// Copyright © 2019-2023
// 
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
// http://www.apache.org/licenses/LICENSE-2.0
// 
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#ifndef __VX_INTRINSICS_H__
#define __VX_INTRINSICS_H__

#include <VX_config.h>
#include <VX_types.h>

#if defined(__clang__)
#define __UNIFORM__   __attribute__((annotate("vortex.uniform")))
#else
#define __UNIFORM__
#endif

#ifdef __cplusplus
extern "C" {
#endif

#ifdef __ASSEMBLY__
#define __ASM_STR(x)	x
#else
#define __ASM_STR(x)	#x
#endif

#define RISCV_CUSTOM0   0x0B
#define RISCV_CUSTOM1   0x2B
#define RISCV_CUSTOM2   0x5B
#define RISCV_CUSTOM3   0x7B

#define csr_read(csr) ({                        \
	unsigned __r;	               		        \
	__asm__ __volatile__ ("csrr %0, %1" : "=r" (__r) : "i" (csr)); \
	__r;							            \
})

#define csr_write(csr, val)	({                  \
	unsigned __v = (unsigned)(val);             \
	if (__builtin_constant_p(val) && __v < 32)  \
        __asm__ __volatile__ ("csrw %0, %1"	:: "i" (csr), "i" (__v));  \
    else                                        \
        __asm__ __volatile__ ("csrw %0, %1"	:: "i" (csr), "r" (__v));  \
})

#define csr_swap(csr, val) ({                   \
    unsigned __r;                               \
	unsigned __v = (unsigned)(val);	            \
	if (__builtin_constant_p(val) && __v < 32)  \
        __asm__ __volatile__ ("csrrw %0, %1, %2" : "=r" (__r) : "i" (csr), "i" (__v)); \
    else                                        \
        __asm__ __volatile__ ("csrrw %0, %1, %2" : "=r" (__r) : "i" (csr), "r" (__v)); \
	__r;						                \
})

#define csr_read_set(csr, val) ({               \
	unsigned __r;                               \
	unsigned __v = (unsigned)(val);	            \
    if (__builtin_constant_p(val) && __v < 32)  \
	    __asm__ __volatile__ ("csrrs %0, %1, %2" : "=r" (__r) : "i" (csr), "i" (__v)); \
    else                                        \
        __asm__ __volatile__ ("csrrs %0, %1, %2" : "=r" (__r) : "i" (csr), "r" (__v)); \
	__r;							            \
})

#define csr_set(csr, val) ({                    \
	unsigned __v = (unsigned)(val);	            \
    if (__builtin_constant_p(val) && __v < 32)  \
	    __asm__ __volatile__ ("csrs %0, %1"	:: "i" (csr), "i" (__v));  \
    else                                        \
        __asm__ __volatile__ ("csrs %0, %1"	:: "i" (csr), "r" (__v));  \
})

#define csr_read_clear(csr, val) ({             \
	unsigned __r;                               \
	unsigned __v = (unsigned)(val);	            \
    if (__builtin_constant_p(val) && __v < 32)  \
	    __asm__ __volatile__ ("csrrc %0, %1, %2" : "=r" (__r) : "i" (csr), "i" (__v)); \
    else                                        \
        __asm__ __volatile__ ("csrrc %0, %1, %2" : "=r" (__r) : "i" (csr), "r" (__v)); \
	__r;							            \
})

#define csr_clear(csr, val)	({                  \
	unsigned __v = (unsigned)(val);             \
	if (__builtin_constant_p(val) && __v < 32)  \
        __asm__ __volatile__ ("csrc %0, %1"	:: "i" (csr), "i" (__v)); \
    else                                        \
        __asm__ __volatile__ ("csrc %0, %1"	:: "i" (csr), "r" (__v)); \
})

// Conditional move
inline unsigned vx_cmov(unsigned c, unsigned t, unsigned f) {
    unsigned ret;
    asm volatile (".insn r4 %1, 1, 0, %0, %2, %3, %4" : "=r"(ret) : "i"(RISCV_CUSTOM1), "r"(c), "r"(t), "r"(f));
    return ret;
}

// Set thread mask
inline void vx_tmc(unsigned thread_mask) {
    asm volatile (".insn r %0, 0, 0, x0, %1, x0" :: "i"(RISCV_CUSTOM0), "r"(thread_mask));
}

// disable all threads in the current warp
inline void vx_tmc_zero() {
    asm volatile (".insn r %0, 0, 0, x0, x0, x0" :: "i"(RISCV_CUSTOM0));
}

// switch execution to single thread zero
inline void vx_tmc_one() {
    asm volatile (
        "li a0, 1\n\t"  // Load immediate value 1 into a0 (x10) register
        ".insn r %0, 0, 0, x0, a0, x0" :: "i"(RISCV_CUSTOM0) 
        : "a0"          // Indicate that a0 (x10) is clobbered
    );
}

// Set thread predicate
inline void vx_pred(unsigned condition, unsigned thread_mask) {
    asm volatile (".insn r %0, 5, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(condition), "r"(thread_mask));
}

typedef void (*vx_wspawn_pfn)();

// Spawn warps
inline void vx_wspawn(unsigned num_warps, vx_wspawn_pfn func_ptr) {
    asm volatile (".insn r %0, 1, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(num_warps), "r"(func_ptr));
}

// Spawn an explicit warp mask. The current warp bit is ignored by hardware.
inline void vx_wspawn_mask(unsigned warp_mask, vx_wspawn_pfn func_ptr) {
    asm volatile (".insn r %0, 6, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(warp_mask), "r"(func_ptr));
}

inline void vx_spawn_scalar(unsigned warp_mask, vx_wspawn_pfn func_ptr) {
    vx_wspawn_mask(warp_mask & ((1u << NUM_SCALAR_WARPS) - 1u), func_ptr);
}

inline void vx_spawn_tensor(unsigned warp_mask, vx_wspawn_pfn func_ptr) {
    vx_wspawn_mask(warp_mask & (((1u << NUM_TENSOR_WARPS) - 1u) << NUM_SCALAR_WARPS), func_ptr);
}

// Split on a predicate
inline unsigned vx_split(unsigned predicate) {
    unsigned ret;
    asm volatile (".insn r %1, 2, 0, %0, %2, x0" : "=r"(ret) : "i"(RISCV_CUSTOM0), "r"(predicate));
    return ret;
}

// Join
inline void vx_join(unsigned stack_ptr) {
    asm volatile (".insn r %0, 3, 0, x0, %1, x0" :: "i"(RISCV_CUSTOM0), "r"(stack_ptr));
}

// Warp Barrier
__attribute__((convergent))
inline void vx_barrier(unsigned barried_id, unsigned num_warps) {
    unsigned scalar_warps = (num_warps > NUM_SCALAR_WARPS) ? NUM_SCALAR_WARPS : num_warps;
    asm volatile (".insn r %0, 4, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(barried_id), "r"(scalar_warps));
}

#define VX_BARRIER_DOMAIN_SHIFT 28
#define VX_BARRIER_DOMAIN_ALL    0u
#define VX_BARRIER_DOMAIN_SCALAR 1u
#define VX_BARRIER_DOMAIN_TENSOR 2u

__attribute__((convergent))
inline void vx_barrier_domain(unsigned barrier_id, unsigned num_warps, unsigned domain) {
    unsigned encoded_id = barrier_id | (domain << VX_BARRIER_DOMAIN_SHIFT);
    asm volatile (".insn r %0, 4, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(encoded_id), "r"(num_warps));
}

__attribute__((convergent))
inline void vx_barrier_scalar(unsigned barrier_id, unsigned num_warps) {
    vx_barrier_domain(barrier_id, num_warps, VX_BARRIER_DOMAIN_SCALAR);
}

__attribute__((convergent))
inline void vx_barrier_tensor(unsigned barrier_id, unsigned num_warps) {
    vx_barrier_domain(barrier_id, num_warps, VX_BARRIER_DOMAIN_TENSOR);
}

__attribute__((convergent))
inline void vx_barrier_mask(unsigned barrier_id, unsigned warp_mask) {
    asm volatile (".insn r %0, 7, 0, x0, %1, %2" :: "i"(RISCV_CUSTOM0), "r"(barrier_id), "r"(warp_mask));
}

// Return current thread identifier
inline int vx_thread_id() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_THREAD_ID));
    return ret;
}

// Return current warp identifier
inline int vx_warp_id() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_WARP_ID));
    return ret;
}

// Return current core identifier
inline int vx_core_id() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_CORE_ID));
    return ret;
}

// Return current thread mask
inline int vx_thread_mask() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_THREAD_MASK));
    return ret;
}

// Return number of active warps
inline int vx_active_warps() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_WARP_MASK));
    return ret;
}

// Return the number of threads per warp
inline int vx_num_threads() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_NUM_THREADS));
    return ret;
}

// Return the number of warps per core
inline int vx_num_warps() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_NUM_WARPS));
    return ret;   
}

inline int vx_num_scalar_warps() {
    return NUM_SCALAR_WARPS;
}

inline int vx_num_tensor_warps() {
    return NUM_TENSOR_WARPS;
}

inline unsigned vx_scalar_warp_mask() {
    return (1u << NUM_SCALAR_WARPS) - 1u;
}

inline unsigned vx_tensor_warp_mask() {
    return ((1u << NUM_TENSOR_WARPS) - 1u) << NUM_SCALAR_WARPS;
}

// Return the number of cores per cluster
inline int vx_num_cores() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_NUM_CORES));
    return ret;
}

// Return the hart identifier (thread id accross the processor)
inline int vx_hart_id() {
    int ret;
    asm volatile ("csrr %0, %1" : "=r"(ret) : "i"(VX_CSR_MHARTID));
    return ret;
}

inline void vx_fence() {
    asm volatile ("fence iorw, iorw");
}

#ifdef __cplusplus
}
#endif

#endif // __VX_INTRINSICS_H__