Ready!

software/source/bfloat16nnlib.c

@@ -50,7 +50,7 @@ extern int key_eval(void);
 #define DRAMDATASIZE2 DRAMDATASIZE1 
 
 static uint32_t fpgastate, fpustates;
-static int fpgaload(uint32_t *mempt1, uint32_t *mempt2, int16_t calclen, int bReload1, int bReload2)
+static int fpgaload(uint32_t *mempt1, uint32_t *mempt2, uint16_t calclen, int bReload1, int bReload2)
 {
 	uint32_t *sentinel1 = (uint32_t *)(DRAMDATABASE1 + ((DRAMDATASIZE1-1) * sizeof(uint32_t)));
 	uint32_t *sentinel2 = (uint32_t *)(DRAMDATABASE2 + ((DRAMDATASIZE2-1) * sizeof(uint32_t)));	
@@ -62,7 +62,7 @@ static int fpgaload(uint32_t *mempt1, uint32_t *mempt2, int16_t calclen, int bRe
 	bfloat16nn_b32Sentinel1_write(*sentinel1);
 	bfloat16nn_b32Sentinel2_write(*sentinel2);
 	flush_l2_cache(); // Strictly nec. for longer transfers	
-	bfloat16nn_b10ArrayWordLen_write(calclen); // Indicate array length for calc.												
+	bfloat16nn_b10ArrayWordLen_write(calclen>>1); // Indicate array length for calc. but honour split!
 	bfloat16nn_b32DRAMLoadAddress1_write((uint32_t)mempt1); // Indicate memory to load from	
 	bfloat16nn_b32DRAMLoadAddress2_write((uint32_t)mempt2); // Indicate memory to load from	
 	bfloat16nn_bReload1_write(bReload1 ? 1 : 0); // Reload mem#1
@@ -113,14 +113,14 @@ void dumpfloat(float f)
 int key_eval(void)
 {
 	extern void printf1(const char *fmt, float f1);	
-	uint32_t *ui32ptr1, *ui32ptr2; 	
+	//uint32_t *ui32ptr1, *ui32ptr2; 	
 	uint16_t *ui16ptr1, *ui16ptr2;	
 	float fpResult1, fpResult2;	
 	uint32_t starttime;
 	uint32_t deltatime;
 	int i;		
 
-#define MAXCALCLEN 4 //784 //16 OK
+#define MAXCALCLEN 784 //784 //16 OK
 
 	switch(kbhit()) {
 		case 'r': // Reload
@@ -156,32 +156,22 @@ int key_eval(void)
 				printf("S/W Delta t: %dms ", deltatime);				
 				printf1("\t\t\tS/W SUM=%8.4f\n", sum);				
 				
-				// If not cleared fails???
-				for(i=0, ui32ptr1 = (uint32_t *)DRAMDATABASE1;i<DRAMDATASIZE1;i++) // Setup test data
-					*ui32ptr1++ = 0;	// Clear all memory ...
-				for(i=0, ui32ptr2 = (uint32_t *)DRAMDATABASE2;i<DRAMDATASIZE2;i++) // Setup test data
-					*ui32ptr2++ = 0;	// Clear all memory ...
-
 				ui16ptr1 = (uint16_t *)DRAMDATABASE1; // Absolute: bytes! Matrice/row
-				for(i=1;i<=MAXCALCLEN+1;i++)
+				for(i=1;i<=MAXCALCLEN;i++)
 					*ui16ptr1++ = f2ui16(1.0 * (float)i );					
 				ui16ptr2 = (uint16_t *)DRAMDATABASE2; // Absolute: bytes! Vector
-				for(i=1;i<=MAXCALCLEN+1;i++)
+				for(i=1;i<=MAXCALCLEN;i++)
 					*ui16ptr2++ = f2ui16(1.0 * (float)i );				
-
-				// BOTH:  1*1+2*2+3*3+4*4 (+5*5) = 1+4+9+16+25 = 5+9+16+25 = 14+16=30+25 = 55 WRONG!
-				// FPU#1: 1*1    +3*3     = 1+9  = 10 OK
-				// FPU#2:     2*2    +4*4  (+5*5)= 4+16(+5*5) = 20(+25) = 45 WRONG!
-
+				
 				starttime = systime(0);
-				if(fpgaload((uint32_t *)DRAMDATABASE1, (uint32_t *)DRAMDATABASE2, MAXCALCLEN, 1, 1)) { // 800*32-bit=3200 bytes, 400 Words/FPU to calc.
+				if(fpgaload((uint32_t *)DRAMDATABASE1, (uint32_t *)DRAMDATABASE2, MAXCALCLEN, 1, 1)) {
 					deltatime = systime(0)-starttime;
 					printf("H/W Delta t: %dms ", deltatime);				
 					fpResult1 = fpResult1_read();
 					fpResult2 = fpResult2_read();
 					printf("(S=%04Xh: FS=%04Xh)", fpgastate, fpustates);
-					printf1("\tH/W SUM=%8.4f", fpResult1);
-					printf1("\t(FPU#2=%8.4f)\n", fpResult2);
+					printf1("\tH/W SUM=%8.4f\n", fpResult1);
+					//printf1("\t(FPU#2=%8.4f)\n", fpResult2);
 				}						
 				else {					
 					printf("CURRENT TIMEOUT: S=%04Xh: FS=%04Xh ", fpgastate, fpustates);