Merge pull request #310 from GreenWaves-Technologies/4.9_dev

4.9 dev

Makefile

@@ -178,11 +178,19 @@ openocd.checkout:
 	fi
 
 openocd.build: openocd.checkout
-	cd utils/openocd && ./bootstrap && ./configure --enable-jtag_dpi --prefix=$(INSTALL_DIR)/openocd && make && make install
+	cd utils/openocd && ./bootstrap && ./configure --enable-jtag_dpi --prefix=$(INSTALL_DIR)/openocd && $(MAKE) && $(MAKE) install
 
 openocd.clean:
 	rm -rf $(INSTALL_DIR)/openocd tools/openocd
 
+PROFILER_V2_DIR = $(GAP_SDK_HOME)/tools/profiler_v2
+PROFILER_V2_BUILD_DIR = $(GAP_SDK_HOME)/build/profiler_v2
+
+profiler_v2:
+	cmake -S $(PROFILER_V2_DIR) -B $(PROFILER_V2_BUILD_DIR)
+	cmake --build $(PROFILER_V2_BUILD_DIR)
+	cmake --install $(PROFILER_V2_BUILD_DIR) --prefix $(INSTALL_DIR)
+
 profiler:
 	$(MAKE) -C tools/profiler all
 	mkdir -p $(INSTALL_DIR)/bin

configs/common.sh

@@ -21,14 +21,14 @@ export NNTOOL_KERNELS_PATH=$NNTOOL_DIR/autotiler/kernels
 export NNTOOL_MATH_PATH=$NNTOOL_DIR/autotiler/math_funcs
 export NNTOOL_GENERATOR_PATH=$NNTOOL_DIR/autotiler/generators
 export PATH=$PATH:"$GAP_SDK_HOME"
-export PATH=$PATH:"$NNTOOL_DIR"
+export PATH="$NNTOOL_DIR":$PATH
 
 # OpenMP
 export OPENMP_DIR="$GAP_SDK_HOME/libs/openmp"
 
 # PulpOS 2
 export PULPOS_HOME=$GAP_SDK_HOME/rtos/pulp/pulpos-2
-export PULPOS_MODULES="$GAP_SDK_HOME/rtos/pulp/pulpos-2_gap8 $GAP_SDK_HOME/rtos/pulp/pulpos-2_gap9 $GAP_SDK_HOME/rtos/pmsis/pmsis_bsp $OPENMP_DIR $GAP_SDK_HOME/rtos/sfu"
+export PULPOS_MODULES="$GAP_SDK_HOME/rtos/pmsis/pmsis_implem $GAP_SDK_HOME/rtos/pulp/pulpos-2_gap8 $GAP_SDK_HOME/rtos/pulp/pulpos-2_gap9 $GAP_SDK_HOME/rtos/pmsis/pmsis_bsp $OPENMP_DIR $GAP_SDK_HOME/rtos/sfu"
 export PULPOS_GAP8_HOME=$GAP_SDK_HOME/rtos/pulp/pulpos-2_gap8
 export PULPOS_GAP9_HOME=$GAP_SDK_HOME/rtos/pulp/pulpos-2_gap9
 export GAP_PULPOS_ARCHI=$GAP_SDK_HOME/rtos/pulp/gap_archi
@@ -47,6 +47,9 @@ export RUNTIME_PATH=$GAP_SDK_HOME/pulp-os
 # For FreeRTOS
 export FREERTOS_PATH=$GAP_SDK_HOME/rtos/freeRTOS
 
+# For PMSIS
+export PMSIS_HOME=$GAP_SDK_HOME/rtos/pmsis
+
 export PATH="$INSTALL_DIR/bin":$PATH
 export LD_LIBRARY_PATH="$INSTALL_DIR/lib":$LD_LIBRARY_PATH
 export PYTHONPATH=$INSTALL_DIR/python:$PYTHONPATH
@@ -62,7 +65,7 @@ export GVSOC_SRC_PATH=$GAP_SDK_HOME/gvsoc/gvsoc
 export GVSOC_GAP_SRC_PATH=$GAP_SDK_HOME/gvsoc/gvsoc_gap
 export GVSOC_SFU_PATH=$GAP_SDK_HOME/gvsoc/gvsoc_gap_sfu
 source $GAP_SDK_HOME/gvsoc/setup_gvsoc.sh
-if [ -e "$GAP_SDK_HOME/configs/skip_udma_build" ]; then
+if [ -d "$GAP_SDK_HOME/gvsoc/gvsoc_libs" ]; then
     export CONFIG_GVSOC_SKIP_UDMA_BUILD=1
 fi
 export PYTHONPATH=$GAP_SDK_HOME/gvsoc/gvsoc_gap/models:$PYTHONPATH
@@ -73,7 +76,8 @@ export PYTHONPATH=$GAP_SDK_HOME/gvsoc/gvsoc/engine/python:$PYTHONPATH
 export PATH="$GAP_SDK_HOME/utils/gaptest":$PATH
 
 # Audio framework
-export PYTHONPATH=$GAP_SDK_HOME/tools/audio-framework/frontends/python_graph_generator:$GAP_SDK_HOME/tools/audio-framework/components:$PYTHONPATH
+export GAP_AUDIO_FRAMEWORK_HOME=$GAP_SDK_HOME/tools/audio-framework
+export PYTHONPATH=$GAP_AUDIO_FRAMEWORK_HOME/frontends/python_graph_generator:$GAP_AUDIO_FRAMEWORK_HOME/components:$PYTHONPATH
 
 
 # Autotiler

configs/gapuino_v3.sh

@@ -23,4 +23,6 @@ export OPENOCD_CABLE=interface/ftdi/gapuino_ftdi.cfg
 
 export GAPY_TARGET=gapuino_v3
 
+export PLPTEST_DEFAULT_PROPERTIES="chip=gap8_v3 chip_family=gap8 board=gapuino_v3 duration=50 test_duration=50"
+
 source $GAP_SDK_HOME/configs/common.sh

configs/openocd.sh

@@ -9,3 +9,7 @@ else
 fi
 
 export PATH=$GAP_SDK_HOME/install/workstation/openocd/bin:$PATH
+
+# Path to openocd scripts
+export OPENOCD_SCRIPTS=$GAP_SDK_HOME/utils/openocd_tools
+

doc/conf.py

@@ -35,6 +35,7 @@ def configure_doxyfile(file_in, file_out, replace_dict):
         "../rtos/pmsis/pmsis_api/include/pmsis/rtos/",
         "../rtos/pmsis/pmsis_api/include/pmsis/cluster/",
         "../rtos/pmsis/pmsis_api/include/pmsis/platforms/",
+        "../rtos/pmsis/pmsis_api/include/pmsis/",
         "../rtos/pmsis/pmsis_bsp/include/",
         "source/reference/builtins/headers/",
     ]
@@ -86,6 +87,10 @@ def configure_doxyfile(file_in, file_out, replace_dict):
 html_theme = "sphinx_rtd_theme"
 html_logo = "_static/logo.png"
 
+html_theme_options = {
+        'navigation_depth' : -1,
+}
+
 # Add any paths that contain custom static files (such as style sheets) here,
 # relative to this directory. They are copied after the builtin static files,
 # so a file named "default.css" will overwrite the builtin "default.css".

examples/autotiler/BilinearResize/Bilinear_Resize.c

@@ -115,13 +115,12 @@ void run_Bilinear_Resize(void)
     cluster_call.ImageOut    = ImageOut;
 
     /* Prepare task to be offload to Cluster. */
-    struct pi_cluster_task task = {0};
-    task.entry = (void *) cluster_main;
-    task.arg = &cluster_call;
+    struct pi_cluster_task task;
+    pi_cluster_task(&task, (void *) cluster_main, &cluster_call);
     task.stack_size = (uint32_t) STACK_SIZE;
 
     /* Execute the function "cluster_main" on the Core 0 of cluster. */
-    pi_cluster_send_task_to_cl(&cluster_dev, &task);
+    pi_cluster_send_task(&cluster_dev, &task);
 
     pi_l1_free(&cluster_dev, Resize_L1_Memory, _Resize_L1_Memory_SIZE);
 

examples/autotiler/Cifar10/Cifar10.c

@@ -336,12 +336,10 @@ void test_cifar10(void)
     }
 
     struct pi_cluster_task *task = pmsis_l2_malloc(sizeof(struct pi_cluster_task));
-    memset(task, 0, sizeof(struct pi_cluster_task));
-    task->entry = RunCifar10;
-    task->arg = NULL;
+    pi_cluster_task(task, RunCifar10, NULL);
 //    task->stack_size = 2048*2;
 
-    pi_cluster_send_task_to_cl(&cluster_dev, task);
+    pi_cluster_send_task(&cluster_dev, task);
 
     pmsis_l1_malloc_free(Cifar10_L1_Memory, _Cifar10_L1_Memory_SIZE);
 

examples/autotiler/FFT2DModel/TestFFT2D.c

@@ -103,12 +103,10 @@ void fft_2d(void)
     }
 
     struct pi_cluster_task *task = pmsis_l2_malloc(sizeof(struct pi_cluster_task));
-    memset(task, 0, sizeof(struct pi_cluster_task));
-    task->entry = (void *)Process;
-    task->arg = (void *) NULL;
+    pi_cluster_task(task, (void *)Process, (void *) NULL);
     task->stack_size = (uint32_t) STACK_SIZE;
 
-    pi_cluster_send_task_to_cl(&cluster_dev, task);
+    pi_cluster_send_task(&cluster_dev, task);
 
     // Close the cluster
     pi_cluster_close(&cluster_dev);

examples/autotiler/FFTL1/FFTRunTest.c

@@ -4,6 +4,12 @@
 	#define pmsis_exit(a)	exit(a)		
 #endif
 
+#ifndef SILENT
+    #define PRINTF printf
+#else
+    #define PRINTF(...) ((void) 0)
+#endif
+
 #define __XSTR(__s) __STR(__s)
 #define __STR(__s) #__s
 #include <stdlib.h>
@@ -21,7 +27,8 @@
 #endif
 #define  STACK_SIZE 			   2048
 typedef void (*FFTFun_T )(void *Data, void *Twiddles, signed char *shift, unsigned int Nfft, unsigned int Inverse);
-
+PI_L2 int PERF_ARR[6][3][2];
+PI_L2 float MSE_ARR[6][2];
 
 short int *InBuff_q16;
 float     *InBuff_f32, *InBuff_f32R4, *OutBuff_f32;
@@ -58,14 +65,14 @@ float MSE_f32(float* real, float* calc, int Size){
 void CallFFT(int Nfft, int Type){
 
 	// FFT: reset buffers, run and check mse
-	int start, elapsed, elapsedFFT, Q;
+	int start, elapsed, elapsedFFT, Q = 0;
 	FFT_InstallArg_T ArgIns;
 	FFT_Arg_T FFTArg;
 	AT_L2_EVENT DmaR_Evt1;
-	void (*FFTFun)(FFT_Arg_T*);
-	void (*SwapFun)(SwapSamples_Arg_T*);
+	void (*FFTFun)(FFT_Arg_T*) = 0;
+	void (*SwapFun)(SwapSamples_Arg_T*) = 0;
 	char *FFTDataType = 0;
-	void *InBuff;
+	void *InBuff = 0;
 
   	ArgIns.Nfft = Nfft; 
   	ArgIns.Radix = ((Nfft)==64 || (Nfft)==256 || (Nfft)==1024)?4:2;
@@ -96,11 +103,11 @@ void CallFFT(int Nfft, int Type){
 			FFTDataType = "Q16";
 			switch (Nfft) {
 				case 64:   ArgIns.Twiddles = R4_Twiddles_fix_64;   ArgIns.SwapLUT = R4_SwapTable_fix_64;   Q = 10; break;
-				case 128:  ArgIns.Twiddles = R2_Twiddles_fix_128;  ArgIns.SwapLUT = R2_SwapTable_fix_128;  Q = 8; break;
-				case 256:  ArgIns.Twiddles = R4_Twiddles_fix_256;  ArgIns.SwapLUT = R4_SwapTable_fix_256;  Q = 8; break;
-				case 512:  ArgIns.Twiddles = R2_Twiddles_fix_512;  ArgIns.SwapLUT = R2_SwapTable_fix_512;  Q = 6; break;
-				case 1024: ArgIns.Twiddles = R4_Twiddles_fix_1024; ArgIns.SwapLUT = R4_SwapTable_fix_1024; Q = 6; break;
-				case 2048: ArgIns.Twiddles = R2_Twiddles_fix_2048; ArgIns.SwapLUT = R2_SwapTable_fix_2048; Q = 4; break;
+				case 128:  ArgIns.Twiddles = R2_Twiddles_fix_128;  ArgIns.SwapLUT = R2_SwapTable_fix_128;  Q = 7; break;
+				case 256:  ArgIns.Twiddles = R4_Twiddles_fix_256;  ArgIns.SwapLUT = R4_SwapTable_fix_256;  Q = 6; break;
+				case 512:  ArgIns.Twiddles = R2_Twiddles_fix_512;  ArgIns.SwapLUT = R2_SwapTable_fix_512;  Q = 5; break;
+				case 1024: ArgIns.Twiddles = R4_Twiddles_fix_1024; ArgIns.SwapLUT = R4_SwapTable_fix_1024; Q = 4; break;
+				case 2048: ArgIns.Twiddles = R2_Twiddles_fix_2048; ArgIns.SwapLUT = R2_SwapTable_fix_2048; Q = 3; break;
 			}
 			if (ArgIns.Radix == 2) FFTFun = &Radix2FFT_DIF_Par_Fix16;
 			else 				   FFTFun = &Radix4FFT_DIF_Par_Fix16;
@@ -139,24 +146,32 @@ void CallFFT(int Nfft, int Type){
   	__CALL((*FFTFun), &FFTArg);
   	AT_FORK(gap_ncore(), (void *) (*SwapFun), (void *) &SwapArg);
   	__CALL((*SwapFun), &SwapArg);
-  	elapsed = gap_cl_readhwtimer() - start; printf("|     %4d | %3s %6s | %6d | %5d | %6d", Nfft, FFTDataType, ArgIns.Radix==2?"Radix2":"Radix4", elapsedFFT, elapsed, elapsed+elapsedFFT);
+  	elapsed = gap_cl_readhwtimer() - start;
+
+  	PERF_ARR[Nfft/128][Type][0] = elapsedFFT;
+  	PERF_ARR[Nfft/128][Type][1] = elapsed;
+
+
+  	PRINTF("|     %4d | %3s %6s | %6d | %5d | %6d", Nfft, FFTDataType, ArgIns.Radix==2?"Radix2":"Radix4", elapsedFFT, elapsed, elapsed+elapsedFFT);
 	#if !defined(__EMUL__) && defined(PERF_ALL)
-  	printf(" | %7d | %7d | %7d | %8d | %7d |", pi_perf_read(PI_PERF_INSTR), pi_perf_read(PI_PERF_ACTIVE_CYCLES), pi_perf_read(PI_PERF_TCDM_CONT), pi_perf_read(PI_PERF_LD_STALL), pi_perf_read(PI_PERF_IMISS));
+  	PRINTF(" | %7d | %7d | %7d | %8d | %7d |", pi_perf_read(PI_PERF_INSTR), pi_perf_read(PI_PERF_ACTIVE_CYCLES), pi_perf_read(PI_PERF_TCDM_CONT), pi_perf_read(PI_PERF_LD_STALL), pi_perf_read(PI_PERF_IMISS));
   	#else
-  	printf(" |         |         |         |          |         |");
+  	PRINTF(" |         |         |         |          |         |");
   	#endif
   	if (Type == 0) {
-  		printf("          |\n");
-  		// printf("\nOutFFT%d_f32 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) printf("%f%+fj, ", InBuff_f32[2*i], InBuff_f32[2*i+1]); printf("])\n");
+  		PRINTF("          |\n");
+  		// PRINTF("\nOutFFT%d_f32 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) PRINTF("%f%+fj, ", InBuff_f32[2*i], InBuff_f32[2*i+1]); PRINTF("])\n");
   	} else if (Type == 1) {
-  		// printf("\nOutFFT%d_q16 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) printf("%d%+dj, ", ((short int*)InBuff_q16)[2*i], ((short int*)InBuff_q16)[2*i+1]); printf("])\n");
-  		printf(" %f |\n", MSE_16(InBuff_f32, (short int*) InBuff_q16, Nfft, Q));
+  		// PRINTF("\nOutFFT%d_q16 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) PRINTF("%d%+dj, ", ((short int*)InBuff_q16)[2*i], ((short int*)InBuff_q16)[2*i+1]); PRINTF("])\n");
+  		MSE_ARR[Nfft/128][0] = MSE_16(InBuff_f32, (short int*) InBuff_q16, Nfft, Q);
+  		PRINTF(" %f |\n", MSE_ARR[Nfft/128][0]);
   	} else if (Type == 2) {
   		#ifdef __gap9__
-  		// printf("\nOutFFT%d_f16 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) printf("%f%+fj, ", ((f16*)OutBuff)[2*i], ((f16*)OutBuff)[2*i+1]); printf("])\n");
-  		printf(" %f |\n", MSE_f16(InBuff_f32, (f16 *) InBuff_f16, Nfft));
+  		// PRINTF("\nOutFFT%d_f16 = np.array([\n", Nfft); for(int i=0;i<(Nfft); i++) PRINTF("%f%+fj, ", ((f16*)OutBuff)[2*i], ((f16*)OutBuff)[2*i+1]); PRINTF("])\n");
+  		MSE_ARR[Nfft/128][1] = MSE_f16(InBuff_f32, (f16 *) InBuff_f16, Nfft);
+  		PRINTF(" %f |\n", MSE_ARR[Nfft/128][1]);
   		#else
-  		printf("\n");
+  		PRINTF("\n");
   		#endif
   	}
 }
@@ -171,31 +186,31 @@ static void RunFFT()
     #endif
     gap_cl_resethwtimer();
     int	start, elapsed, timef32;
-    printf("Initializing inputs....\n");
+    PRINTF("Initializing inputs....\n");
 	//InitData4      (InDataQ16, MAXDIM, 37, 15, 23, 73, 0.1, 0.5, 0.6, 0.8);
 	//InitData4_float(InDataf32, MAXDIM, 37, 15, 23, 73, 0.1, 0.5, 0.6, 0.8);
 	#ifdef __gap9__
 	for (int i=0; i<MAXDIM; i++) InDataf16[i] = (f16) InDataf32[i];
 	#endif
-	printf("Done!\n");
+	PRINTF("Done!\n");
 
   	gap_cl_resethwtimer();
 
   	int FFTBins = 64;
-  	printf("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
-  	printf("| FFT BINS | Type       | FFT    | Swap  | Tot    | Instr   | Act Cyc | TCDM Co | LD Stall | Imiss   | MSE Err  |\n");
-  	printf("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
+  	PRINTF("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
+  	PRINTF("| FFT BINS | Type       | FFT    | Swap  | Tot    | Instr   | Act Cyc | TCDM Co | LD Stall | Imiss   | MSE Err  |\n");
+  	PRINTF("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
 	while (FFTBins < MAXDIM){
-		//printf("FFT: %4d\n", FFTBins);
+		//PRINTF("FFT: %4d\n", FFTBins);
 	  	CallFFT(FFTBins, 0);
 	  	CallFFT(FFTBins, 1);
 	  	#ifdef __gap9__
 	  	CallFFT(FFTBins, 2);
 	  	#endif
 	  	FFTBins *= 2;
-  		printf("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
+  		PRINTF("|----------+------------+--------+-------+--------+---------+---------+---------+----------+---------+----------|\n");
   	}
-  	printf("Finished\n");
+  	PRINTF("Finished\n");
 }
 
 void test_kickoff(void *arg)
@@ -231,14 +246,31 @@ void test_kickoff(void *arg)
 	#ifdef __EMUL__
 		RunFFT();
 	#else
-		struct pi_cluster_task task = {0};
-		task.entry = RunFFT;
-		task.arg = NULL;
+		struct pi_cluster_task task;
+		pi_cluster_task(&task, RunFFT, NULL);
 		task.stack_size = (unsigned int) STACK_SIZE;
 		task.slave_stack_size = (unsigned int) 1048;
-		pi_cluster_send_task_to_cl(&cluster_dev, &task);
+		pi_cluster_send_task(&cluster_dev, &task);
 	#endif
-	printf("Exiting\n");
+
+  	int FFTBins = 64;
+	while (FFTBins < MAXDIM){
+		if (MSE_ARR[FFTBins/128][0] > 0.016) {
+			printf("Error: MSE too large for %d FFT Q16\n", FFTBins);
+			printf("Test FAILED\n");
+			pmsis_exit(-1);
+		}
+		#ifdef __gap9__
+		if (MSE_ARR[FFTBins/128][1] > 0.000048) {
+			printf("Error: MSE too large for %d FFT F16\n", FFTBins);
+			printf("Test FAILED\n");
+			pmsis_exit(-1);
+		}
+		#endif
+	  	FFTBins *= 2;
+	}
+
+	printf("Test PASSED\n");
     pmsis_exit(0);
 }
 

examples/autotiler/FFTL1/Makefile

@@ -1,7 +1,7 @@
 # User Test
 #------------------------------------
 
-PMSIS_OS?=pulpos
+#PMSIS_OS?=pulpos
 APP              = test
 APP_SRCS        += FFTRunTest.c $(AT_HOME)/DSP_Libraries/FFT_Library.c $(AT_HOME)/DSP_Libraries/LUT_Tables/TwiddlesDef.c $(AT_HOME)/DSP_Libraries/LUT_Tables/SwapTablesDef.c 
 APP_INC	        += 

examples/autotiler/Fir/main.c

@@ -194,12 +194,10 @@ void test_fir()
 
 	printf ("Call cluster\n");
 	struct pi_cluster_task *task = pmsis_l2_malloc(sizeof(struct pi_cluster_task));
-    memset(task, 0, sizeof(struct pi_cluster_task));
-    task->entry = cluster_main;
-    task->arg = (void *) NULL;
+	pi_cluster_task(task, cluster_main, NULL);
     task->stack_size = (uint32_t) STACK_SIZE;
 
-    pi_cluster_send_task_to_cl(&cluster_dev, task);
+    pi_cluster_send_task(&cluster_dev, task);
 
 	pi_cluster_close(&cluster_dev);
 

examples/autotiler/IRFFT2D/FFTRunTest.c

@@ -102,11 +102,10 @@ void test_kickoff(void *arg)
 	#ifdef __EMUL__
 		RunFFT();
 	#else
-		struct pi_cluster_task task = {0};
-		task.entry = RunFFT;
-		task.arg = NULL;
+		struct pi_cluster_task task;
+		pi_cluster_task(&task, RunFFT, NULL);
 		task.stack_size = (unsigned int) STACK_SIZE;
-		pi_cluster_send_task_to_cl(&cluster_dev, &task);
+		pi_cluster_send_task(&cluster_dev, &task);
 	#endif
 
 

examples/autotiler/IntegralImage/main.c

@@ -126,7 +126,7 @@ void integral_image(int argc, char *argv[])
 
 	pi_cluster_task(task, (void (*)(void *))cluster_main, (void *) &ClusterCall);
 
-    pi_cluster_send_task_to_cl(&cluster_dev, task);
+    pi_cluster_send_task(&cluster_dev, task);
 
 	//Enable Debug to print to stdout the result image
 	if(DEBUG){

examples/autotiler/MatMult/MatMult.c

@@ -107,13 +107,12 @@ void run_MatMult(void)
     }
 
     /* Prepare cluster task and send it to cluster. */
-    struct pi_cluster_task task = {0};
-    task.entry = cluster_main;
-    task.arg = NULL;
+    struct pi_cluster_task task;
+    pi_cluster_task(&task, cluster_main, NULL);
     task.stack_size = (uint32_t) STACK_SIZE;
 
     /* Offloading Task to cluster. */
-    pi_cluster_send_task_to_cl(&cluster_dev, &task);
+    pi_cluster_send_task(&cluster_dev, &task);
 
     pi_l1_free(&cluster_dev, L1_Memory, _L1_Memory_SIZE);