Sync with local branch, phase 2 done

Xtra-Computing · Aug 28, 2020 · c512fd2 · c512fd2
1 parent 5ee345a
commit c512fd2
Show file tree

Hide file tree

Showing 25 changed files with 433 additions and 58 deletions.
diff --git a/.gitignore b/.gitignore
@@ -19,4 +19,6 @@ csr
 *.xclbin
 *.sh
 tmp_fpga_top
-code_gen
+tmp_para
+code_gen
+para_gen
diff --git a/application/ar/main.cpp b/application/ar/main.cpp
@@ -55,7 +55,7 @@ int main(int argc, char **argv) {
         endStamp = getCurrentTimestamp();
 
         /* profile */
-        accelratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
+        acceleratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
     }
     acceleratorDeinit();
 

diff --git a/application/bfs/main.cpp b/application/bfs/main.cpp
@@ -63,7 +63,7 @@ int main(int argc, char **argv) {
 
 
         /* profile */
-        accelratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
+        acceleratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
         runCounter++;
     }
     acceleratorDeinit();

diff --git a/application/cc/main.cpp b/application/cc/main.cpp
@@ -66,7 +66,7 @@ int main(int argc, char **argv) {
         }
 
         /* profile */
-        accelratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
+        acceleratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
         runCounter ++;
     }
     acceleratorDeinit();

diff --git a/application/common.mk b/application/common.mk
@@ -31,7 +31,8 @@ else
 	HOST_SRCS +=  $(APPCONFIG)/main.cpp
 endif
 
-HOST_SRCS += ./libgraph/memory/he_mem.cpp 
+HOST_SRCS += ./libgraph/memory/he_mem.cpp
+HOST_SRCS += ./libgraph/memory/he_mapping.cpp
 HOST_SRCS += ./libgraph/host_graph_sw_mem.cpp
 HOST_SRCS += ./libgraph/host_graph_sw_partition.cpp
 HOST_SRCS += ./libgraph/host_graph_sw_kernel.cpp

diff --git a/docs/api_details.md b/docs/api_details.md
@@ -182,6 +182,6 @@ inline prop_t applyCalculation( prop_t tProp,
 |acceleratorDataPrepare  | load the graph data |
 |acceleratorDataPreprocess  | graph partitioning |
 |acceleratorSuperStep  | process all of the partitions once |
-|accelratorProfile  | profile the execution time and verify the result |
+|acceleratorProfile  | profile the execution time and verify the result |
 |acceleratorDeinit  | release all the dynamic resources |
 
diff --git a/docs/verification.md b/docs/verification.md
@@ -1,3 +1,3 @@
 # Verification
 
-As we have a cache in ThunderGP, the static verification method (emu-sw) can not be used for the verification and validation of our framework. ThunderGP provides an automatic verification scheme, by calling the function ```accelratorProfile``` after the calcuation, the verification will be automatically performed, and it will output the difference between the result from hardware unit and the golden. Currently it only supports the verification on the fisrt super step.
+As we have a cache in ThunderGP, the static verification method (emu-sw) can not be used for the verification and validation of our framework. ThunderGP provides an automatic verification scheme, by calling the function ```acceleratorProfile``` after the calcuation, the verification will be automatically performed, and it will output the difference between the result from hardware unit and the golden. Currently it only supports the verification on the fisrt super step.
diff --git a/libfpga/fpga_edge_prop.h b/libfpga/fpga_edge_prop.h
@@ -55,6 +55,7 @@ void edgePropCouple (
             break;
         }
     }
+    clear_stream(input);
     clear_stream(edgeProp);
 }
 

diff --git a/libfpga/generator/devices/device_common.h b/libfpga/generator/devices/device_common.h
@@ -0,0 +1,15 @@
+#ifndef __DEVICE_COMMON_H__
+#define __DEVICE_COMMON_H__
+
+typedef struct 
+{
+	int slr_id;
+	int luts;
+	int ffs;
+	int dsp;
+	int bram;
+	int uram;
+} slr_resource_info_t;
+
+
+#endif /* __DEVICE_COMMON_H__ */
diff --git a/libfpga/generator/devices/xilinx_u200_xdma_201830_2.h b/libfpga/generator/devices/xilinx_u200_xdma_201830_2.h
@@ -0,0 +1,36 @@
+#include "device_common.h"
+
+const char * board_name = "u200";
+
+const slr_resource_info_t slrs[] =
+{
+    {
+        .slr_id  = 0,
+        .luts    = 354831,
+        .ffs     = 723372,
+        .dsp     = 2265,
+        .bram    = 638,
+        .uram    = 320,
+    },
+    {
+        .slr_id  = 1,
+        .luts    = 159854,
+        .ffs     = 331711,
+        .dsp     = 1317,
+        .bram    = 326,
+        .uram    = 160,
+    },
+    {
+        .slr_id  = 2,
+        .luts    = 354962,
+        .ffs     = 723353,
+        .dsp     = 2265,
+        .bram    = 638,
+        .uram    = 320,
+    },
+};
+
+const int mem_cu_map[] =
+{
+    0, 1, 1, 2
+};
diff --git a/libfpga/generator/devices/xilinx_u250_xdma_201830_2.h b/libfpga/generator/devices/xilinx_u250_xdma_201830_2.h
@@ -0,0 +1,45 @@
+#include "device_common.h"
+
+const char * board_name = "u250";
+
+const slr_resource_info_t slrs[] =
+{
+    {
+        .slr_id  = 0,
+        .luts    = 345171,
+        .ffs     = 704801,
+        .dsp     = 2877,
+        .bram    = 500,
+        .uram    = 320,
+    },
+    {
+        .slr_id  = 1,
+        .luts    = 344533,
+        .ffs     = 702517,
+        .dsp     = 2877,
+        .bram    = 500,
+        .uram    = 320,
+    },
+    {
+        .slr_id  = 2,
+        .luts    = 344878,
+        .ffs     = 703253,
+        .dsp     = 2877,
+        .bram    = 500,
+        .uram    = 0,
+    },
+    {
+        .slr_id  = 3,
+        .luts    = 345158,
+        .ffs     = 703761,
+        .dsp     = 2877,
+        .bram    = 500,
+        .uram    = 320,
+    },
+};
+
+const int mem_cu_map[] =
+{
+    0, 1, 2, 3
+};
+
diff --git a/libfpga/generator/devices/xilinx_vcu1525_xdma_201830_1.h b/libfpga/generator/devices/xilinx_vcu1525_xdma_201830_1.h
@@ -0,0 +1,36 @@
+#include "device_common.h"
+
+const char * board_name = "vcu1525";
+
+const slr_resource_info_t slrs[] =
+{
+    {
+        .slr_id  = 0,
+        .luts    = 354830,
+        .ffs     = 723371,
+        .dsp     = 2265,
+        .bram    = 638,
+        .uram    = 320,
+    },
+    {
+        .slr_id  = 1,
+        .luts    = 159088,
+        .ffs     = 329162,
+        .dsp     = 1317,
+        .bram    = 326,
+        .uram    = 160,
+    },
+    {
+        .slr_id  = 2,
+        .luts    = 354934,
+        .ffs     = 723328,
+        .dsp     = 2265,
+        .bram    = 638,
+        .uram    = 320,
+    },
+};
+
+const int mem_cu_map[] =
+{
+    0, 1, 1, 2
+};
diff --git a/libfpga/generator/gs_gen.cpp b/libfpga/generator/gs_gen.cpp
diff --git a/libfpga/generator/para_gen.cpp b/libfpga/generator/para_gen.cpp
@@ -0,0 +1,13 @@
+#include <stdio.h>
+
+#include "device_common.h"
+
+#ifndef DEVICE_HEADER
+#error "no board name"
+#endif
+
+#include DEVICE_HEADER
+
+int main(int argc, char **argv) {
+	printf("%s \n", board_name);
+}
diff --git a/libfpga/generator/parser.mk b/libfpga/generator/parser.mk
@@ -13,3 +13,9 @@ code_gen:
 	./code_gen libfpga/common/apply_top.cpp             tmp_fpga_top/apply_top
 	./code_gen libfpga/common/scatter_gather_top.cpp    tmp_fpga_top/scatter_gather_top
 
+.PHONY: para_gen
+para_gen:
+	rm -rf tmp_para
+	mkdir -p tmp_para
+	g++  -static-libstdc++ -I libfpga/generator/devices -DDEVICE_HEADER="$(DEVICES).h"  $(CODE_GEN_PATH)/para_gen.cpp -o para_gen
+
diff --git a/libgraph/common.h b/libgraph/common.h
@@ -9,10 +9,20 @@
 
 #include "global_config.h"
 
+#define EXPAND(x)                   x
 
 
-#define FLAG_SET                (1u)
-#define FLAG_RESET              (0u)
+#define STRINGIFY_MACRO(x)          STR(x)
+#define STR(x)                      #x
+
+#define CONCAT4(n1, n2, n3, n4)     STRINGIFY_MACRO(EXPAND(n1)EXPAND(n2)EXPAND(n3)EXPAND(n4))
+
+#define CAT_SECOND_LEVLE(x, y)      x ## y
+#define VAR_CONCAT2(x, y)           CAT_SECOND_LEVLE(x, y)
+
+
+#define FLAG_SET                    (1u)
+#define FLAG_RESET                  (0u)
 
 
 #if 1

diff --git a/libgraph/default_entry.cpp b/libgraph/default_entry.cpp
@@ -49,7 +49,7 @@ int main(int argc, char **argv) {
     /* for verification */
     acceleratorCModelDataPreprocess(&graphDataInfo);
 
-    for (int runCounter = 0 ; runCounter < 2 ; runCounter ++)
+    for (int runCounter = 0 ; runCounter < 3 ; runCounter ++)
     {
         double startStamp, endStamp;
         startStamp = getCurrentTimestamp();
@@ -61,7 +61,7 @@ int main(int argc, char **argv) {
         acceleratorCModelSuperStep(runCounter, &graphDataInfo);
 
         /* for profile */
-        accelratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
+        acceleratorProfile(runCounter, runCounter, &graphDataInfo, endStamp - startStamp);
     }
     acceleratorDeinit();
 

diff --git a/libgraph/host_graph_sw.h b/libgraph/host_graph_sw.h
@@ -154,7 +154,7 @@ void* acceleratorQueryRegister(void);
 prop_t* acceleratorQueryProperty(int step);
 
 
-int accelratorProfile (int superStep, int runCounter, graphInfo *info, double exeTime);
+int acceleratorProfile (int superStep, int runCounter, graphInfo *info, double exeTime);
 
 int acceleratorCModelDataPreprocess(graphInfo *info);
 
@@ -168,6 +168,10 @@ int dataPrepareProperty(graphInfo *info);
 int getStartIndex(void);
 
 
+inline int getCuIDbyInterface(int order)
+{
+    return he_get_interface_id(order);
+}
 
 
 double getCurrentTimestamp(void);

diff --git a/libgraph/host_graph_sw_kernel.cpp b/libgraph/host_graph_sw_kernel.cpp
@@ -83,9 +83,7 @@ void setGsKernel(int partId, int superStep, graphInfo *info)
 #if HW_EMU_DEBUG
         edgeEnd         = HW_EMU_DEBUG_SIZE;
 #endif
-        //DEBUG_PRINTF("gs task in cu [%d] info:\n", i);
-        //DEBUG_PRINTF("\tedge  %d %d \n", 0, edgeEnd);
-        //DEBUG_PRINTF("\tsink  %d %d \n", sinkStart, sinkEnd);
+
         clSetKernelArg(gsHandler->kernel, argvi++, sizeof(cl_mem), get_cl_mem_pointer(partition->edgeHead.id));
         clSetKernelArg(gsHandler->kernel, argvi++, sizeof(cl_mem), get_cl_mem_pointer(gsHandler->prop[currentPropId].id));
         clSetKernelArg(gsHandler->kernel, argvi++, sizeof(cl_mem), get_cl_mem_pointer(partition->edgeTail.id));
@@ -105,10 +103,6 @@ void setGsKernel(int partId, int superStep, graphInfo *info)
 
 #if  CUSTOMIZE_APPLY == 0
 
-const int applyGlobalMemoryIndex[] =
-{
-    2, 1, 0, 3
-};
 
 void setApplyKernel(int partId, int superStep, graphInfo *info)
 {
@@ -127,20 +121,20 @@ void setApplyKernel(int partId, int superStep, graphInfo *info)
     int offset = p_partition->dstVertexStart;
 
 
-    clSetKernelArg(applyHandler->kernel, argvi++, sizeof(cl_mem), get_cl_mem_pointer(getGatherScatter(applyGlobalMemoryIndex[0])->prop[currentPropId].id));
+    clSetKernelArg(applyHandler->kernel, argvi++, sizeof(cl_mem), get_cl_mem_pointer(getGatherScatter(getCuIDbyInterface(0))->prop[currentPropId].id));
 
-    for (int i = 0; i < 4; i++)
+    for (int i = 0; i < SUB_PARTITION_NUM; i++)
     {
         clSetKernelArg(applyHandler->kernel, argvi++, sizeof(cl_mem),
-                       get_cl_mem_pointer(getSubPartition(partId * SUB_PARTITION_NUM + applyGlobalMemoryIndex[i])->tmpProp.id)
+                       get_cl_mem_pointer(getSubPartition(partId * SUB_PARTITION_NUM + getCuIDbyInterface(i))->tmpProp.id)
                       );
     }
-    for (int i = 0; i < 4; i++)
+    for (int i = 0; i < SUB_PARTITION_NUM; i++)
     {
         clSetKernelArg(applyHandler->kernel, argvi++, sizeof(cl_mem),
-                       get_cl_mem_pointer(getGatherScatter(applyGlobalMemoryIndex[i])->prop[updatePropId].id)
+                       get_cl_mem_pointer(getGatherScatter(getCuIDbyInterface(i))->prop[updatePropId].id)
                       );
-        he_set_dirty(getGatherScatter(applyGlobalMemoryIndex[i])->prop[updatePropId].id);
+        he_set_dirty(getGatherScatter(getCuIDbyInterface(i))->prop[updatePropId].id);
     }
 
 #if HAVE_APPLY_OUTDEG