diff --git a/Appl/CEP/CS1/CS1_BGLProc/src/BGL_Processing.cc b/Appl/CEP/CS1/CS1_BGLProc/src/BGL_Processing.cc
index 04c23602d594a733a1676213340343607cd7624c..7733d158cf71a1023c724e94c39568e697824045 100644
--- a/Appl/CEP/CS1/CS1_BGLProc/src/BGL_Processing.cc
+++ b/Appl/CEP/CS1/CS1_BGLProc/src/BGL_Processing.cc
@@ -191,7 +191,7 @@ void BGL_Processing::checkConsistency(CS1_Parset *parset) const
#if !defined C_IMPLEMENTATION
ASSERT(parset->BGLintegrationSteps() % 16 == 0);
- ASSERT(_FIR_constants_used.input_type == INPUT_TYPE);
+ ASSERT(_FIR_constants_used.nr_bits_per_sample == NR_BITS_PER_SAMPLE);
ASSERT(_FIR_constants_used.nr_subband_channels == NR_SUBBAND_CHANNELS);
ASSERT(_FIR_constants_used.nr_taps == NR_TAPS);
ASSERT(_FIR_constants_used.nr_polarizations == NR_POLARIZATIONS);
diff --git a/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.S b/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.S
index 20283314686229d825b3e841d29a6c912979ab92..a99f849c02a2ef54db7be7dd9c1af063307c530f 100644
--- a/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.S
+++ b/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.S
@@ -22,13 +22,15 @@
#include <CS1_Interface/CS1_Config.h>
#define I4COMPLEX_SIZE 1
+#define I8COMPLEX_SIZE 2
#define I16COMPLEX_SIZE 4
#define FCOMPLEX_SIZE 8
#define DCOMPLEX_SIZE 16
+#define FLOAT_SIZE 4
.global _FIR_constants_used
_FIR_constants_used:
- .long INPUT_TYPE
+ .long NR_BITS_PER_SAMPLE
.long NR_SUBBAND_CHANNELS
.long NR_TAPS
.long NR_POLARIZATIONS
@@ -208,266 +210,10 @@ L5:
blr
-#if INPUT_TYPE == I4COMPLEX_TYPE
+#if NR_BITS_PER_SAMPLE == 4
.align 5
-fp_table:
- .long 0x00000000, 0x00000000
- .long 0x3F800000, 0x00000000
- .long 0x40000000, 0x00000000
- .long 0x40400000, 0x00000000
- .long 0x40800000, 0x00000000
- .long 0x40A00000, 0x00000000
- .long 0x40C00000, 0x00000000
- .long 0x40E00000, 0x00000000
- .long 0xC1000000, 0x00000000
- .long 0xC0E00000, 0x00000000
- .long 0xC0C00000, 0x00000000
- .long 0xC0A00000, 0x00000000
- .long 0xC0800000, 0x00000000
- .long 0xC0400000, 0x00000000
- .long 0xC0000000, 0x00000000
- .long 0xBF800000, 0x00000000
- .long 0x00000000, 0x3F800000
- .long 0x3F800000, 0x3F800000
- .long 0x40000000, 0x3F800000
- .long 0x40400000, 0x3F800000
- .long 0x40800000, 0x3F800000
- .long 0x40A00000, 0x3F800000
- .long 0x40C00000, 0x3F800000
- .long 0x40E00000, 0x3F800000
- .long 0xC1000000, 0x3F800000
- .long 0xC0E00000, 0x3F800000
- .long 0xC0C00000, 0x3F800000
- .long 0xC0A00000, 0x3F800000
- .long 0xC0800000, 0x3F800000
- .long 0xC0400000, 0x3F800000
- .long 0xC0000000, 0x3F800000
- .long 0xBF800000, 0x3F800000
- .long 0x00000000, 0x40000000
- .long 0x3F800000, 0x40000000
- .long 0x40000000, 0x40000000
- .long 0x40400000, 0x40000000
- .long 0x40800000, 0x40000000
- .long 0x40A00000, 0x40000000
- .long 0x40C00000, 0x40000000
- .long 0x40E00000, 0x40000000
- .long 0xC1000000, 0x40000000
- .long 0xC0E00000, 0x40000000
- .long 0xC0C00000, 0x40000000
- .long 0xC0A00000, 0x40000000
- .long 0xC0800000, 0x40000000
- .long 0xC0400000, 0x40000000
- .long 0xC0000000, 0x40000000
- .long 0xBF800000, 0x40000000
- .long 0x00000000, 0x40400000
- .long 0x3F800000, 0x40400000
- .long 0x40000000, 0x40400000
- .long 0x40400000, 0x40400000
- .long 0x40800000, 0x40400000
- .long 0x40A00000, 0x40400000
- .long 0x40C00000, 0x40400000
- .long 0x40E00000, 0x40400000
- .long 0xC1000000, 0x40400000
- .long 0xC0E00000, 0x40400000
- .long 0xC0C00000, 0x40400000
- .long 0xC0A00000, 0x40400000
- .long 0xC0800000, 0x40400000
- .long 0xC0400000, 0x40400000
- .long 0xC0000000, 0x40400000
- .long 0xBF800000, 0x40400000
- .long 0x00000000, 0x40800000
- .long 0x3F800000, 0x40800000
- .long 0x40000000, 0x40800000
- .long 0x40400000, 0x40800000
- .long 0x40800000, 0x40800000
- .long 0x40A00000, 0x40800000
- .long 0x40C00000, 0x40800000
- .long 0x40E00000, 0x40800000
- .long 0xC1000000, 0x40800000
- .long 0xC0E00000, 0x40800000
- .long 0xC0C00000, 0x40800000
- .long 0xC0A00000, 0x40800000
- .long 0xC0800000, 0x40800000
- .long 0xC0400000, 0x40800000
- .long 0xC0000000, 0x40800000
- .long 0xBF800000, 0x40800000
- .long 0x00000000, 0x40A00000
- .long 0x3F800000, 0x40A00000
- .long 0x40000000, 0x40A00000
- .long 0x40400000, 0x40A00000
- .long 0x40800000, 0x40A00000
- .long 0x40A00000, 0x40A00000
- .long 0x40C00000, 0x40A00000
- .long 0x40E00000, 0x40A00000
- .long 0xC1000000, 0x40A00000
- .long 0xC0E00000, 0x40A00000
- .long 0xC0C00000, 0x40A00000
- .long 0xC0A00000, 0x40A00000
- .long 0xC0800000, 0x40A00000
- .long 0xC0400000, 0x40A00000
- .long 0xC0000000, 0x40A00000
- .long 0xBF800000, 0x40A00000
- .long 0x00000000, 0x40C00000
- .long 0x3F800000, 0x40C00000
- .long 0x40000000, 0x40C00000
- .long 0x40400000, 0x40C00000
- .long 0x40800000, 0x40C00000
- .long 0x40A00000, 0x40C00000
- .long 0x40C00000, 0x40C00000
- .long 0x40E00000, 0x40C00000
- .long 0xC1000000, 0x40C00000
- .long 0xC0E00000, 0x40C00000
- .long 0xC0C00000, 0x40C00000
- .long 0xC0A00000, 0x40C00000
- .long 0xC0800000, 0x40C00000
- .long 0xC0400000, 0x40C00000
- .long 0xC0000000, 0x40C00000
- .long 0xBF800000, 0x40C00000
- .long 0x00000000, 0x40E00000
- .long 0x3F800000, 0x40E00000
- .long 0x40000000, 0x40E00000
- .long 0x40400000, 0x40E00000
- .long 0x40800000, 0x40E00000
- .long 0x40A00000, 0x40E00000
- .long 0x40C00000, 0x40E00000
- .long 0x40E00000, 0x40E00000
- .long 0xC1000000, 0x40E00000
- .long 0xC0E00000, 0x40E00000
- .long 0xC0C00000, 0x40E00000
- .long 0xC0A00000, 0x40E00000
- .long 0xC0800000, 0x40E00000
- .long 0xC0400000, 0x40E00000
- .long 0xC0000000, 0x40E00000
- .long 0xBF800000, 0x40E00000
- .long 0x00000000, 0xC1000000
- .long 0x3F800000, 0xC1000000
- .long 0x40000000, 0xC1000000
- .long 0x40400000, 0xC1000000
- .long 0x40800000, 0xC1000000
- .long 0x40A00000, 0xC1000000
- .long 0x40C00000, 0xC1000000
- .long 0x40E00000, 0xC1000000
- .long 0xC1000000, 0xC1000000
- .long 0xC0E00000, 0xC1000000
- .long 0xC0C00000, 0xC1000000
- .long 0xC0A00000, 0xC1000000
- .long 0xC0800000, 0xC1000000
- .long 0xC0400000, 0xC1000000
- .long 0xC0000000, 0xC1000000
- .long 0xBF800000, 0xC1000000
- .long 0x00000000, 0xC0E00000
- .long 0x3F800000, 0xC0E00000
- .long 0x40000000, 0xC0E00000
- .long 0x40400000, 0xC0E00000
- .long 0x40800000, 0xC0E00000
- .long 0x40A00000, 0xC0E00000
- .long 0x40C00000, 0xC0E00000
- .long 0x40E00000, 0xC0E00000
- .long 0xC1000000, 0xC0E00000
- .long 0xC0E00000, 0xC0E00000
- .long 0xC0C00000, 0xC0E00000
- .long 0xC0A00000, 0xC0E00000
- .long 0xC0800000, 0xC0E00000
- .long 0xC0400000, 0xC0E00000
- .long 0xC0000000, 0xC0E00000
- .long 0xBF800000, 0xC0E00000
- .long 0x00000000, 0xC0C00000
- .long 0x3F800000, 0xC0C00000
- .long 0x40000000, 0xC0C00000
- .long 0x40400000, 0xC0C00000
- .long 0x40800000, 0xC0C00000
- .long 0x40A00000, 0xC0C00000
- .long 0x40C00000, 0xC0C00000
- .long 0x40E00000, 0xC0C00000
- .long 0xC1000000, 0xC0C00000
- .long 0xC0E00000, 0xC0C00000
- .long 0xC0C00000, 0xC0C00000
- .long 0xC0A00000, 0xC0C00000
- .long 0xC0800000, 0xC0C00000
- .long 0xC0400000, 0xC0C00000
- .long 0xC0000000, 0xC0C00000
- .long 0xBF800000, 0xC0C00000
- .long 0x00000000, 0xC0A00000
- .long 0x3F800000, 0xC0A00000
- .long 0x40000000, 0xC0A00000
- .long 0x40400000, 0xC0A00000
- .long 0x40800000, 0xC0A00000
- .long 0x40A00000, 0xC0A00000
- .long 0x40C00000, 0xC0A00000
- .long 0x40E00000, 0xC0A00000
- .long 0xC1000000, 0xC0A00000
- .long 0xC0E00000, 0xC0A00000
- .long 0xC0C00000, 0xC0A00000
- .long 0xC0A00000, 0xC0A00000
- .long 0xC0800000, 0xC0A00000
- .long 0xC0400000, 0xC0A00000
- .long 0xC0000000, 0xC0A00000
- .long 0xBF800000, 0xC0A00000
- .long 0x00000000, 0xC0800000
- .long 0x3F800000, 0xC0800000
- .long 0x40000000, 0xC0800000
- .long 0x40400000, 0xC0800000
- .long 0x40800000, 0xC0800000
- .long 0x40A00000, 0xC0800000
- .long 0x40C00000, 0xC0800000
- .long 0x40E00000, 0xC0800000
- .long 0xC1000000, 0xC0800000
- .long 0xC0E00000, 0xC0800000
- .long 0xC0C00000, 0xC0800000
- .long 0xC0A00000, 0xC0800000
- .long 0xC0800000, 0xC0800000
- .long 0xC0400000, 0xC0800000
- .long 0xC0000000, 0xC0800000
- .long 0xBF800000, 0xC0800000
- .long 0x00000000, 0xC0400000
- .long 0x3F800000, 0xC0400000
- .long 0x40000000, 0xC0400000
- .long 0x40400000, 0xC0400000
- .long 0x40800000, 0xC0400000
- .long 0x40A00000, 0xC0400000
- .long 0x40C00000, 0xC0400000
- .long 0x40E00000, 0xC0400000
- .long 0xC1000000, 0xC0400000
- .long 0xC0E00000, 0xC0400000
- .long 0xC0C00000, 0xC0400000
- .long 0xC0A00000, 0xC0400000
- .long 0xC0800000, 0xC0400000
- .long 0xC0400000, 0xC0400000
- .long 0xC0000000, 0xC0400000
- .long 0xBF800000, 0xC0400000
- .long 0x00000000, 0xC0000000
- .long 0x3F800000, 0xC0000000
- .long 0x40000000, 0xC0000000
- .long 0x40400000, 0xC0000000
- .long 0x40800000, 0xC0000000
- .long 0x40A00000, 0xC0000000
- .long 0x40C00000, 0xC0000000
- .long 0x40E00000, 0xC0000000
- .long 0xC1000000, 0xC0000000
- .long 0xC0E00000, 0xC0000000
- .long 0xC0C00000, 0xC0000000
- .long 0xC0A00000, 0xC0000000
- .long 0xC0800000, 0xC0000000
- .long 0xC0400000, 0xC0000000
- .long 0xC0000000, 0xC0000000
- .long 0xBF800000, 0xC0000000
- .long 0x00000000, 0xBF800000
- .long 0x3F800000, 0xBF800000
- .long 0x40000000, 0xBF800000
- .long 0x40400000, 0xBF800000
- .long 0x40800000, 0xBF800000
- .long 0x40A00000, 0xBF800000
- .long 0x40C00000, 0xBF800000
- .long 0x40E00000, 0xBF800000
- .long 0xC1000000, 0xBF800000
- .long 0xC0E00000, 0xBF800000
- .long 0xC0C00000, 0xBF800000
- .long 0xC0A00000, 0xBF800000
- .long 0xC0800000, 0xBF800000
- .long 0xC0400000, 0xBF800000
- .long 0xC0000000, 0xBF800000
- .long 0xBF800000, 0xBF800000
+.comm _FIR_fp_table,16*16*FCOMPLEX_SIZE,8
.global _filter
_filter:
@@ -484,7 +230,7 @@ _filter:
# internally used:
# r9 : 8
# r10 : 512
-# r11 : pointer to fp_table
+# r11 : pointer to _FIR_fp_table
# r28-r31 : prefetched samples
# f0-f15 : delay line (real in primary, imaginary in secondary unit)
# f16-f23 : weights (these are real values alternately stored in primary
@@ -523,8 +269,8 @@ _filter:
# convert 15 i4complex numbers to fcomplex
li 10,I4COMPLEX_SIZE*NR_SUBBAND_CHANNELS*NR_POLARIZATIONS
; lbzx 29,0,5 # fetch FIR history samples
- lis 11,fp_table@ha ; lbzux 30,5,10
- la 11,fp_table@l(11);lbzux 31,5,10
+ lis 11,_FIR_fp_table@ha ; lbzux 30,5,10
+ la 11,_FIR_fp_table@l(11);lbzux 31,5,10
slwi 29,29,3 ; lbzux 28,5,10
; lfpsx 1,11,29
slwi 30,30,3 ; lbzux 29,5,10
@@ -571,19 +317,54 @@ _filter:
loop:
# time steps 0-5
- fxcpmadd 24,17,14,24
- fxcsmadd 25,17,14,25
- fxcpmadd 26,18,14,26
- fxcsmadd 27,18,14,27
- fxcpmadd 28,19,14,28
- fxcsmadd 29,19,14,29
+ fxcsmadd 24,23,1,24 ; slwi 28,28,3
+ fxcpmadd 25,20,9,25 ; lfpsx 0,11,28
+ fxcsmadd 26,20,9,26
+ fxcpmadd 27,21,9,27
+ fxcsmadd 28,21,9,28
+ fxcpmadd 29,22,9,29 ; lbzux 28,5,10
+
+ fxcpmadd 24,23,2,24
+ fxcsmadd 25,23,2,25
+ fxcpmadd 26,20,10,26
+ fxcsmadd 27,20,10,27
+ fxcpmadd 28,21,10,28
+ fxcsmadd 29,21,10,29
+
+ fxcsmadd 24,22,3,24 ; slwi 29,29,3
+ fxcpmadd 25,23,3,25 ; lfpsx 1,11,29
+ fxcsmadd 26,23,3,26
+ fxcpmadd 27,20,11,27
+ fxcsmadd 28,20,11,28
+ fxcpmadd 29,21,11,29 ; lbzux 29,5,10
+
+ fxcpmadd 24,22,4,24
+ fxcsmadd 25,22,4,25
+ fxcpmadd 26,23,4,26
+ fxcsmadd 27,23,4,27
+ fxcpmadd 28,20,12,28
+ fxcsmadd 29,20,12,29
- fxcpmadd 24,21,6,24 ; slwi 28,28,3
- fxcsmadd 25,21,6,25 ; lfpsx 0,11,28
- fxcpmadd 26,22,6,26 ; lbzux 28,5,10
+ fxcsmadd 24,21,5,24 ; slwi 30,30,3
+ fxcpmadd 25,22,5,25 ; lfpsx 2,11,30
+ fxcsmadd 26,22,5,26
+ fxcpmadd 27,23,5,27
+ fxcsmadd 28,23,5,28
+ fxcpmadd 29,20,13,29 ; lbzux 30,5,10
+
+ fxcpmadd 24,21,6,24
+ fxcsmadd 25,21,6,25
+ fxcpmadd 26,22,6,26
fxcsmadd 27,22,6,27
fxcpmadd 28,23,6,28
fxcsmadd 29,23,6,29
+
+ fxcsmadd 24,20,7,24 ; slwi 31,31,3
+ fxcpmadd 25,21,7,25 ; lfpsx 3,11,31
+ fxcsmadd 26,21,7,26
+ fxcpmadd 27,22,7,27
+ fxcsmadd 28,22,7,28
+ fxcpmadd 29,23,7,29 ; lbzux 31,5,10
fxcpmadd 24,20,8,24
fxcsmadd 25,20,8,25
@@ -592,40 +373,26 @@ loop:
fxcpmadd 28,22,8,28
fxcsmadd 29,22,8,29
+ fxcpmadd 24,17,14,24 ; slwi 28,28,3
+ fxcsmadd 25,17,14,25 ; lfpsx 4,11,28
+ fxcpmadd 26,18,14,26
+ fxcsmadd 27,18,14,27
+ fxcpmadd 28,19,14,28
+ fxcsmadd 29,19,14,29 ; lbzux 28,5,10
+
fxcpmadd 24,16,0,24
fxcsmadd 25,16,0,25
fxcpmadd 26,17,0,26
fxcsmadd 27,17,0,27
fxcpmadd 28,18,0,28
- fxcsmadd 29,18,0,29 ; slwi 29,29,3
-
- fxcsmadd 24,23,1,24 ; lfpsx 1,11,29
- fxcpmadd 25,20,9,25 ; lbzux 29,5,10
- fxcsmadd 26,20,9,26
- fxcpmadd 27,21,9,27
- fxcsmadd 28,21,9,28
- fxcpmadd 29,22,9,29
+ fxcsmadd 29,18,0,29
- fxcsmadd 24,19,9,24
- fxcpmadd 25,16,1,25
+ fxcsmadd 24,19,9,24 ; slwi 29,29,3
+ fxcpmadd 25,16,1,25 ; lfpsx 5,11,29
fxcsmadd 26,16,1,26
fxcpmadd 27,17,1,27
fxcsmadd 28,17,1,28
- fxcpmadd 29,18,1,29
-
- fxcpmadd 24,23,2,24 ; slwi 30,30,3
- fxcsmadd 25,23,2,25 ; lfpsx 2,11,30
- fxcpmadd 26,20,10,26 ; lbzux 30,5,10
- fxcsmadd 27,20,10,27
- fxcpmadd 28,21,10,28
- fxcsmadd 29,21,10,29
-
- fxcsmadd 24,20,7,24
- fxcpmadd 25,21,7,25
- fxcsmadd 26,21,7,26
- fxcpmadd 27,22,7,27
- fxcsmadd 28,22,7,28
- fxcpmadd 29,23,7,29
+ fxcpmadd 29,18,1,29 ; lbzux 29,5,10
fxcpmadd 24,19,10,24
fxcsmadd 25,19,10,25
@@ -634,13 +401,6 @@ loop:
fxcpmadd 28,17,2,28
fxcsmadd 29,17,2,29
- fxcsmadd 24,22,3,24
- fxcpmadd 25,23,3,25 ; slwi 31,31,3
- fxcsmadd 26,23,3,26 ; lfpsx 3,11,31
- fxcpmadd 27,20,11,27 ; lbzux 31,5,10
- fxcsmadd 28,20,11,28
- fxcpmadd 29,21,11,29
-
fxcsmadd 24,18,11,24
fxcpmadd 25,19,11,25
fxcsmadd 26,19,11,26
@@ -648,13 +408,6 @@ loop:
fxcsmadd 28,16,3,28
fxcpmadd 29,17,3,29
- fxcpmadd 24,22,4,24
- fxcsmadd 25,22,4,25
- fxcpmadd 26,23,4,26 ; slwi 28,28,3
- fxcsmadd 27,23,4,27 ; lfpsx 4,11,28
- fxcpmadd 28,20,12,28 ; lbzux 28,5,10
- fxcsmadd 29,20,12,29
-
fxcpmadd 24,18,12,24
fxcsmadd 25,18,12,25
fxcpmadd 26,19,12,26
@@ -662,13 +415,6 @@ loop:
fxcpmadd 28,16,4,28
fxcsmadd 29,16,4,29
- fxcsmadd 24,21,5,24
- fxcpmadd 25,22,5,25
- fxcsmadd 26,22,5,26
- fxcpmadd 27,23,5,27 ; slwi 29,29,3
- fxcsmadd 28,23,5,28 ; lfpsx 5,11,29
- fxcpmadd 29,20,13,29 ; lbzux 29,5,10
-
fxcsmadd 24,17,13,24
fxcpmadd 25,18,13,25
fxcsmadd 26,18,13,26
@@ -802,112 +548,1200 @@ loop:
fxcpmadd 27,18,10,27
fxcsmadd 28,18,10,28
- fxcpmadd 24,20,3,24
- fxcsmadd 25,20,3,25
- fxcpmadd 26,21,3,26
- fxcsmadd 27,21,3,27
- fxcpmadd 28,22,3,28
-
- fxcpmadd 24,16,11,24
- fxcsmadd 25,16,11,25
- fxcpmadd 26,17,11,26
- fxcsmadd 27,17,11,27
- fxcpmadd 28,18,11,28 ; slwi 28,28,3
+ fxcpmadd 24,20,3,24
+ fxcsmadd 25,20,3,25
+ fxcpmadd 26,21,3,26
+ fxcsmadd 27,21,3,27
+ fxcpmadd 28,22,3,28
+
+ fxcpmadd 24,16,11,24
+ fxcsmadd 25,16,11,25
+ fxcpmadd 26,17,11,26
+ fxcsmadd 27,17,11,27
+ fxcpmadd 28,18,11,28 ; slwi 28,28,3
+
+ fxcsmadd 24,23,12,24 ; lfpsx 12,11,28
+ fxcpmadd 25,20,4,25 ; lbzux 28,5,10
+ fxcsmadd 26,20,4,26
+ fxcpmadd 27,21,4,27
+ fxcsmadd 28,21,4,28
+
+ fxcsmadd 24,19,4,24
+ fxcpmadd 25,16,12,25
+ fxcsmadd 26,16,12,26
+ fxcpmadd 27,17,12,27
+ fxcsmadd 28,17,12,28
+
+ fxcpmadd 24,23,13,24 ; slwi 29,29,3
+ fxcsmadd 25,23,13,25 ; lfpsx 13,11,29
+ fxcpmadd 26,20,5,26 ; lbzux 29,5,10
+ fxcsmadd 27,20,5,27
+ fxcpmadd 28,21,5,28
+
+ fxcsmadd 24,17,8,24
+ fxcpmadd 25,18,8,25
+ fxcsmadd 26,18,8,26
+ fxcpmadd 27,19,8,27
+ fxcsmadd 28,19,8,28
+
+ fxcpmadd 24,19,5,24
+ fxcsmadd 25,19,5,25
+ fxcpmadd 26,16,13,26
+ fxcsmadd 27,16,13,27
+ fxcpmadd 28,17,13,28
+
+ fxcsmadd 24,22,14,24
+ fxcpmadd 25,23,14,25 ; slwi 30,30,3
+ fxcsmadd 26,23,14,26 ; lfpsx 14,11,30
+ fxcpmadd 27,20,6,27 ; lbzux 30,5,10
+ fxcsmadd 28,20,6,28
+
+ fxcsmadd 24,18,6,24
+ fxcpmadd 25,19,6,25
+ fxcsmadd 26,19,6,26
+ fxcpmadd 27,16,14,27
+ fxcsmadd 28,16,14,28
+
+ fxcpmadd 24,21,1,24
+ fxcsmadd 25,21,1,25
+ fxcpmadd 26,22,1,26
+ fxcsmadd 27,22,1,27
+ fxcpmadd 28,23,1,28
+
+ fxcpmadd 24,22,15,24
+ fxcsmadd 25,22,15,25
+ fxcpmadd 26,23,15,26 ; slwi 31,31,3
+ fxcsmadd 27,23,15,27 ; lfpsx 15,11,31
+ fxcpmadd 28,20,7,28 ; lbzux 31,5,10
+
+ fxcpmadd 24,18,7,24
+ fxcsmadd 25,18,7,25
+ fxcpmadd 26,19,7,26
+ fxcsmadd 27,19,7,27
+ fxcpmadd 28,16,15,28
+
+ fxpmul 29,19,15 ; stfpsux 24,6,9
+ fxsmul 24,16,15 ; stfpsux 25,6,9
+ fxpmul 25,17,15 ; stfpsux 26,6,9
+ fxsmul 26,17,15 ; stfpsux 27,6,9
+ fxpmul 27,18,15 ; stfpsux 28,6,9
+ fxsmul 28,18,15
+
+ bdnz loop
+
+ la 8,16(1) # restore call-saved registers
+ li 9,DCOMPLEX_SIZE
+
+ lfpdx 14,0,8
+ lfpdux 15,8,9
+ lfpdux 16,8,9
+ lfpdux 17,8,9
+ lfpdux 18,8,9
+ lfpdux 19,8,9
+ lfpdux 20,8,9
+ lfpdux 21,8,9
+ lfpdux 22,8,9
+ lfpdux 23,8,9
+ lfpdux 24,8,9
+ lfpdux 25,8,9
+ lfpdux 26,8,9
+ lfpdux 27,8,9
+ lfpdux 28,8,9
+ lfpdux 29,8,9
+ lmw 28,0(1) # restore r28 ... r31
+
+ addi 1,1,272 # restore stack pointer
+ blr # return
+
+#elif NR_BITS_PER_SAMPLE == 8
+
+.align 5
+.comm _FIR_fp_table,256*256*FCOMPLEX_SIZE,8
+
+.global _filter
+_filter:
+# filters all samples for one station, one polarization
+
+# arguments:
+# r3 : pointer to delay line (fcomplex[16]) NOTE: USE OF THE DELAY
+# LINE IS COMMENTED OUT!!!!!!!!!!!!!!!!!!!!
+# r4 : pointer to weights line (const fcomplex[16])
+# r5 : pointer to first sample (const i8complex[16*r7])
+# r6 : pointer to result (fcomplex *)
+# r7 : number of samples / 16
+
+# internally used:
+# r9 : 8
+# r10 : 1024
+# r11 : ptr to constant table
+# r28-r31 : prefetched samples
+# f0-f15 : delay line (real in primary, imaginary in secondary unit)
+# f16-f23 : weights (these are real values alternately stored in primary
+# and secondary units)
+# f24-f29 : sums
+
+# The implementation works on 5 or 6 time samples concurrently, to avoid
+# stalls in the double hummer. This unfortunately leads to totally
+# incomprehensible code. The loop processes 16 samples at a time.
+# The input is converted from int16complex to dcomplex by black magic,
+# making the code even harder to understand.
+
+ mtctr 7 # set number of iterations
+
+ li 9,-DCOMPLEX_SIZE
+ stfpdux 14,1,9 # save call-saved registers
+ stfpdux 15,1,9
+ stfpdux 16,1,9
+ stfpdux 17,1,9
+ stfpdux 18,1,9
+ stfpdux 19,1,9
+ stfpdux 20,1,9
+ stfpdux 21,1,9
+ stfpdux 22,1,9
+ stfpdux 23,1,9
+ stfpdux 24,1,9
+ stfpdux 25,1,9
+ stfpdux 26,1,9
+ stfpdux 27,1,9
+ stfpdux 28,1,9
+ stfpdux 29,1,9
+ stfpdux 30,1,9
+ stfpdux 31,1,9
+
+ subi 1,1,16
+ stmw 28,0(1) # save r28 ... r31
+
+ lis 11,_FIR_fp_table@ha
+ li 9,FCOMPLEX_SIZE
+ la 11,_FIR_fp_table@l(11)
+ li 10,I8COMPLEX_SIZE*NR_SUBBAND_CHANNELS*NR_POLARIZATIONS
+ subi 6,6,FCOMPLEX_SIZE
+
+
+#if 0
+ lfpsx 1,0,3 # load delay line
+ lfpsux 2,3,9
+ lfpsux 3,3,9
+ lfpsux 4,3,9
+ lfpsux 5,3,9
+ lfpsux 6,3,9
+ lfpsux 7,3,9
+ lfpsux 8,3,9
+ lfpsux 9,3,9
+ lfpsux 10,3,9
+ lfpsux 11,3,9
+ lfpsux 12,3,9
+ lfpsux 13,3,9
+ lfpsux 14,3,9
+ lfpsux 15,3,9
+#endif
+
+ # convert 15 i16complex numbers to fcomplex
+ lhzx 29,0,5 # fetch FIR history samples
+ lhzux 30,5,10
+ lhzux 31,5,10
+ lhzux 28,5,10
+
+ rlwinm 12,29,3,13,28
+ lfpsx 1,11,12
+ lhzux 29,5,10
+
+ rlwinm 12,30,3,13,28
+ lfpsx 2,11,12
+ lhzux 30,5,10
+
+ rlwinm 12,31,3,13,28
+ lfpsx 3,11,12
+ lhzux 31,5,10
+
+ rlwinm 12,28,3,13,28
+ lfpsx 4,11,12
+ lhzux 28,5,10
+
+ rlwinm 12,29,3,13,28
+ lfpsx 5,11,12
+ lhzux 29,5,10
+
+ rlwinm 12,30,3,13,28
+ lfpsx 6,11,12
+ lhzux 30,5,10
+
+ rlwinm 12,31,3,13,28
+ lfpsx 7,11,12
+ lhzux 31,5,10
+
+ rlwinm 12,28,3,13,28
+ lfpsx 8,11,12
+ lhzux 28,5,10
+
+ rlwinm 12,29,3,13,28
+ lfpsx 9,11,12
+ lhzux 29,5,10
+
+ rlwinm 12,30,3,13,28
+ lfpsx 10,11,12
+ lhzux 30,5,10
+
+ rlwinm 12,31,3,13,28
+ lfpsx 11,11,12
+ lhzux 31,5,10
+
+ rlwinm 12,28,3,13,28
+ lfpsx 12,11,12
+ lhzux 28,5,10 # prefetch samples before entering loop
+
+ rlwinm 12,29,3,13,28
+ lfpsx 13,11,12
+ lhzux 29,5,10
+
+ rlwinm 12,30,3,13,28
+ lfpsx 14,11,12
+ lhzux 30,5,10
+
+ rlwinm 12,31,3,13,28
+ lfpsx 15,11,12
+ lhzux 31,5,10
+
+ ; lfpsx 16,0,4 # load weights line
+ ; lfpsux 17,4,9
+ ; lfpsux 18,4,9
+ ; lfpsux 19,4,9
+ ; lfpsux 20,4,9
+ ; lfpsux 21,4,9
+ ; lfpsux 22,4,9
+ ; lfpsux 23,4,9
+
+ # essentially part of loop, but interleaved with
+ # the stores in the tail of the loop
+ fxpmul 24,20,8
+ fxsmul 25,20,8
+ fxpmul 26,21,8
+ fxsmul 27,21,8
+ fxpmul 28,22,8
+ fxsmul 29,22,8
+ fxpmul 30,23,8
+ fxsmul 31,23,8
+
+loop:
+ # time steps 0-7
+
+ # load 0
+
+ fxcsmadd 24,23,1,24 ; rlwinm 12,28,3,13,28
+ # load 1
+ fxcpmadd 25,20,9,25 ; lfpsx 0,11,12
+ fxcsmadd 26,20,9,26 ; lhzux 28,5,10
+ fxcpmadd 27,21,9,27
+ fxcsmadd 28,21,9,28
+ fxcpmadd 29,22,9,29
+ fxcsmadd 30,22,9,30
+ fxcpmadd 31,23,9,31
+
+ fxcpmadd 24,23,2,24 ; rlwinm 12,29,3,13,28
+ fxcsmadd 25,23,2,25 ; lfpsx 1,11,12
+ # load 2
+ fxcpmadd 26,20,10,26 ; lhzux 29,5,10
+ fxcsmadd 27,20,10,27
+ fxcpmadd 28,21,10,28
+ fxcsmadd 29,21,10,29
+ fxcpmadd 30,22,10,30
+ fxcsmadd 31,22,10,31
+
+ fxcsmadd 24,22,3,24 ; rlwinm 12,30,3,13,28
+ fxcpmadd 25,23,3,25 ; lfpsx 2,11,12
+ fxcsmadd 26,23,3,26 ; lhzux 30,5,10
+ # load 3
+ fxcpmadd 27,20,11,27
+ fxcsmadd 28,20,11,28
+ fxcpmadd 29,21,11,29
+ fxcsmadd 30,21,11,30
+ fxcpmadd 31,22,11,31
+
+ fxcpmadd 24,22,4,24 ; rlwinm 12,31,3,13,28
+ fxcsmadd 25,22,4,25 ; lfpsx 3,11,12
+ fxcpmadd 26,23,4,26 ; lhzux 31,5,10
+ fxcsmadd 27,23,4,27
+ # load 4
+ fxcpmadd 28,20,12,28
+ fxcsmadd 29,20,12,29
+ fxcpmadd 30,21,12,30
+ fxcsmadd 31,21,12,31
+
+ fxcsmadd 24,21,5,24
+ fxcpmadd 25,22,5,25
+ fxcsmadd 26,22,5,26
+ fxcpmadd 27,23,5,27
+ fxcsmadd 28,23,5,28
+ # load 5
+ fxcpmadd 29,20,13,29
+ fxcsmadd 30,20,13,30
+ fxcpmadd 31,21,13,31
+
+ fxcpmadd 24,21,6,24
+ fxcsmadd 25,21,6,25
+ fxcpmadd 26,22,6,26
+ fxcsmadd 27,22,6,27
+ fxcpmadd 28,23,6,28
+ fxcsmadd 29,23,6,29
+ # load 6
+ fxcpmadd 30,20,14,30
+ fxcsmadd 31,20,14,31
+
+ fxcsmadd 24,20,7,24
+ fxcpmadd 25,21,7,25
+ fxcsmadd 26,21,7,26
+ fxcpmadd 27,22,7,27 ; rlwinm 12,28,3,13,28
+ fxcsmadd 28,22,7,28 ; lfpsx 4,11,12
+ fxcpmadd 29,23,7,29 ; lhzux 28,5,10
+ fxcsmadd 30,23,7,30
+ # load 7
+ fxcpmadd 31,20,15,31
+
+ # after loads
+ fxcpmadd 24,16,0,24
+ fxcsmadd 25,16,0,25
+ fxcpmadd 26,17,0,26
+ fxcsmadd 27,17,0,27
+ fxcpmadd 28,18,0,28
+ fxcsmadd 29,18,0,29
+ fxcpmadd 30,19,0,30
+ fxcsmadd 31,19,0,31 ; rlwinm 12,29,3,13,28
+
+ fxcsmadd 24,19,9,24 ; lfpsx 5,11,12
+ fxcpmadd 25,16,1,25 ; lhzux 29,5,10
+ fxcsmadd 26,16,1,26
+ fxcpmadd 27,17,1,27
+ fxcsmadd 28,17,1,28
+ fxcpmadd 29,18,1,29
+ fxcsmadd 30,18,1,30
+ fxcpmadd 31,19,1,31
+
+ fxcpmadd 24,19,10,24
+ fxcsmadd 25,19,10,25
+ fxcpmadd 26,16,2,26
+ fxcsmadd 27,16,2,27 ; rlwinm 12,30,3,13,28
+ fxcpmadd 28,17,2,28 ; lfpsx 6,11,12
+ fxcsmadd 29,17,2,29 ; lhzux 30,5,10
+ fxcpmadd 30,18,2,30
+ fxcsmadd 31,18,2,31
+
+ fxcsmadd 24,18,11,24
+ fxcpmadd 25,19,11,25
+ fxcsmadd 26,19,11,26
+ fxcpmadd 27,16,3,27
+ fxcsmadd 28,16,3,28
+ fxcpmadd 29,17,3,29
+ fxcsmadd 30,17,3,30
+ fxcpmadd 31,18,3,31 ; rlwinm 12,31,3,13,28
+
+ fxcpmadd 24,18,12,24 ; lfpsx 7,11,12
+ fxcsmadd 25,18,12,25 ; lhzux 31,5,10
+ fxcpmadd 26,19,12,26
+ fxcsmadd 27,19,12,27
+ fxcpmadd 28,16,4,28
+ fxcsmadd 29,16,4,29
+ fxcpmadd 30,17,4,30
+ fxcsmadd 31,17,4,31
+
+ fxcsmadd 24,17,13,24
+ fxcpmadd 25,18,13,25
+ fxcsmadd 26,18,13,26
+ fxcpmadd 27,19,13,27
+ fxcsmadd 28,19,13,28
+ fxcpmadd 29,16,5,29
+ fxcsmadd 30,16,5,30
+ fxcpmadd 31,17,5,31 ; rlwinm 12,28,3,13,28
+
+ fxcpmadd 24,17,14,24 ; lfpsx 8,11,12
+ fxcsmadd 25,17,14,25 ; lhzux 28,5,10
+ fxcpmadd 26,18,14,26
+ fxcsmadd 27,18,14,27
+ fxcpmadd 28,19,14,28
+ fxcsmadd 29,19,14,29
+ fxcpmadd 30,16,6,30
+ fxcsmadd 31,16,6,31
+
+ fxcsmadd 24,16,15,24
+ fxcpmadd 25,17,15,25
+ fxcsmadd 26,17,15,26
+ fxcpmadd 27,18,15,27
+ fxcsmadd 28,18,15,28
+ fxcpmadd 29,19,15,29 ; stfpsux 24,6,9
+ fxcsmadd 30,19,15,30 ; stfpsux 25,6,9
+ fxcpmadd 31,16,7,31 ; stfpsux 26,6,9
+
+
+ # time steps 8-15
+ fxpmul 24,20,0
+ fxsmul 25,20,0
+ fxpmul 26,21,0 ; stfpsux 27,6,9
+ fxsmul 27,21,0 ; stfpsux 28,6,9
+ fxpmul 28,22,0 ; stfpsux 29,6,9
+ fxsmul 29,22,0 ; stfpsux 30,6,9
+ fxpmul 30,23,0 ; stfpsux 31,6,9
+ fxsmul 31,23,0
+
+ # already loaded 8
+
+ fxcsmadd 24,23,9,24
+ # load 9
+ fxcpmadd 25,20,1,25
+ fxcsmadd 26,20,1,26
+ fxcpmadd 27,21,1,27
+ fxcsmadd 28,21,1,28
+ fxcpmadd 29,22,1,29
+ fxcsmadd 30,22,1,30
+ fxcpmadd 31,23,1,31
+
+ fxcpmadd 24,23,10,24
+ fxcsmadd 25,23,10,25
+ # load 10
+ fxcpmadd 26,20,2,26
+ fxcsmadd 27,20,2,27
+ fxcpmadd 28,21,2,28
+ fxcsmadd 29,21,2,29
+ fxcpmadd 30,22,2,30
+ fxcsmadd 31,22,2,31 ; rlwinm 12,29,3,13,28
+
+ fxcsmadd 24,22,11,24 ; lfpsx 9,11,12
+ fxcpmadd 25,23,11,25 ; lhzux 29,5,10
+ fxcsmadd 26,23,11,26
+ # load 11
+ fxcpmadd 27,20,3,27
+ fxcsmadd 28,20,3,28
+ fxcpmadd 29,21,3,29
+ fxcsmadd 30,21,3,30
+ fxcpmadd 31,22,3,31
+
+ fxcpmadd 24,22,12,24
+ fxcsmadd 25,22,12,25
+ fxcpmadd 26,23,12,26
+ fxcsmadd 27,23,12,27 ; rlwinm 12,30,3,13,28
+ # load 12
+ fxcpmadd 28,20,4,28 ; lfpsx 10,11,12
+ fxcsmadd 29,20,4,29 ; lhzux 30,5,10
+ fxcpmadd 30,21,4,30
+ fxcsmadd 31,21,4,31
+
+ fxcsmadd 24,21,13,24
+ fxcpmadd 25,22,13,25
+ fxcsmadd 26,22,13,26
+ fxcpmadd 27,23,13,27
+ fxcsmadd 28,23,13,28
+ # load 13
+ fxcpmadd 29,20,5,29
+ fxcsmadd 30,20,5,30
+ fxcpmadd 31,21,5,31 ; rlwinm 12,31,3,13,28
+
+ fxcpmadd 24,21,14,24 ; lfpsx 11,11,12
+ fxcsmadd 25,21,14,25 ; lhzux 31,5,10
+ fxcpmadd 26,22,14,26
+ fxcsmadd 27,22,14,27
+ fxcpmadd 28,23,14,28
+ fxcsmadd 29,23,14,29
+ # load 14
+ fxcpmadd 30,20,6,30
+ fxcsmadd 31,20,6,31
+
+ fxcsmadd 24,20,15,24
+ fxcpmadd 25,21,15,25
+ fxcsmadd 26,21,15,26
+ fxcpmadd 27,22,15,27 ; rlwinm 12,28,3,13,28
+ fxcsmadd 28,22,15,28 ; lfpsx 12,11,12
+ fxcpmadd 29,23,15,29 ; lhzux 28,5,10
+ fxcsmadd 30,23,15,30
+ # load 15
+ fxcpmadd 31,20,7,31
+
+ # after loads
+
+ fxcpmadd 24,16,8,24
+ fxcsmadd 25,16,8,25
+ fxcpmadd 26,17,8,26
+ fxcsmadd 27,17,8,27
+ fxcpmadd 28,18,8,28
+ fxcsmadd 29,18,8,29
+ fxcpmadd 30,19,8,30
+ fxcsmadd 31,19,8,31 ; rlwinm 12,29,3,13,28
+
+ fxcsmadd 24,19,1,24 ; lfpsx 13,11,12
+ fxcpmadd 25,16,9,25 ; lhzux 29,5,10
+ fxcsmadd 26,16,9,26
+ fxcpmadd 27,17,9,27
+ fxcsmadd 28,17,9,28
+ fxcpmadd 29,18,9,29
+ fxcsmadd 30,18,9,30
+ fxcpmadd 31,19,9,31
+
+ fxcpmadd 24,19,2,24
+ fxcsmadd 25,19,2,25
+ fxcpmadd 26,16,10,26
+ fxcsmadd 27,16,10,27 ; rlwinm 12,30,3,13,28
+ fxcpmadd 28,17,10,28 ; lfpsx 14,11,12
+ fxcsmadd 29,17,10,29 ; lhzux 30,5,10
+ fxcpmadd 30,18,10,30
+ fxcsmadd 31,18,10,31
+
+ fxcsmadd 24,18,3,24
+ fxcpmadd 25,19,3,25
+ fxcsmadd 26,19,3,26
+ fxcpmadd 27,16,11,27
+ fxcsmadd 28,16,11,28
+ fxcpmadd 29,17,11,29
+ fxcsmadd 30,17,11,30
+ fxcpmadd 31,18,11,31 ; rlwinm 12,31,3,13,28
+
+ fxcpmadd 24,18,4,24 ; lfpsx 15,11,12
+ fxcsmadd 25,18,4,25 ; lhzux 31,5,10
+ fxcpmadd 26,19,4,26
+ fxcsmadd 27,19,4,27
+ fxcpmadd 28,16,12,28
+ fxcsmadd 29,16,12,29
+ fxcpmadd 30,17,12,30
+ fxcsmadd 31,17,12,31
+
+ fxcsmadd 24,17,5,24
+ fxcpmadd 25,18,5,25
+ fxcsmadd 26,18,5,26
+ fxcpmadd 27,19,5,27
+ fxcsmadd 28,19,5,28
+ fxcpmadd 29,16,13,29
+ fxcsmadd 30,16,13,30
+ fxcpmadd 31,17,13,31
+
+ fxcpmadd 24,17,6,24
+ fxcsmadd 25,17,6,25
+ fxcpmadd 26,18,6,26
+ fxcsmadd 27,18,6,27
+ fxcpmadd 28,19,6,28
+ fxcsmadd 29,19,6,29
+ fxcpmadd 30,16,14,30
+ fxcsmadd 31,16,14,31
+
+ fxcsmadd 24,16,7,24
+ fxcpmadd 25,17,7,25
+ fxcsmadd 26,17,7,26
+ fxcpmadd 27,18,7,27
+ fxcsmadd 28,18,7,28
+ fxcpmadd 29,19,7,29 ; stfpsux 24,6,9
+ fxcsmadd 30,19,7,30 ; stfpsux 25,6,9
+ fxcpmadd 31,16,15,31 ; stfpsux 26,6,9
+
+ fxpmul 24,20,8 # part of next loop
+ fxsmul 25,20,8
+ fxpmul 26,21,8 ; stfpsux 27,6,9
+ fxsmul 27,21,8 ; stfpsux 28,6,9
+ fxpmul 28,22,8 ; stfpsux 29,6,9
+ fxsmul 29,22,8 ; stfpsux 30,6,9
+ fxpmul 30,23,8 ; stfpsux 31,6,9
+ fxsmul 31,23,8
+ bdnz loop
+
+#if 0
+ addi 3,3,-120 # store delay line
+ stfpsux 1,3,9
+ stfpsux 2,3,9
+ stfpsux 3,3,9
+ stfpsux 4,3,9
+ stfpsux 5,3,9
+ stfpsux 6,3,9
+ stfpsux 7,3,9
+ stfpsux 8,3,9
+ stfpsux 9,3,9
+ stfpsux 10,3,9
+ stfpsux 11,3,9
+ stfpsux 12,3,9
+ stfpsux 13,3,9
+ stfpsux 14,3,9
+ stfpsux 15,3,9
+#endif
+
+ li 9,DCOMPLEX_SIZE # restore call-saved registers
+ lmw 28,0(1) # restore r28 ... r31
+
+ #addi 1,1,16
+ lfpdux 31,1,9
+ lfpdux 30,1,9
+ lfpdux 29,1,9
+ lfpdux 28,1,9
+ lfpdux 27,1,9
+ lfpdux 26,1,9
+ lfpdux 25,1,9
+ lfpdux 24,1,9
+ lfpdux 23,1,9
+ lfpdux 22,1,9
+ lfpdux 21,1,9
+ lfpdux 20,1,9
+ lfpdux 19,1,9
+ lfpdux 18,1,9
+ lfpdux 17,1,9
+ lfpdux 16,1,9
+ lfpdux 15,1,9
+ lfpdux 14,1,9
+
+ addi 1,1,16 # restore stack pointer
+ blr # return
+
+#elif 0 && NR_BITS_PER_SAMPLE == 16
+
+.align 5
+.comm _FIR_fp_table,65536*FLOAT_SIZE,4
+
+.global _filter
+_filter:
+# filters all samples for one station, one polarization
+
+# arguments:
+# r3 : pointer to delay line (fcomplex[16]) NOTE: USE OF THE DELAY
+# LINE IS COMMENTED OUT!!!!!!!!!!!!!!!!!!!!
+# r4 : pointer to weights line (const fcomplex[16])
+# r5 : pointer to first sample (const i16complex[16*r7])
+# r6 : pointer to result (fcomplex *)
+# r7 : number of samples / 16
+
+# internally used:
+# r9 : 8
+# r10 : 2048
+# r11 : ptr to constant table
+# r28-r31 : prefetched samples
+# f0-f15 : delay line (real in primary, imaginary in secondary unit)
+# f16-f23 : weights (these are real values alternately stored in primary
+# and secondary units)
+# f24-f29 : sums
+
+# The implementation works on 5 or 6 time samples concurrently, to avoid
+# stalls in the double hummer. This unfortunately leads to totally
+# incomprehensible code. The loop processes 16 samples at a time.
+# The input is converted from int16complex to dcomplex by black magic,
+# making the code even harder to understand.
+
+ mtctr 7 # set number of iterations
+
+ li 9,-DCOMPLEX_SIZE
+ stfpdux 14,1,9 # save call-saved registers
+ stfpdux 15,1,9
+ stfpdux 16,1,9
+ stfpdux 17,1,9
+ stfpdux 18,1,9
+ stfpdux 19,1,9
+ stfpdux 20,1,9
+ stfpdux 21,1,9
+ stfpdux 22,1,9
+ stfpdux 23,1,9
+ stfpdux 24,1,9
+ stfpdux 25,1,9
+ stfpdux 26,1,9
+ stfpdux 27,1,9
+ stfpdux 28,1,9
+ stfpdux 29,1,9
+ stfpdux 30,1,9
+ stfpdux 31,1,9
+
+ subi 1,1,16
+ stmw 28,0(1) # save r28 ... r31
+
+ lis 11,_FIR_fp_table@ha
+ li 9,FCOMPLEX_SIZE
+ la 11,_FIR_fp_table@l(11)
+ li 10,I16COMPLEX_SIZE*NR_SUBBAND_CHANNELS*NR_POLARIZATIONS
+ subi 6,6,FCOMPLEX_SIZE
+
+
+#if 0
+ lfpsx 1,0,3 # load delay line
+ lfpsux 2,3,9
+ lfpsux 3,3,9
+ lfpsux 4,3,9
+ lfpsux 5,3,9
+ lfpsux 6,3,9
+ lfpsux 7,3,9
+ lfpsux 8,3,9
+ lfpsux 9,3,9
+ lfpsux 10,3,9
+ lfpsux 11,3,9
+ lfpsux 12,3,9
+ lfpsux 13,3,9
+ lfpsux 14,3,9
+ lfpsux 15,3,9
+#endif
+
+ # convert 15 i16complex numbers to fcomplex
+ lwzx 29,0,5 # fetch FIR history samples
+ lwzux 30,5,10
+ lwzux 31,5,10
+ lwzux 28,5,10
+
+ rlwinm 12,29,2,14,29
+ lfssx 1,11,12
+ rlwinm 12,29,18,14,29
+ lfsx 1,11,12
+ lwzux 29,5,10
+
+ rlwinm 12,30,2,14,29
+ lfssx 2,11,12
+ rlwinm 12,30,18,14,29
+ lfsx 2,11,12
+ lwzux 30,5,10
+
+ rlwinm 12,31,2,14,29
+ lfssx 3,11,12
+ rlwinm 12,31,18,14,29
+ lfsx 3,11,12
+ lwzux 31,5,10
+
+ rlwinm 12,28,2,14,29
+ lfssx 4,11,12
+ rlwinm 12,28,18,14,29
+ lfsx 4,11,12
+ lwzux 28,5,10
+
+ rlwinm 12,29,2,14,29
+ lfssx 5,11,12
+ rlwinm 12,29,18,14,29
+ lfsx 5,11,12
+ lwzux 29,5,10
+
+ rlwinm 12,30,2,14,29
+ lfssx 6,11,12
+ rlwinm 12,30,18,14,29
+ lfsx 6,11,12
+ lwzux 30,5,10
+
+ rlwinm 12,31,2,14,29
+ lfssx 7,11,12
+ rlwinm 12,31,18,14,29
+ lfsx 7,11,12
+ lwzux 31,5,10
+
+ rlwinm 12,28,2,14,29
+ lfssx 8,11,12
+ rlwinm 12,28,18,14,29
+ lfsx 8,11,12
+ lwzux 28,5,10
+
+ rlwinm 12,29,2,14,29
+ lfssx 9,11,12
+ rlwinm 12,29,18,14,29
+ lfsx 9,11,12
+ lwzux 29,5,10
+
+ rlwinm 12,30,2,14,29
+ lfssx 10,11,12
+ rlwinm 12,30,18,14,29
+ lfsx 10,11,12
+ lwzux 30,5,10
+
+ rlwinm 12,31,2,14,29
+ lfssx 11,11,12
+ rlwinm 12,31,18,14,29
+ lfsx 11,11,12
+ lwzux 31,5,10
+
+ rlwinm 12,28,2,14,29
+ lfssx 12,11,12
+ rlwinm 12,28,18,14,29
+ lfsx 12,11,12
+ lwzux 28,5,10 # prefetch samples before entering loop
+
+ rlwinm 12,29,2,14,29
+ lfssx 13,11,12
+ rlwinm 12,29,18,14,29
+ lfsx 13,11,12
+ lwzux 29,5,10
+
+ rlwinm 12,30,2,14,29
+ lfssx 14,11,12
+ rlwinm 12,30,18,14,29
+ lfsx 14,11,12
+ lwzux 30,5,10
+
+ rlwinm 12,31,2,14,29
+ lfssx 15,11,12
+ rlwinm 12,31,18,14,29
+ lfsx 15,11,12
+ lwzux 31,5,10
+
+ ; lfpsx 16,0,4 # load weights line
+ ; lfpsux 17,4,9
+ ; lfpsux 18,4,9
+ ; lfpsux 19,4,9
+ ; lfpsux 20,4,9
+ ; lfpsux 21,4,9
+ ; lfpsux 22,4,9
+ ; lfpsux 23,4,9
+
+ # essentially part of loop, but interleaved with
+ # the stores in the tail of the loop
+ fxpmul 24,20,8
+ fxsmul 25,20,8
+ fxpmul 26,21,8
+ fxsmul 27,21,8
+ fxpmul 28,22,8
+ fxsmul 29,22,8
+ fxpmul 30,23,8
+ fxsmul 31,23,8
+
+loop:
+ # time steps 0-7
+
+ # load 0
- fxcsmadd 24,23,12,24 ; lfpsx 12,11,28
- fxcpmadd 25,20,4,25 ; lbzux 28,5,10
- fxcsmadd 26,20,4,26
- fxcpmadd 27,21,4,27
- fxcsmadd 28,21,4,28
+ fxcsmadd 24,23,1,24 ; rlwinm 12,28,2,14,29
+ # load 1
+ fxcpmadd 25,20,9,25 ; lfssx 0,11,12
+ fxcsmadd 26,20,9,26 ; rlwinm 12,28,18,14,29
+ fxcpmadd 27,21,9,27 ; lfsx 0,11,12
+ fxcsmadd 28,21,9,28 ; lwzux 28,5,10
+ fxcpmadd 29,22,9,29
+ fxcsmadd 30,22,9,30
+ fxcpmadd 31,23,9,31
+
+ fxcpmadd 24,23,2,24 ; rlwinm 12,29,2,14,29
+ fxcsmadd 25,23,2,25 ; lfssx 1,11,12
+ # load 2
+ fxcpmadd 26,20,10,26 ; rlwinm 12,29,18,14,29
+ fxcsmadd 27,20,10,27 ; lfsx 1,11,12
+ fxcpmadd 28,21,10,28 ; lwzux 29,5,10
+ fxcsmadd 29,21,10,29
+ fxcpmadd 30,22,10,30
+ fxcsmadd 31,22,10,31
+
+ fxcsmadd 24,22,3,24 ; rlwinm 12,30,2,14,29
+ fxcpmadd 25,23,3,25 ; lfssx 2,11,12
+ fxcsmadd 26,23,3,26 ; rlwinm 12,30,18,14,29
+ # load 3
+ fxcpmadd 27,20,11,27 ; lfsx 2,11,12
+ fxcsmadd 28,20,11,28 ; lwzux 30,5,10
+ fxcpmadd 29,21,11,29
+ fxcsmadd 30,21,11,30
+ fxcpmadd 31,22,11,31
+
+ fxcpmadd 24,22,4,24 ; rlwinm 12,31,2,14,29
+ fxcsmadd 25,22,4,25 ; lfssx 3,11,12
+ fxcpmadd 26,23,4,26 ; rlwinm 12,31,18,14,29
+ fxcsmadd 27,23,4,27 ; lfsx 3,11,12
+ # load 4
+ fxcpmadd 28,20,12,28 ; lwzux 31,5,10
+ fxcsmadd 29,20,12,29
+ fxcpmadd 30,21,12,30
+ fxcsmadd 31,21,12,31
- fxcsmadd 24,19,4,24
- fxcpmadd 25,16,12,25
- fxcsmadd 26,16,12,26
- fxcpmadd 27,17,12,27
- fxcsmadd 28,17,12,28
+ fxcsmadd 24,21,5,24
+ fxcpmadd 25,22,5,25
+ fxcsmadd 26,22,5,26
+ fxcpmadd 27,23,5,27
+ fxcsmadd 28,23,5,28
+ # load 5
+ fxcpmadd 29,20,13,29
+ fxcsmadd 30,20,13,30
+ fxcpmadd 31,21,13,31
- fxcpmadd 24,23,13,24 ; slwi 29,29,3
- fxcsmadd 25,23,13,25 ; lfpsx 13,11,29
- fxcpmadd 26,20,5,26 ; lbzux 29,5,10
- fxcsmadd 27,20,5,27
- fxcpmadd 28,21,5,28
-
- fxcsmadd 24,17,8,24
- fxcpmadd 25,18,8,25
- fxcsmadd 26,18,8,26
- fxcpmadd 27,19,8,27
- fxcsmadd 28,19,8,28
+ fxcpmadd 24,21,6,24
+ fxcsmadd 25,21,6,25
+ fxcpmadd 26,22,6,26
+ fxcsmadd 27,22,6,27
+ fxcpmadd 28,23,6,28
+ fxcsmadd 29,23,6,29
+ # load 6
+ fxcpmadd 30,20,14,30
+ fxcsmadd 31,20,14,31
- fxcpmadd 24,19,5,24
- fxcsmadd 25,19,5,25
- fxcpmadd 26,16,13,26
- fxcsmadd 27,16,13,27
- fxcpmadd 28,17,13,28
+ fxcsmadd 24,20,7,24
+ fxcpmadd 25,21,7,25
+ fxcsmadd 26,21,7,26
+ fxcpmadd 27,22,7,27 ; rlwinm 12,28,2,14,29
+ fxcsmadd 28,22,7,28 ; lfssx 4,11,12
+ fxcpmadd 29,23,7,29
+ fxcsmadd 30,23,7,30
+ # load 7
+ fxcpmadd 31,20,15,31
- fxcsmadd 24,22,14,24
- fxcpmadd 25,23,14,25 ; slwi 30,30,3
- fxcsmadd 26,23,14,26 ; lfpsx 14,11,30
- fxcpmadd 27,20,6,27 ; lbzux 30,5,10
- fxcsmadd 28,20,6,28
-
- fxcsmadd 24,18,6,24
- fxcpmadd 25,19,6,25
- fxcsmadd 26,19,6,26
- fxcpmadd 27,16,14,27
- fxcsmadd 28,16,14,28
+ # after loads
+ fxcpmadd 24,16,0,24 ; rlwinm 12,28,18,14,29
+ fxcsmadd 25,16,0,25 ; lfsx 4,11,12
+ fxcpmadd 26,17,0,26
+ fxcsmadd 27,17,0,27 ; lwzux 28,5,10
+ fxcpmadd 28,18,0,28
+ fxcsmadd 29,18,0,29
+ fxcpmadd 30,19,0,30
+ fxcsmadd 31,19,0,31 ; rlwinm 12,29,2,14,29
- fxcpmadd 24,21,1,24
- fxcsmadd 25,21,1,25
- fxcpmadd 26,22,1,26
- fxcsmadd 27,22,1,27
- fxcpmadd 28,23,1,28
+ fxcsmadd 24,19,9,24 ; lfssx 5,11,12
+ fxcpmadd 25,16,1,25
+ fxcsmadd 26,16,1,26
+ fxcpmadd 27,17,1,27
+ fxcsmadd 28,17,1,28 ; rlwinm 12,29,18,14,29
+ fxcpmadd 29,18,1,29 ; lfsx 5,11,12
+ fxcsmadd 30,18,1,30
+ fxcpmadd 31,19,1,31 ; lwzux 29,5,10
- fxcpmadd 24,22,15,24
- fxcsmadd 25,22,15,25
- fxcpmadd 26,23,15,26 ; slwi 31,31,3
- fxcsmadd 27,23,15,27 ; lfpsx 15,11,31
- fxcpmadd 28,20,7,28 ; lbzux 31,5,10
+ fxcpmadd 24,19,10,24
+ fxcsmadd 25,19,10,25
+ fxcpmadd 26,16,2,26
+ fxcsmadd 27,16,2,27 ; rlwinm 12,30,2,14,29
+ fxcpmadd 28,17,2,28 ; lfssx 6,11,12
+ fxcsmadd 29,17,2,29
+ fxcpmadd 30,18,2,30
+ fxcsmadd 31,18,2,31
- fxcpmadd 24,18,7,24
- fxcsmadd 25,18,7,25
- fxcpmadd 26,19,7,26
- fxcsmadd 27,19,7,27
- fxcpmadd 28,16,15,28
+ fxcsmadd 24,18,11,24 ; rlwinm 12,30,18,14,29
+ fxcpmadd 25,19,11,25 ; lfsx 6,11,12
+ fxcsmadd 26,19,11,26
+ fxcpmadd 27,16,3,27 ; lwzux 30,5,10
+ fxcsmadd 28,16,3,28
+ fxcpmadd 29,17,3,29
+ fxcsmadd 30,17,3,30
+ fxcpmadd 31,18,3,31 ; rlwinm 12,31,2,14,29
+
+ fxcpmadd 24,18,12,24 ; lfssx 7,11,12
+ fxcsmadd 25,18,12,25
+ fxcpmadd 26,19,12,26
+ fxcsmadd 27,19,12,27
+ fxcpmadd 28,16,4,28 ; rlwinm 12,31,18,14,29
+ fxcsmadd 29,16,4,29 ; lfsx 7,11,12
+ fxcpmadd 30,17,4,30
+ fxcsmadd 31,17,4,31 ; lwzux 31,5,10
- fxpmul 29,19,15 ; stfpsux 24,6,9
- fxsmul 24,16,15 ; stfpsux 25,6,9
- fxpmul 25,17,15 ; stfpsux 26,6,9
- fxsmul 26,17,15 ; stfpsux 27,6,9
- fxpmul 27,18,15 ; stfpsux 28,6,9
- fxsmul 28,18,15
+ fxcsmadd 24,17,13,24
+ fxcpmadd 25,18,13,25
+ fxcsmadd 26,18,13,26
+ fxcpmadd 27,19,13,27
+ fxcsmadd 28,19,13,28
+ fxcpmadd 29,16,5,29
+ fxcsmadd 30,16,5,30
+ fxcpmadd 31,17,5,31 ; rlwinm 12,28,2,14,29
+
+ fxcpmadd 24,17,14,24 ; lfssx 8,11,12
+ fxcsmadd 25,17,14,25
+ fxcpmadd 26,18,14,26
+ fxcsmadd 27,18,14,27
+ fxcpmadd 28,19,14,28 ; rlwinm 12,28,18,14,29
+ fxcsmadd 29,19,14,29 ; lfsx 8,11,12
+ fxcpmadd 30,16,6,30
+ fxcsmadd 31,16,6,31 ; lwzux 28,5,10
+
+ fxcsmadd 24,16,15,24
+ fxcpmadd 25,17,15,25
+ fxcsmadd 26,17,15,26
+ fxcpmadd 27,18,15,27
+ fxcsmadd 28,18,15,28
+ fxcpmadd 29,19,15,29 ; stfpsux 24,6,9
+ fxcsmadd 30,19,15,30 ; stfpsux 25,6,9
+ fxcpmadd 31,16,7,31 ; stfpsux 26,6,9
+
+ # time steps 8-15
+ fxpmul 24,20,0
+ fxsmul 25,20,0
+ fxpmul 26,21,0 ; stfpsux 27,6,9
+ fxsmul 27,21,0 ; stfpsux 28,6,9
+ fxpmul 28,22,0 ; stfpsux 29,6,9
+ fxsmul 29,22,0 ; stfpsux 30,6,9
+ fxpmul 30,23,0 ; stfpsux 31,6,9
+ fxsmul 31,23,0
+
+ # already loaded 8
+
+ fxcsmadd 24,23,9,24
+ # load 9
+ fxcpmadd 25,20,1,25
+ fxcsmadd 26,20,1,26
+ fxcpmadd 27,21,1,27
+ fxcsmadd 28,21,1,28
+ fxcpmadd 29,22,1,29
+ fxcsmadd 30,22,1,30
+ fxcpmadd 31,23,1,31
+
+ fxcpmadd 24,23,10,24
+ fxcsmadd 25,23,10,25
+ # load 10
+ fxcpmadd 26,20,2,26
+ fxcsmadd 27,20,2,27
+ fxcpmadd 28,21,2,28
+ fxcsmadd 29,21,2,29
+ fxcpmadd 30,22,2,30
+ fxcsmadd 31,22,2,31 ; rlwinm 12,29,2,14,29
+
+ fxcsmadd 24,22,11,24 ; lfssx 9,11,12
+ fxcpmadd 25,23,11,25
+ fxcsmadd 26,23,11,26
+ # load 11
+ fxcpmadd 27,20,3,27
+ fxcsmadd 28,20,3,28 ; rlwinm 12,29,18,14,29
+ fxcpmadd 29,21,3,29 ; lfsx 9,11,12
+ fxcsmadd 30,21,3,30
+ fxcpmadd 31,22,3,31 ; lwzux 29,5,10
+
+ fxcpmadd 24,22,12,24
+ fxcsmadd 25,22,12,25
+ fxcpmadd 26,23,12,26
+ fxcsmadd 27,23,12,27 ; rlwinm 12,30,2,14,29
+ # load 12
+ fxcpmadd 28,20,4,28 ; lfssx 10,11,12
+ fxcsmadd 29,20,4,29
+ fxcpmadd 30,21,4,30
+ fxcsmadd 31,21,4,31
+
+ fxcsmadd 24,21,13,24 ; rlwinm 12,30,18,14,29
+ fxcpmadd 25,22,13,25 ; lfsx 10,11,12
+ fxcsmadd 26,22,13,26
+ fxcpmadd 27,23,13,27 ; lwzux 30,5,10
+ fxcsmadd 28,23,13,28
+ # load 13
+ fxcpmadd 29,20,5,29
+ fxcsmadd 30,20,5,30
+ fxcpmadd 31,21,5,31 ; rlwinm 12,31,2,14,29
+
+ fxcpmadd 24,21,14,24 ; lfssx 11,11,12
+ fxcsmadd 25,21,14,25
+ fxcpmadd 26,22,14,26
+ fxcsmadd 27,22,14,27
+ fxcpmadd 28,23,14,28 ; rlwinm 12,31,18,14,29
+ fxcsmadd 29,23,14,29 ; lfsx 11,11,12
+ # load 14
+ fxcpmadd 30,20,6,30
+ fxcsmadd 31,20,6,31 ; lwzux 31,5,10
+
+ fxcsmadd 24,20,15,24
+ fxcpmadd 25,21,15,25
+ fxcsmadd 26,21,15,26
+ fxcpmadd 27,22,15,27 ; rlwinm 12,28,2,14,29
+ fxcsmadd 28,22,15,28 ; lfssx 12,11,12
+ fxcpmadd 29,23,15,29
+ fxcsmadd 30,23,15,30
+ # load 15
+ fxcpmadd 31,20,7,31
+
+ # after loads
+
+ fxcpmadd 24,16,8,24 ; rlwinm 12,28,18,14,29
+ fxcsmadd 25,16,8,25 ; lfsx 12,11,12
+ fxcpmadd 26,17,8,26
+ fxcsmadd 27,17,8,27 ; lwzux 28,5,10
+ fxcpmadd 28,18,8,28
+ fxcsmadd 29,18,8,29
+ fxcpmadd 30,19,8,30
+ fxcsmadd 31,19,8,31 ; rlwinm 12,29,2,14,29
+
+ fxcsmadd 24,19,1,24 ; lfssx 13,11,12
+ fxcpmadd 25,16,9,25
+ fxcsmadd 26,16,9,26
+ fxcpmadd 27,17,9,27
+ fxcsmadd 28,17,9,28 ; rlwinm 12,29,18,14,29
+ fxcpmadd 29,18,9,29 ; lfsx 13,11,12
+ fxcsmadd 30,18,9,30
+ fxcpmadd 31,19,9,31 ; lwzux 29,5,10
+
+ fxcpmadd 24,19,2,24
+ fxcsmadd 25,19,2,25
+ fxcpmadd 26,16,10,26
+ fxcsmadd 27,16,10,27 ; rlwinm 12,30,2,14,29
+ fxcpmadd 28,17,10,28 ; lfssx 14,11,12
+ fxcsmadd 29,17,10,29
+ fxcpmadd 30,18,10,30
+ fxcsmadd 31,18,10,31
+
+ fxcsmadd 24,18,3,24 ; rlwinm 12,30,18,14,29
+ fxcpmadd 25,19,3,25 ; lfsx 14,11,12
+ fxcsmadd 26,19,3,26
+ fxcpmadd 27,16,11,27 ; lwzux 30,5,10
+ fxcsmadd 28,16,11,28
+ fxcpmadd 29,17,11,29
+ fxcsmadd 30,17,11,30
+ fxcpmadd 31,18,11,31 ; rlwinm 12,31,2,14,29
+
+ fxcpmadd 24,18,4,24 ; lfssx 15,11,12
+ fxcsmadd 25,18,4,25
+ fxcpmadd 26,19,4,26
+ fxcsmadd 27,19,4,27
+ fxcpmadd 28,16,12,28 ; rlwinm 12,31,18,14,29
+ fxcsmadd 29,16,12,29 ; lfsx 15,11,12
+ fxcpmadd 30,17,12,30
+ fxcsmadd 31,17,12,31 ; lwzux 31,5,10
+
+ fxcsmadd 24,17,5,24
+ fxcpmadd 25,18,5,25
+ fxcsmadd 26,18,5,26
+ fxcpmadd 27,19,5,27
+ fxcsmadd 28,19,5,28
+ fxcpmadd 29,16,13,29
+ fxcsmadd 30,16,13,30
+ fxcpmadd 31,17,13,31
+
+ fxcpmadd 24,17,6,24
+ fxcsmadd 25,17,6,25
+ fxcpmadd 26,18,6,26
+ fxcsmadd 27,18,6,27
+ fxcpmadd 28,19,6,28
+ fxcsmadd 29,19,6,29
+ fxcpmadd 30,16,14,30
+ fxcsmadd 31,16,14,31
+
+ fxcsmadd 24,16,7,24
+ fxcpmadd 25,17,7,25
+ fxcsmadd 26,17,7,26
+ fxcpmadd 27,18,7,27
+ fxcsmadd 28,18,7,28
+ fxcpmadd 29,19,7,29 ; stfpsux 24,6,9
+ fxcsmadd 30,19,7,30 ; stfpsux 25,6,9
+ fxcpmadd 31,16,15,31 ; stfpsux 26,6,9
+
+ fxpmul 24,20,8 # part of next loop
+ fxsmul 25,20,8
+ fxpmul 26,21,8 ; stfpsux 27,6,9
+ fxsmul 27,21,8 ; stfpsux 28,6,9
+ fxpmul 28,22,8 ; stfpsux 29,6,9
+ fxsmul 29,22,8 ; stfpsux 30,6,9
+ fxpmul 30,23,8 ; stfpsux 31,6,9
+ fxsmul 31,23,8
bdnz loop
- la 8,16(1) # restore call-saved registers
- li 9,DCOMPLEX_SIZE
+#if 0
+ addi 3,3,-120 # store delay line
+ stfpsux 1,3,9
+ stfpsux 2,3,9
+ stfpsux 3,3,9
+ stfpsux 4,3,9
+ stfpsux 5,3,9
+ stfpsux 6,3,9
+ stfpsux 7,3,9
+ stfpsux 8,3,9
+ stfpsux 9,3,9
+ stfpsux 10,3,9
+ stfpsux 11,3,9
+ stfpsux 12,3,9
+ stfpsux 13,3,9
+ stfpsux 14,3,9
+ stfpsux 15,3,9
+#endif
- lfpdx 14,0,8
- lfpdux 15,8,9
- lfpdux 16,8,9
- lfpdux 17,8,9
- lfpdux 18,8,9
- lfpdux 19,8,9
- lfpdux 20,8,9
- lfpdux 21,8,9
- lfpdux 22,8,9
- lfpdux 23,8,9
- lfpdux 24,8,9
- lfpdux 25,8,9
- lfpdux 26,8,9
- lfpdux 27,8,9
- lfpdux 28,8,9
- lfpdux 29,8,9
- lmw 28,0(1) # restore r28 ... r31
+ li 9,DCOMPLEX_SIZE # restore call-saved registers
+ lmw 28,0(1) # restore r28 ... r31
- addi 1,1,272 # restore stack pointer
+ #addi 1,1,16
+ lfpdux 31,1,9
+ lfpdux 30,1,9
+ lfpdux 29,1,9
+ lfpdux 28,1,9
+ lfpdux 27,1,9
+ lfpdux 26,1,9
+ lfpdux 25,1,9
+ lfpdux 24,1,9
+ lfpdux 23,1,9
+ lfpdux 22,1,9
+ lfpdux 21,1,9
+ lfpdux 20,1,9
+ lfpdux 19,1,9
+ lfpdux 18,1,9
+ lfpdux 17,1,9
+ lfpdux 16,1,9
+ lfpdux 15,1,9
+ lfpdux 14,1,9
+
+ addi 1,1,16 # restore stack pointer
blr # return
-#elif INPUT_TYPE == I16COMPLEX_TYPE
+#elif NR_BITS_PER_SAMPLE == 16
.align 5
sub_value:
@@ -1158,116 +1992,118 @@ _filter:
loop:
# time steps 0-5
- fxcpmadd 24,17,14,24
- fxcsmadd 25,17,14,25 ; xor 28,28,12
- fxcpmadd 26,18,14,26 ; sthbrx 28,7,1
- fxcsmadd 27,18,14,27
- fxcpmadd 28,19,14,28
- fxcsmadd 29,19,14,29 ; srawi 28,28,16
+ fxcsmadd 24,23,1,24 ; xor 28,28,12
+ fxcpmadd 25,20,9,25 ; sthbrx 28,7,1
+ fxcsmadd 26,20,9,26 ; srawi 28,28,16
+ fxcpmadd 27,21,9,27 ; sthbrx 28,11,1
+ fxcsmadd 28,21,9,28
+ fxcpmadd 29,22,9,29
- fxcpmadd 24,21,6,24 ; sthbrx 28,11,1
- fxcsmadd 25,21,6,25
- fxcpmadd 26,22,6,26
- fxcsmadd 27,22,6,27 ; lfpdx 30,8,1
- fxcpmadd 28,23,6,28
- fxcsmadd 29,23,6,29 ; lwzux 28,5,10
+ fxcpmadd 24,23,2,24
+ fxcsmadd 25,23,2,25 ; lfpdx 30,8,1
+ fxcpmadd 26,20,10,26
+ fxcsmadd 27,20,10,27
+ fxcpmadd 28,21,10,28
+ fxcsmadd 29,21,10,29 ; lwzux 28,5,10
- fxcpmadd 24,20,8,24
- fxcsmadd 25,20,8,25
- fpsub 0,30,31
- fxcpmadd 26,21,8,26 ; xor 29,29,12
- fxcsmadd 27,21,8,27 ; sthbrx 29,7,1
- fxcpmadd 28,22,8,28
- fxcsmadd 29,22,8,29
+ fxcsmadd 24,22,3,24
+ fxcpmadd 25,23,3,25
+ fxcsmadd 26,23,3,26
+ fxcpmadd 27,20,11,27 ; xor 29,29,12
+ fxcsmadd 28,20,11,28 ; sthbrx 29,7,1
+ fxcpmadd 29,21,11,29 ; srawi 29,29,16
- fxcpmadd 24,16,0,24 ; srawi 29,29,16
- fxcsmadd 25,16,0,25 ; sthbrx 29,11,1
- fxcpmadd 26,17,0,26
- fxcsmadd 27,17,0,27
- fxcpmadd 28,18,0,28 ; lfpdx 30,8,1
- fxcsmadd 29,18,0,29
+ fxcpmadd 24,22,4,24 ; sthbrx 29,11,1
+ fxcsmadd 25,22,4,25
+ fxcpmadd 26,23,4,26
+ fxcsmadd 27,23,4,27
+ fpsub 0,30,31
+ fxcpmadd 28,20,12,28 ; lfpdx 30,8,1
+ fxcsmadd 29,20,12,29
- fxcsmadd 24,23,1,24 ; lwzux 29,5,10
- fxcpmadd 25,20,9,25
- fxcsmadd 26,20,9,26
- fpsub 1,30,31
- fxcpmadd 27,21,9,27 ; xor 30,30,12
- fxcsmadd 28,21,9,28 ; sthbrx 30,7,1
- fxcpmadd 29,22,9,29
+ fxcsmadd 24,21,5,24
+ fxcpmadd 25,22,5,25
+ fxcsmadd 26,22,5,26 ; lwzux 29,5,10
+ fxcpmadd 27,23,5,27
+ fxcsmadd 28,23,5,28
+ fxcpmadd 29,20,13,29
- fxcsmadd 24,19,9,24
- fxcpmadd 25,16,1,25 ; srawi 30,30,16
- fxcsmadd 26,16,1,26 ; sthbrx 30,11,1
- fxcpmadd 27,17,1,27
- fxcsmadd 28,17,1,28
- fxcpmadd 29,18,1,29 ; lfpdx 30,8,1
+ fxcpmadd 24,21,6,24 ; xor 30,30,12
+ fxcsmadd 25,21,6,25 ; sthbrx 30,7,1
+ fxcpmadd 26,22,6,26 ; srawi 30,30,16
+ fxcsmadd 27,22,6,27 ; sthbrx 30,11,1
+ fxcpmadd 28,23,6,28
+ fxcsmadd 29,23,6,29
- fxcpmadd 24,23,2,24
- fxcsmadd 25,23,2,25 ; lwzux 30,5,10
- fxcpmadd 26,20,10,26
- fxcsmadd 27,20,10,27
- fpsub 2,30,31
- fxcpmadd 28,21,10,28 ; xor 31,31,12
- fxcsmadd 29,21,10,29 ; sthbrx 31,7,1
-
fxcsmadd 24,20,7,24
fxcpmadd 25,21,7,25
- fxcsmadd 26,21,7,26
+ fpsub 1,30,31
+ fxcsmadd 26,21,7,26 ; lfpdx 30,8,1
fxcpmadd 27,22,7,27
fxcsmadd 28,22,7,28
- fxcpmadd 29,23,7,29
+ fxcpmadd 29,23,7,29 ; lwzux 30,5,10
+
+ fxcpmadd 24,20,8,24
+ fxcsmadd 25,20,8,25
+ fxcpmadd 26,21,8,26
+ fxcsmadd 27,21,8,27
+ fxcpmadd 28,22,8,28
+ fxcsmadd 29,22,8,29 ; xor 31,31,12
+
+ fxcpmadd 24,17,14,24 ; sthbrx 31,7,1
+ fxcsmadd 25,17,14,25
+ fxcpmadd 26,18,14,26
+ fxcsmadd 27,18,14,27 ; srawi 31,31,16
+ fxcpmadd 28,19,14,28 ; sthbrx 31,11,1
+ fxcsmadd 29,19,14,29
+
+ fxcpmadd 24,16,0,24
+ fxcsmadd 25,16,0,25
+ fpsub 2,30,31
+ fxcpmadd 26,17,0,26 ; lfpdx 30,8,1
+ fxcsmadd 27,17,0,27
+ fxcpmadd 28,18,0,28
+ fxcsmadd 29,18,0,29 ; lwzux 31,5,10
+
+ fpsub 3,30,31 ; xor 28,28,12
+ fxcsmadd 24,19,9,24 ; sthbrx 28,7,1
+ fxcpmadd 25,16,1,25
+ fxcsmadd 26,16,1,26
+ fxcpmadd 27,17,1,27 ; srawi 28,28,16
+ fxcsmadd 28,17,1,28 ; sthbrx 28,11,1
+ fxcpmadd 29,18,1,29
fxcpmadd 24,19,10,24
fxcsmadd 25,19,10,25
- fxcpmadd 26,16,2,26 ; srawi 31,31,16
- fxcsmadd 27,16,2,27 ; sthbrx 31,11,1
+ fxcpmadd 26,16,2,26 ; lfpdx 30,8,1
+ fxcsmadd 27,16,2,27
fxcpmadd 28,17,2,28
- fxcsmadd 29,17,2,29
-
- fxcsmadd 24,22,3,24 ; lfpdx 30,8,1
- fxcpmadd 25,23,3,25
- fxcsmadd 26,23,3,26 ; lwzux 31,5,10
- fxcpmadd 27,20,11,27
- fxcsmadd 28,20,11,28
- fpsub 3,30,31
- fxcpmadd 29,21,11,29 ; xor 28,28,12
+ fxcsmadd 29,17,2,29 ; lwzux 28,5,10
- fxcsmadd 24,18,11,24 ; sthbrx 28,7,1
+ fpsub 4,30,31 ; xor 29,29,12
+ fxcsmadd 24,18,11,24 ; sthbrx 29,7,1
fxcpmadd 25,19,11,25
fxcsmadd 26,19,11,26
- fxcpmadd 27,16,3,27 ; srawi 28,28,16
- fxcsmadd 28,16,3,28 ; sthbrx 28,11,1
+ fxcpmadd 27,16,3,27 ; srawi 29,29,16
+ fxcsmadd 28,16,3,28 ; sthbrx 29,11,1
fxcpmadd 29,17,3,29
- fxcpmadd 24,22,4,24
- fxcsmadd 25,22,4,25 ; lfpdx 30,8,1
- fxcpmadd 26,23,4,26
- fxcsmadd 27,23,4,27 ; lwzux 28,5,10
- fxcpmadd 28,20,12,28
- fxcsmadd 29,20,12,29
-
- fpsub 4,30,31
- fxcpmadd 24,18,12,24 ; xor 29,29,12
- fxcsmadd 25,18,12,25 ; sthbrx 29,7,1
- fxcpmadd 26,19,12,26
+ fxcpmadd 24,18,12,24
+ fxcsmadd 25,18,12,25
+ fxcpmadd 26,19,12,26 ; lfpdx 30,8,1
fxcsmadd 27,19,12,27
- fxcpmadd 28,16,4,28 ; srawi 29,29,16
- fxcsmadd 29,16,4,29 ; sthbrx 29,11,1
-
- fxcsmadd 24,21,5,24
- fxcpmadd 25,22,5,25
- fxcsmadd 26,22,5,26 ; lfpdx 30,8,1
- fxcpmadd 27,23,5,27
- fxcsmadd 28,23,5,28 ; lwzux 29,5,10
- fxcpmadd 29,20,13,29
+ fxcpmadd 28,16,4,28
+ fxcsmadd 29,16,4,29 ; lwzux 29,5,10
- fxcsmadd 24,17,13,24
fpsub 5,30,31
+ fxcsmadd 24,17,13,24
fxcpmadd 25,18,13,25
fxcsmadd 26,18,13,26
fxcpmadd 27,19,13,27
- fxcsmadd 28,19,13,28 ; stfpsux 24,6,9
- fxcpmadd 29,16,5,29 ; stfpsux 25,6,9
+ fxcsmadd 28,19,13,28
+ fxcpmadd 29,16,5,29
+ ; stfpsux 24,6,9
+ ; stfpsux 25,6,9
# time steps 6-10
@@ -1530,9 +2366,8 @@ loop:
addi 1,1,16 # restore stack pointer
blr # return
-
#else
-#error INPUT_TYPE not supported
+#error unsupported NR_BITS_PER_SAMPLE
#endif
#if 0
diff --git a/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.h b/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.h
index c40ce8e59ffd298c7800dd947e6768db651f9cfb..a5f0eaf25fc09cca82f78ea7d5abb569f75386c1 100644
--- a/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.h
+++ b/Appl/CEP/CS1/CS1_BGLProc/src/FIR_Asm.h
@@ -61,7 +61,7 @@ extern "C" {
void _prefetch(const void *src, size_t count, size_t stride);
extern struct {
- unsigned input_type;
+ unsigned nr_bits_per_sample;
unsigned nr_subband_channels;
unsigned nr_taps;
unsigned nr_polarizations;
@@ -72,6 +72,14 @@ extern "C" {
#endif
unsigned long long _rdtsc();
+
+#if NR_BITS_PER_SAMPLE == 4
+ extern fcomplex _FIR_fp_table[16][16];
+#elif NR_BITS_PER_SAMPLE == 8
+ extern fcomplex _FIR_fp_table[256][256];
+#elif NR_BITS_PER_SAMPLE == 16
+ extern float _FIR_fp_table[65536];
+#endif
};
} // namespace CS1
diff --git a/Appl/CEP/CS1/CS1_BGLProc/src/PPF.cc b/Appl/CEP/CS1/CS1_BGLProc/src/PPF.cc
index de4ea11b9f7c87c854e272809583209f1b4e09ad..4851bd28118dc434587e0b10908f31f62225cd0e 100644
--- a/Appl/CEP/CS1/CS1_BGLProc/src/PPF.cc
+++ b/Appl/CEP/CS1/CS1_BGLProc/src/PPF.cc
@@ -6,6 +6,7 @@
#include <FFT_Asm.h>
#include <FIR_Asm.h>
+#include <Common/DataConvert.h>
#include <Common/Timer.h>
#include <complex>
@@ -57,6 +58,25 @@ PPF::PPF(unsigned nrStations, unsigned nrSamplesPerIntegration, double channelBa
#endif
{
init_fft();
+
+#if !defined PPF_C_IMPLEMENTATION
+#if NR_BITS_PER_SAMPLE == 4
+ static const float map[] = {
+ 0, 1, 2, 3, 4, 5, 6, 7, -8, -7, -6, -5, -4, -3, -2, -1,
+ };
+
+ for (unsigned i = 0; i < 16; i ++)
+ for (unsigned j = 0; j < 16; j ++)
+ _FIR_fp_table[i][j] = makefcomplex(map[j], map[i]);
+#elif NR_BITS_PER_SAMPLE == 8
+ for (unsigned i = 0; i < 256; i ++)
+ for (unsigned j = 0; j < 256; j ++)
+ _FIR_fp_table[i][j] = makefcomplex((float) (signed char) i, (float) (signed char) j);
+#elif 0 && NR_BITS_PER_SAMPLE == 16
+ for (unsigned i = 0; i < 65536; i ++)
+ _FIR_fp_table[i] = (float) byteSwap((signed short) i);
+#endif
+#endif
}
diff --git a/Appl/CEP/CS1/CS1_BGLProc/test/tBGL_Processing.cc b/Appl/CEP/CS1/CS1_BGLProc/test/tBGL_Processing.cc
index 8d82969b78c3302eb5491a5dca7cdd957f3498c0..d86e62629f274112827f8c0bf7b122164bc42dd0 100644
--- a/Appl/CEP/CS1/CS1_BGLProc/test/tBGL_Processing.cc
+++ b/Appl/CEP/CS1/CS1_BGLProc/test/tBGL_Processing.cc
@@ -52,12 +52,14 @@ inline TransposedData::SampleType toComplex(double phi)
double s, c;
sincos(phi, &s, &c);
-#if INPUT_TYPE == I4COMPLEX_TYPE
+#if NR_BITS_PER_SAMPLE == 4
return makei4complex((int) rint(7 * c), (int) rint(7 * s));
-#elif INPUT_TYPE == I16COMPLEX_TYPE
+#elif NR_BITS_PER_SAMPLE == 8
+ return makei8complex((int) rint(127 * c), (int) rint(127 * s));
+#elif NR_BITS_PER_SAMPLE == 16
return makei16complex((int) rint(32767 * c), (int) rint(32767 * s));
#else
-#error Unknown INPUT_TYPE
+#error Unknown NR_BITS_PER_SAMPLE
#endif
}
diff --git a/Appl/CEP/CS1/CS1_Interface/include/CS1_Interface/CS1_Config.h b/Appl/CEP/CS1/CS1_Interface/include/CS1_Interface/CS1_Config.h
index c525af071499ab76ed2b8460b618b3cf8042e104..00cc141693fe9e79e01788fdcd79902a1772cb91 100644
--- a/Appl/CEP/CS1/CS1_Interface/include/CS1_Interface/CS1_Config.h
+++ b/Appl/CEP/CS1/CS1_Interface/include/CS1_Interface/CS1_Config.h
@@ -25,22 +25,21 @@
/* This is included by C++ and assembly files. Do not put anything but
constants here! */
-#define INPUT_TYPE I16COMPLEX_TYPE
+#define NR_BITS_PER_SAMPLE 16
#define NR_POLARIZATIONS 2
#define NR_SUBBAND_CHANNELS 256
#define NR_TAPS 16
/* Do not change anything below this line */
-#define I4COMPLEX_TYPE 1
-#define I16COMPLEX_TYPE 2
-
-#if INPUT_TYPE == I4COMPLEX_TYPE
+#if NR_BITS_PER_SAMPLE == 4
#define INPUT_SAMPLE_TYPE i4complex
-#elif INPUT_TYPE == I16COMPLEX_TYPE
+#elif NR_BITS_PER_SAMPLE == 8
+#define INPUT_SAMPLE_TYPE i8complex
+#elif NR_BITS_PER_SAMPLE == 16
#define INPUT_SAMPLE_TYPE i16complex
#else
-#error Bad INPUT_TYPE
+#error Bad NR_BITS_PER_SAMPLE
#endif
#endif
diff --git a/LCS/Common/include/Common/ComplexStdInt.h b/LCS/Common/include/Common/ComplexStdInt.h
index 845408344675cc450a93cb34ea18bb22e2c3173a..24706e40638fc877019a9a0e0ab424a6eabb1b12 100644
--- a/LCS/Common/include/Common/ComplexStdInt.h
+++ b/LCS/Common/include/Common/ComplexStdInt.h
@@ -31,10 +31,13 @@
namespace LOFAR {
namespace TYPES {
+ typedef std::complex<int8> i8complex;
typedef std::complex<int16> i16complex;
typedef std::complex<uint16> u16complex;
}
+ inline TYPES::i8complex makei8complex (TYPES::int8 re, TYPES::int8 im)
+ { return TYPES::i8complex(re,im); }
inline TYPES::i16complex makei16complex (TYPES::uint16 re, TYPES::uint16 im)
{ return TYPES::i16complex(re,im); }
inline TYPES::u16complex makeu16complex (TYPES::uint16 re, TYPES::uint16 im)
diff --git a/LCS/Common/include/Common/lofar_complex.h b/LCS/Common/include/Common/lofar_complex.h
index 3ed9e6d1d1bc6d7f5291ba4c27834800a8955bbe..287a6dd380a392529c7ad7bbbcd211eb9fb55313 100644
--- a/LCS/Common/include/Common/lofar_complex.h
+++ b/LCS/Common/include/Common/lofar_complex.h
@@ -60,6 +60,7 @@ namespace LOFAR
{
// Define complex types in LOFAR namespace.
using TYPES::i4complex;
+ using TYPES::i8complex;
using TYPES::i16complex;
using TYPES::u16complex;
using TYPES::fcomplex;
@@ -69,6 +70,10 @@ namespace LOFAR
return z;
}
+ inline static i4complex makei4complex(i8complex &z) {
+ return makei4complex(real(z), imag(z));
+ }
+
inline static i4complex makei4complex(i16complex &z) {
return makei4complex(real(z), imag(z));
}
@@ -85,10 +90,38 @@ namespace LOFAR
return makei4complex((int) real(z), (int) imag(z));
}
+ inline static i8complex makei8complex(i4complex &z) {
+ return makei8complex(real(z), imag(z));
+ }
+
+ inline static i8complex makei8complex(i8complex &z) {
+ return z;
+ }
+
+ inline static i8complex makei8complex(i16complex &z) {
+ return makei8complex(real(z), imag(z));
+ }
+
+ inline static i8complex makei8complex(u16complex &z) {
+ return makei8complex(real(z), imag(z));
+ }
+
+ inline static i8complex makei8complex(fcomplex &z) {
+ return makei8complex((int) real(z), (int) imag(z));
+ }
+
+ inline static i8complex makei8complex(dcomplex &z) {
+ return makei8complex((int) real(z), (int) imag(z));
+ }
+
inline static i16complex makei16complex(i4complex &z) {
return makei16complex(real(z), imag(z));
}
+ inline static i16complex makei16complex(i8complex &z) {
+ return makei16complex(real(z), imag(z));
+ }
+
inline static i16complex makei16complex(i16complex &z) {
return z;
}
@@ -109,6 +142,10 @@ namespace LOFAR
return makeu16complex(real(z), imag(z));
}
+ inline static u16complex makeu16complex(i8complex &z) {
+ return makeu16complex(real(z), imag(z));
+ }
+
inline static u16complex makeu16complex(i16complex &z) {
return makeu16complex(real(z), imag(z));
}
@@ -129,6 +166,10 @@ namespace LOFAR
return makefcomplex((float) real(z), (float) imag(z));
}
+ inline static fcomplex makefcomplex(i8complex &z) {
+ return makefcomplex((float) real(z), (float) imag(z));
+ }
+
inline static fcomplex makefcomplex(i16complex &z) {
return makefcomplex((float) real(z), (float) imag(z));
}
@@ -149,6 +190,10 @@ namespace LOFAR
return makedcomplex((double) real(z), (double) imag(z));
}
+ inline static dcomplex makedcomplex(i8complex &z) {
+ return makedcomplex((double) real(z), (double) imag(z));
+ }
+
inline static dcomplex makedcomplex(i16complex &z) {
return makedcomplex((double) real(z), (double) imag(z));
}