X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fadd%2Ffsqrt.py;h=02449b0f75b41b0ba2e004711d9570ccca49ca3d;hb=6bff1a997f3846872cf489c24b5c01426c4dc97c;hp=5aaec4c37be8a52f55f8af742324e3143a2ce1a7;hpb=7072c408c76a9b5a3f0a5749018dbc9f987db119;p=ieee754fpu.git

diff --git a/src/add/fsqrt.py b/src/add/fsqrt.py
index 5aaec4c3..02449b0f 100644
--- a/src/add/fsqrt.py
+++ b/src/add/fsqrt.py
@@ -22,31 +22,25 @@ def sqrtsimple(num):
 
 def sqrt(num):
     D = num # D is input (from num)
-    Q = 0
-    R = 0
-    r = 0 # remainder
-    for i in range(15, -1, -1): # negative ranges are weird...
+    Q = 0 # quotient
+    R = 0 # remainder
+    for i in range(64, -1, -1): # negative ranges are weird...
 
-        if (R>=0):
-
-            R = (R<<2)|((D>>(i+i))&3)
-            R = R-((Q<<2)|1) #/*-Q01*/
+        R = (R<<2)|((D>>(i+i))&3)
 
+        if R >= 0:
+            R -= ((Q<<2)|1) # -Q01
         else:
+            R += ((Q<<2)|3) # +Q11
 
-            R = (R<<2)|((D>>(i+i))&3)
-            R = R+((Q<<2)|3) #/*+Q11*/
-
-        if (R>=0):
-            Q = (Q<<1)|1 #/*new Q:*/
-        else:
-            Q = (Q<<1)|0 #/*new Q:*/
+        Q <<= 1
+        if R >= 0:
+            Q |= 1 # new Q
 
+    if R < 0:
+        R = R + ((Q<<1)|1)
 
-    if (R<0):
-        R = R+((Q<<1)|1)
-        r = R
-    return Q
+    return Q, R
 
 
 # grabbed these from unit_test_single (convenience, this is just experimenting)
@@ -80,10 +74,52 @@ def decode_fp32(x):
 def main(mantissa, exponent):
     if exponent & 1 != 0:
         # shift mantissa up, subtract 1 from exp to compensate
-        return sqrt(mantissa << 1), (exponent - 1) >> 1
-    # mantissa as-is, no compensating needed on exp
-    return sqrt(mantissa), (exponent >> 1)
+        mantissa <<= 1
+        exponent -= 1
+    m, r = sqrt(mantissa)
+    return m, r, exponent >> 1
+
+
+#normalization function
+def normalise(s, m, e, lowbits):
+    if (lowbits >= 2):
+        m += 1
+    if get_mantissa(m) == ((1<<24)-1):
+        e += 1
+    return s, m, e
+
+
+def fsqrt_test(x):
+
+    xbits = x.bits
+    print ("x", x, type(x))
+    sq_test = x.sqrt()
+    print ("sqrt", sq_test)
+
+    print (xbits, type(xbits))
+    s, e, m = decode_fp32(xbits)
+    print("x decode", s, e, m, hex(m))
+
+    m |= 1<<23 # set top bit (the missing "1" from mantissa)
+    m <<= 27
 
+    sm, sr, se = main(m, e)
+    lowbits = sm & 0x3
+    sm >>= 2
+    sm = get_mantissa(sm)
+    #sm += 2
+
+    s, sm, se = normalise(s, sm, se, lowbits)
+
+    print("our  sqrt", s, se, sm, hex(sm), bin(sm), "lowbits", lowbits,
+                                                    "rem", hex(sr))
+    if lowbits >= 2:
+        print ("probably needs rounding (+1 on mantissa)")
+
+    sq_xbits = sq_test.bits
+    s, e, m = decode_fp32(sq_xbits)
+    print ("sf32 sqrt", s, e, m, hex(m), bin(m))
+    print ()
 
 if __name__ == '__main__':
 
@@ -91,29 +127,34 @@ if __name__ == '__main__':
     for Q in range(1, int(1e4)):
         print(Q, sqrt(Q), sqrtsimple(Q), int(Q**0.5))
         assert int(Q**0.5) == sqrtsimple(Q), "Q sqrtsimpl fail %d" % Q
-        assert int(Q**0.5) == sqrt(Q), "Q sqrt fail %d" % Q
+        assert int(Q**0.5) == sqrt(Q)[0], "Q sqrt fail %d" % Q
 
     # quick mantissa/exponent demo
     for e in range(26):
         for m in range(26):
-            ms, es = main(m, e)
-            print("m:%d e:%d sqrt: m:%d e:%d" % (m, e, ms, es))
+            ms, mr, es = main(m, e)
+            print("m:%d e:%d sqrt: m:%d-%d e:%d" % (m, e, ms, mr, es))
 
     x = Float32(1234.123456789)
-    xbits = x.bits
+    fsqrt_test(x)
+    x = Float32(32.1)
+    fsqrt_test(x)
+    x = Float32(16.0)
+    fsqrt_test(x)
+    x = Float32(8.0)
+    fsqrt_test(x)
+    x = Float32(8.5)
+    fsqrt_test(x)
+    x = Float32(3.14159265358979323)
+    fsqrt_test(x)
+    x = Float32(12.99392923123123)
+    fsqrt_test(x)
+    x = Float32(0.123456)
+    fsqrt_test(x)
+
 
-    print (x, type(x))
-    print (xbits, type(xbits))
-    s, e, m = decode_fp32(xbits)
-    print(s, e, m, hex(m))
 
-    se, sm = main(e, m)
-    print("our  sqrt", s, se, sm, hex(sm), bin(sm))
 
-    sq_test = x.sqrt()
-    sq_xbits = sq_test.bits
-    s, e, m = decode_fp32(sq_xbits)
-    print ("sf32 sqrt", s, e, m, hex(m), bin(m))
 """
 
 Notes: