de-tabify; also speculate 1-byte loads to save a cycle or two

amd_d_nrv2b.S amd_d_nrv2e.S l_lx_elf64amd.S

committer: jreiser <jreiser> 1131943590 +0000

src/stub/amd_d_nrv2b.S

@@ -31,16 +31,16 @@
 
         ALIGN(1<<3)
 lit_n2b:
-	movb (%rsi),%al; addq $1,%rsi
-	movb %al,(%rdi); addq $1,%rdi
+        incq %rsi; movb %dl,(%rdi)
+        incq %rdi
 top_n2b:
+        movzbl (%rsi),%edx  # speculate: literal, or bottom 8 bits of offset
         jnextb1y lit_n2b
         lea 1(lenq),off  # [len= 0] off= 1
 offmore_n2b:
         getnextb(off)
         jnextb0n offmore_n2b
 
-	movzbl (%rsi),%edx
         subl $ 3,off; jc len_n2b  # use previous offset
         shll $ 8,off
         orl %edx,off; incq %rsi
@@ -61,3 +61,8 @@ gotlen_n2b:
         call copy
 bot_n2b:  # In: 0==len
         jmp top_n2b
+
+/*
+vi:ts=8:et:nowrap
+*/
+

src/stub/amd_d_nrv2e.S

@@ -31,9 +31,10 @@
 
         ALIGN(1<<3)
 lit_n2e:
-	movb (%rsi),%al; addq $1,%rsi
-	movb %al,(%rdi); addq $1,%rdi
+        incq %rsi; movb %dl,(%rdi)
+        incq %rdi
 top_n2e:
+        movzbl (%rsi),%edx  # speculate: literal, or bottom 8 bits of offset
         jnextb1y lit_n2e
         lea 1(lenq),off  # [len= 0] off= 1
         jmp getoff_n2e
@@ -45,7 +46,6 @@ getoff_n2e:
         getnextb(off)
         jnextb0n off_n2e
 
-	movzbl (%rsi),%edx  #; xorl len,len  # len= 0
         subl $ 3,off; jc offprev_n2e
         shll $ 8,off
         orl %edx,off; incq %rsi
@@ -74,3 +74,8 @@ gotlen_n2e:
         call copy
 bot_n2e:  # In: 0==len
         jmp top_n2e
+
+/*
+vi:ts=8:et:nowrap
+*/
+

src/stub/l_lx_elf64amd.S

@@ -110,20 +110,22 @@ getbit:
 refill:
         movl (%rsi),bits; subq $-4,%rsi  # next 32 bits; set Carry
         adcl bits,bits  # LSB= 1 (CarryIn); CarryOut= next bit
+        movzbl (%rsi),%edx  # speculate: literal, or bottom 8 bits of offset
         rep; ret
 
 copy:  # In: len, %rdi, disp;  Out: 0==len, %rdi, disp;  trashes %rax, %rdx
-	leaq (%rdi,disp),%rdx
+        leaq (%rdi,disp),%rax; movb (%rax),%dl
         cmpl $ 3,len;  jbe copy1  # perhaps extend this to length 5 or less?
         cmpq $-4,disp; ja  copy1  # 4-byte chunks would overlap
         subl $4,len  # adjust for termination cases
 copy4:
-	movl (%rdx),%eax; leaq 4(%rdx),%rdx; subl $4,len
-	movl %eax,(%rdi); leaq 4(%rdi),%rdi; jnc copy4
-	addl $4,len; jz copy0
+        movl (%rax),%edx; addq $4,      %rax; subl $4,len
+        movl %edx,(%rdi); leaq  4(%rdi),%rdi; jnc copy4
+        addl $4,len; movb (%rax),%dl; jz copy0
 copy1:
-	movb (%rdx), %al; leaq 1(%rdx),%rdx; subl $1,len
-	movb  %al,(%rdi); leaq 1(%rdi),%rdi; jnz copy1
+        incq %rax; movb %dl,(%rdi); subl $1,len
+                   movb (%rax),%dl
+        leaq 1(%rdi),%rdi;          jnz copy1
 copy0:
         rep; ret