fixed loader script stack setup

driver/include/vortex.h

@@ -17,11 +17,15 @@ typedef void* vx_buffer_h;
 #define VX_CAPS_MAX_CORES         0x1
 #define VX_CAPS_MAX_WARPS         0x2
 #define VX_CAPS_MAX_THREADS       0x3
-#define VX_CAPS_CACHE_LINESIZE    0x4
+#define VX_CAPS_CACHE_LINE_SIZE   0x4
 #define VX_CAPS_LOCAL_MEM_SIZE    0x5
 #define VX_CAPS_ALLOC_BASE_ADDR   0x6
 #define VX_CAPS_KERNEL_BASE_ADDR  0x7
 
+#define CACHE_BLOCK_SIZE 64
+#define ALLOC_BASE_ADDR  0x00000000
+#define LOCAL_MEM_SIZE   0xffffffff
+
 // open the device and connect to it
 int vx_dev_open(vx_device_h* hdevice);
 

driver/opae/vlsim/opae_sim.cpp

@@ -3,7 +3,7 @@
 #include <fstream>
 #include <iomanip>
 
-#define CCI_LATENCY 8
+#define CCI_LATENCY  8
 #define CCI_RAND_MOD 8
 #define CCI_RQ_SIZE 16
 #define CCI_WQ_SIZE 16
@@ -204,11 +204,11 @@ void opae_sim::sRxPort_bus() {
   if (!mmio_req_enabled 
    && (cci_rd_it != cci_reads_.end())) {
     vortex_afu_->vcp2af_sRxPort_c0_rspValid = 1;
-    memcpy(vortex_afu_->vcp2af_sRxPort_c0_data, cci_rd_it->block.data(), CACHE_BLOCK_SIZE);
+    memcpy(vortex_afu_->vcp2af_sRxPort_c0_data, cci_rd_it->data.data(), CACHE_BLOCK_SIZE);
     vortex_afu_->vcp2af_sRxPort_c0_hdr_mdata = cci_rd_it->mdata;    
     /*printf("%0ld: [sim] CCI Rd Rsp: addr=%ld, mdata=%d, data=", timestamp, cci_rd_it->addr, cci_rd_it->mdata);
     for (int i = 0; i < CACHE_BLOCK_SIZE; ++i)
-      printf("%02x", cci_rd_it->block[CACHE_BLOCK_SIZE-1-i]);
+      printf("%02x", cci_rd_it->data[CACHE_BLOCK_SIZE-1-i]);
     printf("\n");*/
     cci_reads_.erase(cci_rd_it);
   }
@@ -223,7 +223,7 @@ void opae_sim::sTxPort_bus() {
     cci_req.addr = vortex_afu_->af2cp_sTxPort_c0_hdr_address;
     cci_req.mdata = vortex_afu_->af2cp_sTxPort_c0_hdr_mdata;
     auto host_ptr = (uint64_t*)(vortex_afu_->af2cp_sTxPort_c0_hdr_address * CACHE_BLOCK_SIZE);
-    memcpy(cci_req.block.data(), host_ptr, CACHE_BLOCK_SIZE);
+    memcpy(cci_req.data.data(), host_ptr, CACHE_BLOCK_SIZE);
     //printf("%0ld: [sim] CCI Rd Req: addr=%ld, mdata=%d\n", timestamp, vortex_afu_->af2cp_sTxPort_c0_hdr_address, cci_req.mdata);
     cci_reads_.emplace_back(cci_req);    
   }
@@ -262,7 +262,7 @@ void opae_sim::avs_bus() {
   vortex_afu_->avs_readdatavalid = 0;  
   if (dram_rd_it != dram_reads_.end()) {
     vortex_afu_->avs_readdatavalid = 1;
-    memcpy(vortex_afu_->avs_readdata, dram_rd_it->block.data(), CACHE_BLOCK_SIZE);
+    memcpy(vortex_afu_->avs_readdata, dram_rd_it->data.data(), CACHE_BLOCK_SIZE);
     uint32_t addr = dram_rd_it->addr;
     dram_reads_.erase(dram_rd_it);
     /*printf("%0ld: [sim] DRAM Rd Rsp: addr=%x, pending={", timestamp, addr * CACHE_BLOCK_SIZE);
@@ -304,7 +304,7 @@ void opae_sim::avs_bus() {
       assert(0 == vortex_afu_->mem_bank_select);
       dram_rd_req_t dram_req;      
       dram_req.addr = vortex_afu_->avs_address;
-      ram_.read(vortex_afu_->avs_address * CACHE_BLOCK_SIZE, CACHE_BLOCK_SIZE, dram_req.block.data());      
+      ram_.read(vortex_afu_->avs_address * CACHE_BLOCK_SIZE, CACHE_BLOCK_SIZE, dram_req.data.data());      
       dram_req.cycles_left = DRAM_LATENCY;
       for (auto& rsp : dram_reads_) {
         if (dram_req.addr == rsp.addr) {

driver/opae/vlsim/opae_sim.h

@@ -40,13 +40,13 @@ private:
 
   typedef struct {
     int cycles_left;  
-    std::array<uint8_t, CACHE_BLOCK_SIZE> block;
+    std::array<uint8_t, CACHE_BLOCK_SIZE> data;
     uint32_t addr;
   } dram_rd_req_t;
 
   typedef struct {
     int cycles_left;  
-    std::array<uint8_t, CACHE_BLOCK_SIZE> block;
+    std::array<uint8_t, CACHE_BLOCK_SIZE> data;
     uint64_t addr;
     uint32_t mdata;
   } cci_rd_req_t;

driver/opae/vortex.cpp

@@ -22,10 +22,6 @@
 #include "vx_scope.h"
 #endif
 
-#define CACHE_BLOCK_SIZE 64
-#define ALLOC_BASE_ADDR  0x10000000
-#define LOCAL_MEM_SIZE   0xffffffff
-
 #define CHECK_RES(_expr)                                \
    do {                                                 \
      fpga_result res = _expr;                           \
@@ -104,7 +100,7 @@ extern int vx_dev_caps(vx_device_h hdevice, unsigned caps_id, unsigned *value) {
     case VX_CAPS_MAX_THREADS:
         *value = device->num_threads;
         break;
-    case VX_CAPS_CACHE_LINESIZE:
+    case VX_CAPS_CACHE_LINE_SIZE:
         *value = CACHE_BLOCK_SIZE;
         break;
     case VX_CAPS_LOCAL_MEM_SIZE:

driver/rtlsim/vortex.cpp

@@ -11,10 +11,6 @@
 #include <ram.h>
 #include <simulator.h>
 
-#define CACHE_LINESIZE  64
-#define ALLOC_BASE_ADDR 0x10000000
-#define LOCAL_MEM_SIZE  0xffffffff
-
 ///////////////////////////////////////////////////////////////////////////////
 
 inline size_t align_size(size_t size, size_t alignment) {        
@@ -31,7 +27,7 @@ public:
     vx_buffer(size_t size, vx_device* device) 
         : size_(size)
         , device_(device) {
-        auto aligned_asize = align_size(size, CACHE_LINESIZE);
+        auto aligned_asize = align_size(size, CACHE_BLOCK_SIZE);
         data_ = malloc(aligned_asize);
     }
 
@@ -75,7 +71,7 @@ public:
 
     int alloc_local_mem(size_t size, size_t* dev_maddr) {
         auto dev_mem_size = LOCAL_MEM_SIZE;
-        size_t asize = align_size(size, CACHE_LINESIZE);        
+        size_t asize = align_size(size, CACHE_BLOCK_SIZE);        
         if (mem_allocation_ + asize > dev_mem_size)
             return -1;
         *dev_maddr = mem_allocation_;
@@ -84,7 +80,7 @@ public:
     }
 
     int upload(void* src, size_t dest_addr, size_t size, size_t src_offset) {
-        size_t asize = align_size(size, CACHE_LINESIZE);
+        size_t asize = align_size(size, CACHE_BLOCK_SIZE);
         if (dest_addr + asize > ram_.size())
             return -1;
 
@@ -98,7 +94,7 @@ public:
     }
 
     int download(const void* dest, size_t src_addr, size_t size, size_t dest_offset) {
-        size_t asize = align_size(size, CACHE_LINESIZE);
+        size_t asize = align_size(size, CACHE_BLOCK_SIZE);
         if (src_addr + asize > ram_.size())
             return -1;
 
@@ -189,8 +185,8 @@ extern int vx_dev_caps(vx_device_h hdevice, unsigned caps_id, unsigned *value) {
     case VX_CAPS_MAX_THREADS:
         *value = NUM_THREADS;
         break;
-    case VX_CAPS_CACHE_LINESIZE:
-        *value = CACHE_LINESIZE;
+    case VX_CAPS_CACHE_LINE_SIZE:
+        *value = CACHE_BLOCK_SIZE;
         break;
     case VX_CAPS_LOCAL_MEM_SIZE:
         *value = 0xffffffff;

driver/simx/vortex.cpp

@@ -11,10 +11,7 @@
 #include <core.h>
 #include <VX_config.h>
 
-#define CACHE_LINESIZE  64
 #define PAGE_SIZE       4096
-#define ALLOC_BASE_ADDR 0x10000000
-#define LOCAL_MEM_SIZE  0xffffffff
 
 ///////////////////////////////////////////////////////////////////////////////
 
@@ -32,7 +29,7 @@ public:
     vx_buffer(size_t size, vx_device* device) 
         : size_(size)
         , device_(device) {
-        auto aligned_asize = align_size(size, CACHE_LINESIZE);
+        auto aligned_asize = align_size(size, CACHE_BLOCK_SIZE);
         data_ = malloc(aligned_asize);
     }
 
@@ -81,7 +78,7 @@ public:
 
     int alloc_local_mem(size_t size, size_t* dev_maddr) {
         auto dev_mem_size = LOCAL_MEM_SIZE;
-        auto asize = align_size(size, CACHE_LINESIZE);        
+        auto asize = align_size(size, CACHE_BLOCK_SIZE);        
         if (mem_allocation_ + asize > dev_mem_size)
             return -1;
         *dev_maddr = mem_allocation_;
@@ -90,7 +87,7 @@ public:
     }
 
     int upload(void* src, size_t dest_addr, size_t size, size_t src_offset) {
-        auto asize = align_size(size, CACHE_LINESIZE);
+        auto asize = align_size(size, CACHE_BLOCK_SIZE);
         if (dest_addr + asize > ram_.size())
             return -1;
 
@@ -104,7 +101,7 @@ public:
     }
 
     int download(const void* dest, size_t src_addr, size_t size, size_t dest_offset) {
-        size_t asize = align_size(size, CACHE_LINESIZE);
+        size_t asize = align_size(size, CACHE_BLOCK_SIZE);
         if (src_addr + asize > ram_.size())
             return -1;
 
@@ -236,8 +233,8 @@ extern int vx_dev_caps(vx_device_h hdevice, unsigned caps_id, unsigned *value) {
     case VX_CAPS_MAX_THREADS:
         *value = NUM_THREADS;
         break;
-    case VX_CAPS_CACHE_LINESIZE:
-        *value = CACHE_LINESIZE;
+    case VX_CAPS_CACHE_LINE_SIZE:
+        *value = CACHE_BLOCK_SIZE;
         break;
     case VX_CAPS_LOCAL_MEM_SIZE:
         *value = LOCAL_MEM_SIZE;

hw/rtl/VX_config.vh

@@ -32,7 +32,7 @@
 `endif
 
 `ifndef SM_ENABLE
-`define SM_ENABLE 0
+`define SM_ENABLE 1
 `endif
 
 `ifndef GLOBAL_BLOCK_SIZE
@@ -47,14 +47,14 @@
 `define STARTUP_ADDR 32'h80000000
 `endif
 
-`ifndef SHARED_MEM_BASE_ADDR
-`define SHARED_MEM_BASE_ADDR 32'h6FFFF000
-`endif
-
 `ifndef IO_BUS_BASE_ADDR
 `define IO_BUS_BASE_ADDR 32'hFF000000
 `endif
 
+`ifndef SHARED_MEM_BASE_ADDR
+`define SHARED_MEM_BASE_ADDR `IO_BUS_BASE_ADDR
+`endif
+
 `ifndef IO_BUS_ADDR_COUT
 `define IO_BUS_ADDR_COUT 32'hFFFFFFFC
 `endif
@@ -139,16 +139,6 @@
 `define CSR_FRM         12'h002
 `define CSR_FCSR        12'h003
 
-// SIMT CSRs
-`define CSR_LTID        12'h020
-`define CSR_LWID        12'h021
-`define CSR_GTID        12'h022
-`define CSR_GWID        12'h023
-`define CSR_GCID        12'h024
-`define CSR_NT          12'h025
-`define CSR_NW          12'h026
-`define CSR_NC          12'h027
-
 `define CSR_SATP        12'h180
 
 `define CSR_PMPCFG0     12'h3A0
@@ -236,6 +226,19 @@
 `define CSR_MIMPID      12'hF13
 `define CSR_MHARTID     12'hF14
 
+// User SIMT CSRs
+`define CSR_WTID        12'hCC0
+`define CSR_LTID        12'hCC1
+`define CSR_GTID        12'hCC2
+`define CSR_LWID        12'hCC3
+`define CSR_GWID        `CSR_MHARTID
+`define CSR_GCID        12'hCC5
+
+// Machine SIMT CSRs
+`define CSR_NT          12'hFC0
+`define CSR_NW          12'hFC1
+`define CSR_NC          12'hFC2
+
 // Pipeline Queues ////////////////////////////////////////////////////////////
 
 // Size of instruction queue
@@ -324,9 +327,14 @@
 
 // SM Configurable Knobs //////////////////////////////////////////////////////
 
+// Size of cache block in bytes
+`ifndef SM_BLOCK_SIZE
+`define SM_BLOCK_SIZE 1024
+`endif
+
 // Size of cache in bytes
 `ifndef SMEM_SIZE
-`define SMEM_SIZE 8192
+`define SMEM_SIZE (`NUM_WARPS * `NUM_THREADS * `SM_BLOCK_SIZE)
 `endif
 
 // Number of banks

hw/rtl/VX_csr_data.v

@@ -110,9 +110,10 @@ module VX_csr_data #(
             `CSR_FCSR      : read_data_r = 32'(csr_fcsr[read_wid]);
 
             `CSR_LWID      : read_data_r = 32'(read_wid);
+            `CSR_WTID      ,
             `CSR_LTID      ,
             `CSR_GTID      ,
-            `CSR_MHARTID   ,
+            /*`CSR_MHARTID ,*/
             `CSR_GWID      : read_data_r = CORE_ID * `NUM_WARPS + 32'(read_wid);
             `CSR_GCID      : read_data_r = CORE_ID;
             `CSR_NT        : read_data_r = `NUM_THREADS;

hw/rtl/VX_csr_unit.v

@@ -120,8 +120,9 @@ module VX_csr_unit #(
     );
 
     for (genvar i = 0; i < `NUM_THREADS; i++) begin
-        assign csr_pipe_rsp_if.data[i] = (csr_addr_s1 == `CSR_LTID) ? i : 
-                                         (csr_addr_s1 == `CSR_GTID) ? (csr_read_data_s1 * `NUM_THREADS + i) : 
+        assign csr_pipe_rsp_if.data[i] = (csr_addr_s1 == `CSR_WTID) ? i : 
+                                         (csr_addr_s1 == `CSR_LTID 
+                                       || csr_addr_s1 == `CSR_GTID) ? (csr_read_data_s1 * `NUM_THREADS + i) : 
                                                                       csr_read_data_s1;
     end         
 

hw/rtl/VX_define.vh

@@ -319,7 +319,7 @@
 `define SCACHE_ID               (32'(`L3_ENABLE) + 32'(`L2_ENABLE) * `NUM_CLUSTERS + CORE_ID * 3 + 2)
 
 // Block size in bytes
-`define SBANK_LINE_SIZE         4 
+`define SBANK_LINE_SIZE         `SM_BLOCK_SIZE 
 
 // Word size in bytes
 `define SWORD_SIZE              4

runtime/include/vx_intrinsics.h

@@ -20,21 +20,24 @@ void vx_split(int predicate);
 // Join
 void vx_join();
 
-// Return the warp's unique thread id 
+// Return active warp's thread id 
 int vx_thread_id();
 
-// Return the core's unique warp id
-int vx_warp_id();
-
-// Return processsor unique core id
-int vx_core_id();
+// Return active core's local thread id
+int vx_thread_lid();
 
 // Return processsor global thread id
 int vx_thread_gid();
 
-// Return processsor global warp id
+// Return active core's local warp id
+int vx_warp_id();
+
+// Return processsor's global warp id
 int vx_warp_gid();
 
+// Return processsor core id
+int vx_core_id();
+
 // Return the number of threads in a warp
 int vx_num_threads();
 

runtime/linker/vx_link.ld

@@ -9,7 +9,6 @@ OUTPUT_ARCH(riscv)
 ENTRY(_start)
 SECTIONS
 {
-  PROVIDE(__stack_top = 0x6ffff000);
   . = 0x80000000;
   .interp         : { *(.interp) }
   .note.gnu.build-id  : { *(.note.gnu.build-id) }
@@ -191,10 +190,24 @@ SECTIONS
   . = SEGMENT_START("ldata-segment", .);
   . = ALIGN(32 / 8);
   __BSS_END__ = .;
-    __global_pointer$ = MIN(__SDATA_BEGIN__ + 0x800,
+    __global_pointer = MIN(__SDATA_BEGIN__ + 0x800,
                 MAX(__DATA_BEGIN__ + 0x800, __BSS_END__ - 0x800));
   _end = .; PROVIDE (end = .);
-  . = DATA_SEGMENT_END (.);
+  . = DATA_SEGMENT_END (.);  
+  
+  /* .stack_dummy section doesn't contains any symbols. It is only
+	 * used for linker to calculate size of stack sections, and assign
+	 * values to stack symbols later */
+	.stack_dummy (COPY):
+	{
+		KEEP(*(.stack*))
+	}
+  __stack_usage = SIZEOF(.stack_dummy);
+  PROVIDE(__stack_top  = 0xFF000000);
+  PROVIDE(__stack_size = 0x400);
+	PROVIDE(__stack = __stack_top);
+	ASSERT(__stack_usage <= __stack_size, "stack overflow")
+
   /* Stabs debugging sections.  */
   .stab          0 : { *(.stab) }
   .stabstr       0 : { *(.stabstr) }

runtime/src/vx_intrinsics.S

@@ -47,6 +47,12 @@ vx_warp_gid:
 .type vx_thread_id, @function
 .global vx_thread_id
 vx_thread_id:
+	csrr a0, CSR_WTID
+	ret
+
+.type vx_thread_lid, @function
+.global vx_thread_lid
+vx_thread_lid:
 	csrr a0, CSR_LTID
 	ret
 

runtime/src/vx_spawn.c

@@ -6,7 +6,7 @@
 extern "C" {
 #endif
 
-#define NUM_CORES_MAX 8
+#define NUM_CORES_MAX 16
 
 typedef struct {
 	func_t function;

runtime/src/vx_start.S

@@ -57,18 +57,20 @@ vx_set_sp:
   # set global pointer register
   .option push
   .option norelax
-  la gp, __global_pointer$
+  la gp, __global_pointer
   .option pop
 
   # allocate stack region for a threads on the processor 
   # set stack pointer
-  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
-  la sp, __stack_top$  # load stack base address
+  la sp, __stack_top   # load stack base address
+  la a1, __stack_size  # stack size
+  #if SM_ENABLE
+  csrr a2, CSR_LTID    # get lobal thread id
+  #else
+  csrr a2, CSR_GTID    # get global thread id
+  #endif
+  mul  a1, a1, a2
   sub  sp, sp, a1      # sub thread block
-  add  sp, sp, a2      # reduce addr collision for perf
 
   # disable active warps except warp0
   csrr a3, CSR_LWID    # get local wid

runtime/tests/dev/vx_dev_main.c

@@ -1,14 +1,13 @@
-
-#include <vx_intrinsics.h>
-
-
-// #include <utlist.h>
 #include <string.h>
 #include <unistd.h>
 #include <stdint.h>
 #include <stdbool.h>
 #include <stdlib.h>
 
+#include <vx_intrinsics.h>
+#include <vx_print.h>
+#include <vx_spawn.h>
+
 typedef struct
 {
 	unsigned * x;

runtime/tests/nlTest/vx_nl_main.c

@@ -1,16 +1,5 @@
-
 #include <vx_intrinsics.h>
-
-// #include <utlist.h>
-#include <string.h>
-#include <unistd.h>
-#include <stdint.h>
-#include <stdbool.h>
-#include <stdlib.h>
-
-
-// Newlib
-#include <stdio.h>
+#include <vx_print.h>
 
 int main()
 {