runtime static library

runtime/src/vx_intrinsics.S

@@ -0,0 +1,93 @@
+#include <VX_config.h>
+
+.section .text
+
+.type vx_wspawn, @function
+.global vx_wspawn
+vx_wspawn:
+	.word 0x00b5106b # wspawn a0(num_warps), a1(func_ptr)
+	ret
+
+.type vx_tmc, @function
+.global vx_tmc
+vx_tmc:
+	.word 0x0005006b    # tmc a0
+	ret
+
+.type vx_barrier, @function
+.global vx_barrier
+vx_barrier:
+	.word 0x00b5406b # barrier a0(barrier_id), a1(num_warps)
+	ret
+
+.type vx_split, @function
+.global vx_split
+vx_split:
+	.word 0x0005206b    # split a0
+	ret
+
+.type vx_join, @function
+.global vx_join
+vx_join:
+	.word 0x0000306b    #join
+	ret
+
+.type vx_warp_id, @function
+.global vx_warp_id
+vx_warp_id:
+	csrr a0, CSR_LWID
+	ret
+
+.type vx_warp_gid, @function
+.global vx_warp_gid
+vx_warp_gid:
+	csrr a0, CSR_GWID
+	ret
+
+.type vx_thread_id, @function
+.global vx_thread_id
+vx_thread_id:
+	csrr a0, CSR_LTID
+	ret
+
+.type vx_thread_gid, @function
+.global vx_thread_gid
+vx_thread_gid:
+	csrr a0, CSR_GTID
+	ret
+
+.type vx_core_id, @function
+.global vx_core_id
+vx_core_id:
+	csrr a0, CSR_GCID
+	ret
+
+.type vx_num_threads, @function
+.global vx_num_threads
+vx_num_threads:
+    csrr a0, CSR_NT
+    ret
+
+.type vx_num_warps, @function
+.global vx_num_warps
+vx_num_warps:
+    csrr a0, CSR_NW
+    ret
+
+.type vx_num_cores, @function
+.global vx_num_cores
+vx_num_cores:
+    csrr a0, CSR_NC
+    ret
+	
+.type vx_num_cycles, @function
+.global vx_num_cycles
+vx_num_cycles:
+    csrr a0, CSR_CYCLL
+    ret
+    
+.type vx_num_instrs, @function
+.global vx_num_instrs
+vx_num_instrs:
+    csrr a0, CSR_INSTL
+    ret

runtime/src/vx_print.S

@@ -0,0 +1,33 @@
+#include <VX_config.h>
+
+.type vx_print_str, @function
+.global vx_print_str
+vx_print_str:
+    addi sp, sp, -12
+    sw   ra, 0(sp)
+    sw   a1, 4(sp)
+bl:
+    lbu  a1,0(a0)
+    beqz a1,be
+    jal  vx_printc
+    addi a0, a0, 1
+    j bl
+be:
+    lw   ra, 0(sp)
+    lw   a1, 4(sp)
+    addi sp, sp, 12
+    ret
+
+
+.type vx_printc, @function
+.global vx_printc
+vx_printc:
+    la t0, print_addr
+    lw t0, 0(t0)
+    sw a1, 0(t0)
+    ret
+
+.section .data
+print_addr:
+    .word IO_BUS_ADDR_COUT 
+

runtime/src/vx_print.c

@@ -0,0 +1,40 @@
+
+#include <vx_print.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static char * hextoa[] = {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "a", "b", "c", "d", "e", "f"};
+
+void vx_print_hex(unsigned f)
+{
+	// vx_print_str(hextoa[f]);
+	if (f < 16)
+	{
+		vx_print_str(hextoa[f]);
+		return;
+	}
+	int temp;
+	int sf = 32;
+	bool start = false;
+	do
+	{
+		temp = (f >> (sf - 4)) & 0xf;
+		if (temp != 0) start = true;
+		if (start) vx_print_str(hextoa[temp]);
+		sf -= 4;
+	} while(sf > 0);
+}
+
+
+void vx_printf(const char * c, unsigned f)
+{
+	vx_print_str(c);
+	vx_print_hex(f);
+	vx_print_str("\n");
+}
+
+#ifdef __cplusplus
+}
+#endif

runtime/src/vx_spawn.c

@@ -0,0 +1,77 @@
+#include <vx_spawn.h>
+#include <vx_intrinsics.h>
+#include <inttypes.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+func_t global_function_pointer;
+void * global_argument_struct;
+int global_num_threads;
+
+void spawn_warp_runonce() {
+	// active all threads
+	vx_tmc(global_num_threads);
+
+	// call user routine
+	global_function_pointer(global_argument_struct);
+
+	// resume single-thread execution on exit
+	int wid = vx_warp_id();
+	unsigned tmask = (0 == wid) ? 0x1 : 0x0; 
+	vx_tmc(tmask);
+}
+
+void vx_spawn_warps(int numWarps, int numThreads, func_t func_ptr, void * args) {
+	global_function_pointer = func_ptr;
+	global_argument_struct  = args;
+	global_num_threads      = numThreads;
+	if (numWarps > 1) {
+		vx_wspawn(numWarps, (unsigned)spawn_warp_runonce);
+	}
+	spawn_warp_runonce();
+}
+
+int                    pocl_threads;
+struct context_t *     pocl_ctx;
+vx_pocl_workgroup_func pocl_pfn;
+const void *           pocl_args;
+
+void pocl_spawn_warp_runonce() {
+	// active all threads
+	vx_tmc(pocl_threads);
+
+	int x = vx_thread_id();
+	int y = vx_warp_gid();
+
+	// call kernel routine
+	(pocl_pfn)(pocl_args, pocl_ctx, x, y, 0);
+
+	// resume single-thread execution on exit
+	int wid = vx_warp_id();
+	unsigned tmask = (0 == wid) ? 0x1 : 0x0;
+	vx_tmc(tmask);
+}
+
+void pocl_spawn(struct context_t * ctx, vx_pocl_workgroup_func pfn, const void * args) {
+	if (ctx->num_groups[2] > 1)	{
+		printf("ERROR: pocl_spawn doesn't support Z dimension yet!\n");
+		return;
+	}
+
+	pocl_threads = ctx->num_groups[0];
+	pocl_ctx     = ctx;
+	pocl_pfn     = pfn;
+	pocl_args    = args;
+
+	if (ctx->num_groups[1] > 1)	{
+		vx_wspawn(ctx->num_groups[1], (unsigned)&pocl_spawn_warp_runonce);
+	}
+
+	pocl_spawn_warp_runonce();
+}
+
+#ifdef __cplusplus
+}
+#endif

runtime/src/vx_start.S

@@ -0,0 +1,68 @@
+#include <VX_config.h>
+
+.section .init, "ax"
+.global _start
+.type   _start, @function
+_start:
+  la a1, vx_set_sp
+  csrr a0, CSR_NW  # get num warps
+  .word 0x00b5106b # wspawn a0(numWarps), a1(PC SPAWN)
+  jal vx_set_sp
+  li a0, 1
+  .word 0x0005006b    # back to single thread
+  # Initialize global pointerp
+  # call __cxx_global_var_init
+  # Clear the bss segment
+  la      a0, _edata
+  la      a2, _end
+  sub     a2, a2, a0
+  li      a1, 0
+  call    memset
+  la      a0, __libc_fini_array   # Register global termination functions
+  call    atexit                  # to be called upon exit
+  call    __libc_init_array       # Run global initialization functions
+  call    main
+  tail    exit
+.size  _start, .-_start
+
+.section .text
+.type _exit, @function
+.global _exit
+_exit:
+  li a0, 0
+  .word 0x0005006b    # disable all threads
+
+.section .text
+.type vx_set_sp, @function
+.global vx_set_sp
+vx_set_sp:
+  csrr a0, CSR_NT     # get num threads
+  .word 0x0005006b    # activate all threads
+  
+  .option push
+  .option norelax
+  1:auipc gp, %pcrel_hi(__global_pointer$)
+    addi  gp, gp, %pcrel_lo(1b)
+  .option pop
+
+  csrr a1, CSR_GTID    # get global thread id
+  slli a1, a1, 10      # multiply by 1024
+  csrr a2, CSR_LTID    # get local thread id  
+  slli a2, a2, 2       # multiply by 4
+  lui  sp, STACK_BASE_ADDR # load base sp
+  sub  sp, sp, a1      # sub thread block
+  add  sp, sp, a2      # reduce addr collision for perf
+
+  csrr a3, CSR_LWID    # get wid
+  beqz a3, RETURN
+  li a0, 0
+  .word 0x0005006b     # tmc 0
+RETURN:
+  ret
+
+.section .data
+	.global __dso_handle
+	.weak __dso_handle
+__dso_handle:
+	.long	0
+