReg_64>;
+defm FLAT_ATOMIC_AND_X2 : FLAT_ATOMIC <0x59, "flat_atomic_and_x2", VReg_64>;
+defm FLAT_ATOMIC_OR_X2 : FLAT_ATOMIC <0x5a, "flat_atomic_or_x2", VReg_64>;
+defm FLAT_ATOMIC_XOR_X2 : FLAT_ATOMIC <0x5b, "flat_atomic_xor_x2", VReg_64>;
+defm FLAT_ATOMIC_INC_X2 : FLAT_ATOMIC <0x5c, "flat_atomic_inc_x2", VReg_64>;
+defm FLAT_ATOMIC_DEC_X2 : FLAT_ATOMIC <0x5d, "flat_atomic_dec_x2", VReg_64>;
+defm FLAT_ATOMIC_FCMPSWAP_X2 : FLAT_ATOMIC <
+ 0x5e, "flat_atomic_fcmpswap_x2", VReg_64, VReg_128
+>;
+defm FLAT_ATOMIC_FMIN_X2 : FLAT_ATOMIC <0x5f, "flat_atomic_fmin_x2", VReg_64>;
+defm FLAT_ATOMIC_FMAX_X2 : FLAT_ATOMIC <0x60, "flat_atomic_fmax_x2", VReg_64>;
+
+} // End SubtargetPredicate = isCIVI
+
+//===----------------------------------------------------------------------===//
+// Flat Patterns
+//===----------------------------------------------------------------------===//
+
+let Predicates = [HasFlatAddressSpace] in {
+
+class FLATLoad_Pattern <FLAT Instr_ADDR64, ValueType vt,
+ PatFrag flat_ld> :
+ Pat <(vt (flat_ld i64:$ptr)),
+ (Instr_ADDR64 $ptr, 0, 0, 0)
+>;
+
+def : FLATLoad_Pattern <FLAT_LOAD_SBYTE, i32, sextloadi8_flat>;
+def : FLATLoad_Pattern <FLAT_LOAD_UBYTE, i32, az_extloadi8_flat>;
+def : FLATLoad_Pattern <FLAT_LOAD_SSHORT, i32, sextloadi16_flat>;
+def : FLATLoad_Pattern <FLAT_LOAD_USHORT, i32, az_extloadi16_flat>;
+def : FLATLoad_Pattern <FLAT_LOAD_DWORD, i32, flat_load>;
+def : FLATLoad_Pattern <FLAT_LOAD_DWORDX2, i64, flat_load>;
+def : FLATLoad_Pattern <FLAT_LOAD_DWORDX2, i64, az_extloadi32_flat>;
+def : FLATLoad_Pattern <FLAT_LOAD_DWORDX2, v2i32, flat_load>;
+def : FLATLoad_Pattern <FLAT_LOAD_DWORDX4, v4i32, flat_load>;
+
+class FLATStore_Pattern <FLAT Instr, ValueType vt, PatFrag st> :
+ Pat <(st vt:$value, i64:$ptr),
+ (Instr $value, $ptr, 0, 0, 0)
+ >;
+
+def : FLATStore_Pattern <FLAT_STORE_BYTE, i32, truncstorei8_flat>;
+def : FLATStore_Pattern <FLAT_STORE_SHORT, i32, truncstorei16_flat>;
+def : FLATStore_Pattern <FLAT_STORE_DWORD, i32, flat_store>;
+def : FLATStore_Pattern <FLAT_STORE_DWORDX2, i64, flat_store>;
+def : FLATStore_Pattern <FLAT_STORE_DWORDX2, v2i32, flat_store>;
+def : FLATStore_Pattern <FLAT_STORE_DWORDX4, v4i32, flat_store>;
+
+} // End HasFlatAddressSpace predicate
+