bitarray kernels can now be functions instead of macros with good per…

…formance define sort functions with @eval instead of macro export @timed and remove a couple exports add some disabled debug code for printing backtraces during bootstrap
kmsquire · Dec 17, 2012 · c6a3c85 · c6a3c85
1 parent 63dd471
commit c6a3c85
Show file tree

Hide file tree

Showing 5 changed files with 43 additions and 48 deletions.
diff --git a/base/bitarray.jl b/base/bitarray.jl
@@ -1173,17 +1173,14 @@ function rot180(A::BitMatrix)
     return B
 end
 
-# implemented as a macro to improve performance
-macro _jl_reverse_bits(dest, src)
-    quote
-        z    = $(esc(src))
-        z    = ((z >>>  1) & 0x5555555555555555) | ((z <<  1) & 0xaaaaaaaaaaaaaaaa)
-        z    = ((z >>>  2) & 0x3333333333333333) | ((z <<  2) & 0xcccccccccccccccc)
-        z    = ((z >>>  4) & 0x0f0f0f0f0f0f0f0f) | ((z <<  4) & 0xf0f0f0f0f0f0f0f0)
-        z    = ((z >>>  8) & 0x00ff00ff00ff00ff) | ((z <<  8) & 0xff00ff00ff00ff00)
-        z    = ((z >>> 16) & 0x0000ffff0000ffff) | ((z << 16) & 0xffff0000ffff0000)
-        $(esc(dest)) = ((z >>> 32) & 0x00000000ffffffff) | ((z << 32) & 0xffffffff00000000)
-    end
+function _jl_reverse_bits(src::Uint64)
+    z    = src
+    z    = ((z >>>  1) & 0x5555555555555555) | ((z <<  1) & 0xaaaaaaaaaaaaaaaa)
+    z    = ((z >>>  2) & 0x3333333333333333) | ((z <<  2) & 0xcccccccccccccccc)
+    z    = ((z >>>  4) & 0x0f0f0f0f0f0f0f0f) | ((z <<  4) & 0xf0f0f0f0f0f0f0f0)
+    z    = ((z >>>  8) & 0x00ff00ff00ff00ff) | ((z <<  8) & 0xff00ff00ff00ff00)
+    z    = ((z >>> 16) & 0x0000ffff0000ffff) | ((z << 16) & 0xffff0000ffff0000)
+    return ((z >>> 32) & 0x00000000ffffffff) | ((z << 32) & 0xffffffff00000000)
 end
 
 function reverse!(B::BitVector)
@@ -1198,9 +1195,9 @@ function reverse!(B::BitVector)
 
     for i = 1 : hnc
         j = ((i - 1) << 6)
-        @_jl_reverse_bits aux_chunks[1] B.chunks[i]
+        aux_chunks[1] = _jl_reverse_bits(B.chunks[i])
         _jl_copy_chunks(B.chunks, j+1, B.chunks, n-63-j, 64)
-        @_jl_reverse_bits B.chunks[i] B.chunks[i]
+        B.chunks[i] = _jl_reverse_bits(B.chunks[i])
         _jl_copy_chunks(B.chunks, n-63-j, aux_chunks, 1, 64)
     end
 
@@ -1213,7 +1210,7 @@ function reverse!(B::BitVector)
     l = (@_mod64 (n+63)) + 1
     msk = @_mskr l
 
-    @_jl_reverse_bits aux_chunks[1] (B.chunks[i] & msk)
+    aux_chunks[1] = _jl_reverse_bits(B.chunks[i] & msk)
     aux_chunks[1] >>>= (64 - l)
     _jl_copy_chunks(B.chunks, j+1, aux_chunks, 1, l)
 
@@ -1495,17 +1492,14 @@ transpose(B::BitVector) = reshape(copy(B), 1, length(B))
 
 # fast 8x8 bit transpose from Henry S. Warrens's "Hacker's Delight"
 # http://www.hackersdelight.org/HDcode/transpose8.c.txt
-# implemented as a macro to improve performance
-macro _jl_transpose8x8(x)
-    quote
-        y = $(esc(x))
-        t = (y $ (y >>> 7)) & 0x00aa00aa00aa00aa;
-        y = y $ t $ (t << 7)
-        t = (y $ (y >>> 14)) & 0x0000cccc0000cccc
-        y = y $ t $ (t << 14)
-        t = (y $ (y >>> 28)) & 0x00000000f0f0f0f0
-        $(esc(x)) = y $ t $ (t << 28)
-    end
+function _jl_transpose8x8(x::Uint64)
+    y = x
+    t = (y $ (y >>> 7)) & 0x00aa00aa00aa00aa
+    y = y $ t $ (t << 7)
+    t = (y $ (y >>> 14)) & 0x0000cccc0000cccc
+    y = y $ t $ (t << 14)
+    t = (y $ (y >>> 28)) & 0x00000000f0f0f0f0
+    return y $ t $ (t << 28)
 end
 
 function _jl_form_8x8_chunk(B::BitMatrix, i1::Int, i2::Int, m::Int, cgap::Int, cinc::Int, nc::Int, msk8::Uint64)
@@ -1571,7 +1565,7 @@ function transpose(B::BitMatrix)
 
         for j = 1 : 8 : l2
             x = _jl_form_8x8_chunk(B, i, j, l1, cgap1, cinc1, nc, msk8_1)
-            @_jl_transpose8x8 x
+            x = _jl_transpose8x8(x)
 
             msk8_2 = uint64(0xff)
             if (l2 < j + 7)

diff --git a/base/export.jl b/base/export.jl
@@ -1,7 +1,6 @@
 export
 # Modules
     Grisu,
-    Printf,
     PCRE,
     FFTW,
     DSP,
@@ -74,7 +73,6 @@ export
     Regex,
     RegexMatch,
     RegexMatchIterator,
-    Dimspec,
     RemoteRef,
     RepString,
     RevString,
@@ -1307,6 +1305,7 @@ export
     @cmd,
     @time,
     @elapsed,
+    @timed,
     @windows_only,
     @unix_only,
     @osx_only,

diff --git a/base/sort.jl b/base/sort.jl
@@ -11,14 +11,18 @@ _jl_fp_neg_le(x::Float32, y::Float32) = sle_int(unbox(Float32,y),unbox(Float32,x
 _jl_fp_neg_le(x::Float64, y::Float64) = sle_int(unbox(Float64,y),unbox(Float64,x))
 
 ## internal sorting functionality ##
-
-macro _jl_sort_functions(suffix, lt, args...)
-insertionsort = esc(symbol("_jl_insertionsort$suffix"))
-quicksort = esc(symbol("_jl_quicksort$suffix"))
-mergesort = esc(symbol("_jl_mergesort$suffix"))
-pivot_middle = esc(symbol("_jl_pivot_middle$suffix"))
-lt = @eval (a,b)->$lt
-quote
+for (suffix, lt, args) in (("",    (a,b)->:(isless($a,$b)), ()),
+                           ("_r",  (a,b)->:(isless($b,$a)), ()),
+                           ("_lt", (a,b)->:(lt($a,$b)), (:(lt::Function),)),
+                           ("_by", (a,b)->:(isless(by($a),by($b))), (:(by::Function),)),
+                           ## special sorting for floating-point arrays ##
+                           ("_fp_pos", (a,b)->:(_jl_fp_pos_lt($a,$b)), ()),
+                           ("_fp_neg", (a,b)->:(_jl_fp_neg_lt($a,$b)), ()))
+    insertionsort = symbol("_jl_insertionsort$suffix")
+    quicksort = symbol("_jl_quicksort$suffix")
+    mergesort = symbol("_jl_mergesort$suffix")
+    pivot_middle = symbol("_jl_pivot_middle$suffix")
+@eval begin
 
 # sorting should be stable
 # Thus, if a permutation is required, or records are being sorted
@@ -183,12 +187,7 @@ function ($mergesort)($(args...),
     return a, p
 end
 
-end; end # quote / macro
-
-@_jl_sort_functions ""    :(isless($a,$b))
-@_jl_sort_functions "_r"  :(isless($b,$a))
-@_jl_sort_functions "_lt" :(lt($a,$b)) lt::Function
-@_jl_sort_functions "_by" :(isless(by($a),by($b))) by::Function
+end; end # @eval / for
 
 ## external sorting functions ##
 
@@ -202,11 +201,6 @@ sort!{T}(lt::Function, a::AbstractVector{T}) =
 sort_by!{T}(by::Function, a::AbstractVector{T}) =
     _jl_mergesort_by(by, a, 1, length(a), Array(T,length(a)))
 
-## special sorting for floating-point arrays ##
-
-@_jl_sort_functions "_fp_pos" :(_jl_fp_pos_lt($a,$b))
-@_jl_sort_functions "_fp_neg" :(_jl_fp_neg_lt($a,$b))
-
 # push NaNs to the end of a, returning # of non-NaNs
 function _jl_nans_to_end{T<:FloatingPoint}(a::AbstractVector{T})
     n = length(a)

diff --git a/src/task.c b/src/task.c
@@ -501,6 +501,14 @@ static void record_backtrace(void)
     while (unw_step(&cursor) && n < MAX_BT_SIZE) {
         unw_get_reg(&cursor, UNW_REG_IP, &ip);
         bt_data[n++] = ip;
+        /*
+        char *func_name;
+        int line_num;
+        const char *file_name;
+        getFunctionInfo(&func_name, &line_num, &file_name, ip);
+        if (func_name != NULL)
+            ios_printf(ios_stdout, "in %s at %s:%d\n", func_name, file_name, line_num);
+        */
     }
     bt_size = n;
 }

diff --git a/src/toplevel.c b/src/toplevel.c
@@ -84,7 +84,7 @@ jl_value_t *jl_eval_module_expr(jl_expr_t *ex)
         if (table[i] != HT_NOTFOUND) {
             jl_binding_t *b = (jl_binding_t*)table[i];
             // remove non-exported macros
-            if (b->name->name[0]=='@' && !b->exportp)
+            if (b->name->name[0]=='@' && !b->exportp && b->owner==newm)
                 b->value = NULL;
             // error for unassigned exports
             /*