1//===-------- amdgcn_smid.hip - AMDGCN smid implementation -------- HIP -*-===// 2// 3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4// See https://llvm.org/LICENSE.txt for license information. 5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6// 7//===----------------------------------------------------------------------===// 8 9#include "target_impl.h" 10 11// Partially derived fom hcc_detail/device_functions.h 12 13// HW_ID Register bit structure 14// WAVE_ID 3:0 Wave buffer slot number. 0-9. 15// SIMD_ID 5:4 SIMD which the wave is assigned to within the CU. 16// PIPE_ID 7:6 Pipeline from which the wave was dispatched. 17// CU_ID 11:8 Compute Unit the wave is assigned to. 18// SH_ID 12 Shader Array (within an SE) the wave is assigned to. 19// SE_ID 14:13 Shader Engine the wave is assigned to. 20// TG_ID 19:16 Thread-group ID 21// VM_ID 23:20 Virtual Memory ID 22// QUEUE_ID 26:24 Queue from which this wave was dispatched. 23// STATE_ID 29:27 State ID (graphics only, not compute). 24// ME_ID 31:30 Micro-engine ID. 25 26enum { 27 HW_ID = 4, // specify that the hardware register to read is HW_ID 28 29 HW_ID_CU_ID_SIZE = 4, // size of CU_ID field in bits 30 HW_ID_CU_ID_OFFSET = 8, // offset of CU_ID from start of register 31 32 HW_ID_SE_ID_SIZE = 2, // sizeof SE_ID field in bits 33 HW_ID_SE_ID_OFFSET = 13, // offset of SE_ID from start of register 34}; 35 36// The s_getreg_b32 instruction, exposed as an intrinsic, takes a 16 bit 37// immediate and returns a 32 bit value. 38// The encoding of the immediate parameter is: 39// ID 5:0 Which register to read from 40// OFFSET 10:6 Range: 0..31 41// WIDTH 15:11 Range: 1..32 42 43// The asm equivalent is s_getreg_b32 %0, hwreg(HW_REG_HW_ID, Offset, Width) 44// where hwreg forms a 16 bit immediate encoded by the assembler thus: 45// uint64_t encodeHwreg(uint64_t Id, uint64_t Offset, uint64_t Width) { 46// return (Id << 0_) | (Offset << 6) | ((Width - 1) << 11); 47// } 48#define ENCODE_HWREG(WIDTH, OFF, REG) (REG | (OFF << 6) | ((WIDTH - 1) << 11)) 49 50// Note: The results can be changed by a context switch 51// Return value in [0 2^SE_ID_SIZE * 2^CU_ID_SIZE), which is an upper 52// bound on how many compute units are available. Some values in this 53// range may never be returned if there are fewer than 2^CU_ID_SIZE CUs. 54 55DEVICE uint32_t __kmpc_impl_smid() { 56 uint32_t cu_id = __builtin_amdgcn_s_getreg( 57 ENCODE_HWREG(HW_ID_CU_ID_SIZE, HW_ID_CU_ID_OFFSET, HW_ID)); 58 uint32_t se_id = __builtin_amdgcn_s_getreg( 59 ENCODE_HWREG(HW_ID_SE_ID_SIZE, HW_ID_SE_ID_OFFSET, HW_ID)); 60 return (se_id << HW_ID_CU_ID_SIZE) + cu_id; 61} 62