Make asm implementations distinctly named objects, for instance MD5_IA32,

rather than silently replacing the C++ versions. Instead they are silently
replaced (currently, at least) at the lookup level: we switch off the set
of feature macros set to choose the best implementation in the current
build configuration. So you can have (and benchmark) MD5 and MD5_IA32
directly against each other in the same program with no hassles, but if
you ask for "MD5", you'll get maybe an MD5 or maybe MD5_IA32.

Also make the canonical asm names (which aren't guarded by C++ namespaces)
of the form botan_<algo>_<arch>_<func> as in botan_sha160_ia32_compress,
to avoid namespace collisions.

This change has another bonus that it should in many cases be possible to
derive the asm specializations directly from the original implementation,
saving some code (and of course logically SHA_160_IA32 is a SHA_160, just
one with a faster implementation of the compression function, so this seems
reasonable anyway).

4 files changed, 288 insertions, 0 deletions

diff --git a/src/utils/asm_amd64/asm_macr.h b/src/utils/asm_amd64/asm_macr.h
new file mode 100644
index 000000000..3cdd42dc6
--- /dev/null
+++ b/src/utils/asm_amd64/asm_macr.h
@@ -0,0 +1,125 @@
+/*************************************************
+* Assembly Macros Header File                    *
+* (C) 1999-2008 Jack Lloyd                       *
+*************************************************/
+
+#ifndef BOTAN_AMD64_ASM_MACROS_H__
+#define BOTAN_AMD64_ASM_MACROS_H__
+
+#ifdef __ELF__
+.section .note.GNU-stack,"",%progbits
+#endif
+
+/*************************************************
+* General/Global Macros                          *
+*************************************************/
+#define ALIGN .p2align 4,,15
+
+#define START_LISTING(FILENAME) \
+   .file #FILENAME;             \
+   .text;                       \
+   ALIGN;
+
+/*************************************************
+* Function Definitions                           *
+*************************************************/
+#define START_FUNCTION(func_name) \
+   ALIGN;                         \
+   .global  func_name;            \
+   .type    func_name,@function;  \
+func_name:
+
+#define END_FUNCTION(func_name) \
+   ret
+
+/*************************************************
+* Conditional Jumps                              *
+*************************************************/
+#define JUMP_IF_ZERO(REG, LABEL) \
+   cmp IMM(0), REG;              \
+   jz LABEL
+
+#define JUMP_IF_LT(REG, NUM, LABEL) \
+   cmp IMM(NUM), REG;               \
+   jl LABEL
+
+/*************************************************
+* Register Names                                 *
+*************************************************/
+#define R0  %rax
+#define R1  %rbx
+#define R2  %rcx
+#define R2_32  %ecx
+#define R3  %rdx
+#define R3_32  %edx
+#define R4  %rsp
+#define R5  %rbp
+#define R6  %rsi
+#define R6_32  %esi
+#define R7  %rdi
+#define R8  %r8
+#define R9  %r9
+#define R9_32  %r9d
+#define R10 %r10
+#define R11 %r11
+#define R12 %r12
+#define R13 %r13
+#define R14 %r14
+#define R15 %r15
+#define R16 %r16
+
+#define ARG_1 R7
+#define ARG_2 R6
+#define ARG_2_32 R6_32
+#define ARG_3 R3
+#define ARG_3_32 R3_32
+#define ARG_4 R2
+#define ARG_4_32 R2_32
+#define ARG_5 R8
+#define ARG_6 R9
+#define ARG_6_32 R9_32
+
+#define TEMP_1 R10
+#define TEMP_2 R11
+#define TEMP_3 ARG_6
+#define TEMP_4 ARG_5
+#define TEMP_5 ARG_4
+#define TEMP_5_32 ARG_4_32
+#define TEMP_6 ARG_3
+#define TEMP_7 ARG_2
+#define TEMP_8 ARG_1
+#define TEMP_9 R0
+
+/*************************************************
+* Memory Access Operations                       *
+*************************************************/
+#define ARRAY8(REG, NUM) 8*(NUM)(REG)
+#define ARRAY4(REG, NUM) 4*(NUM)(REG)
+
+#define ASSIGN(TO, FROM) mov FROM, TO
+
+/*************************************************
+* ALU Operations                                 *
+*************************************************/
+#define IMM(VAL) $VAL
+
+#define ADD(TO, FROM) add FROM, TO
+#define ADD_LAST_CARRY(REG) adc IMM(0), REG
+#define ADD_IMM(TO, NUM) ADD(TO, IMM(NUM))
+#define ADD_W_CARRY(TO1, TO2, FROM) add FROM, TO1; adc IMM(0), TO2;
+#define SUB_IMM(TO, NUM) sub IMM(NUM), TO
+#define MUL(REG) mul REG
+
+#define XOR(TO, FROM) xor FROM, TO
+#define AND(TO, FROM) and FROM, TO
+#define OR(TO, FROM) or FROM, TO
+#define NOT(REG) not REG
+#define ZEROIZE(REG) XOR(REG, REG)
+
+#define RETURN_VALUE_IS(V) ASSIGN(%rax, V)
+
+#define ROTL_IMM(REG, NUM) rol IMM(NUM), REG
+#define ROTR_IMM(REG, NUM) ror IMM(NUM), REG
+#define ADD3_IMM(TO, FROM, NUM) lea NUM(TO,FROM,1), TO
+
+#endif
diff --git a/src/utils/asm_amd64/info.txt b/src/utils/asm_amd64/info.txt
new file mode 100644
index 000000000..e190dba88
--- /dev/null
+++ b/src/utils/asm_amd64/info.txt
@@ -0,0 +1,16 @@
+realname "Assembler Macros (x86-64)"
+
+load_on required
+
+<add>
+asm_macr.h
+</add>
+
+<arch>
+amd64
+</arch>
+
+<cc>
+gcc
+icc
+</cc>
diff --git a/src/utils/asm_ia32/asm_macr.h b/src/utils/asm_ia32/asm_macr.h
new file mode 100644
index 000000000..6d5dbb59d
--- /dev/null
+++ b/src/utils/asm_ia32/asm_macr.h
@@ -0,0 +1,131 @@
+/*************************************************
+* Assembly Macros Header File                    *
+* (C) 1999-2008 Jack Lloyd                       *
+*************************************************/
+
+#ifndef BOTAN_IA32_ASM_MACROS_H__
+#define BOTAN_IA32_ASM_MACROS_H__
+
+#ifdef __ELF__
+.section .note.GNU-stack,"",%progbits
+#endif
+
+/*************************************************
+* General/Global Macros                          *
+*************************************************/
+#define ALIGN .p2align 4,,15
+
+#define START_LISTING(FILENAME) \
+   .file #FILENAME;             \
+   .text;                       \
+   .p2align 4,,15;
+
+#ifdef __ELF__
+.section .note.GNU-stack,"",%progbits
+#endif
+
+/*************************************************
+* Function Definitions                           *
+*************************************************/
+#define START_FUNCTION(func_name) \
+   .align   8;                    \
+   ALIGN;                         \
+   .global  func_name;            \
+   .type    func_name,@function;  \
+func_name:
+
+#define END_FUNCTION(func_name) \
+   ret
+
+/*************************************************
+* Loop Control                                   *
+*************************************************/
+#define START_LOOP(LABEL) \
+   ALIGN;                 \
+   LABEL##_LOOP:
+
+#define LOOP_UNTIL_EQ(REG, NUM, LABEL) \
+   cmpl IMM(NUM), REG;                 \
+   jne LABEL##_LOOP
+
+#define LOOP_UNTIL_LT(REG, NUM, LABEL) \
+   cmpl IMM(NUM), REG;                 \
+   jge LABEL##_LOOP
+
+/*************************************************
+ Conditional Jumps                              *
+*************************************************/
+#define JUMP_IF_ZERO(REG, LABEL) \
+   cmpl IMM(0), REG;             \
+   jz LABEL
+
+#define JUMP_IF_LT(REG, NUM, LABEL) \
+   cmpl IMM(NUM), REG;              \
+   jl LABEL
+
+/*************************************************
+* Register Names                                 *
+*************************************************/
+#define EAX %eax
+#define EBX %ebx
+#define ECX %ecx
+#define EDX %edx
+#define EBP %ebp
+#define EDI %edi
+#define ESI %esi
+#define ESP %esp
+
+/*************************************************
+* Memory Access Operations                       *
+*************************************************/
+#define ARRAY1(REG, NUM) (NUM)(REG)
+#define ARRAY4(REG, NUM) 4*(NUM)(REG)
+#define ARRAY4_INDIRECT(BASE, OFFSET, NUM) 4*(NUM)(BASE,OFFSET,4)
+#define ARG(NUM) 4*(PUSHED) + ARRAY4(ESP, NUM)
+
+#define ASSIGN(TO, FROM) movl FROM, TO
+#define ASSIGN_BYTE(TO, FROM) movzbl FROM, TO
+
+#define PUSH(REG) pushl REG
+#define POP(REG) popl REG
+
+#define SPILL_REGS() \
+   PUSH(EBP) ; \
+   PUSH(EDI) ; \
+   PUSH(ESI) ; \
+   PUSH(EBX)
+
+#define RESTORE_REGS() \
+   POP(EBX) ;  \
+   POP(ESI) ;  \
+   POP(EDI) ;  \
+   POP(EBP)
+
+/*************************************************
+* ALU Operations                                 *
+*************************************************/
+#define IMM(VAL) $VAL
+
+#define ADD(TO, FROM) addl FROM, TO
+#define ADD_IMM(TO, NUM) ADD(TO, IMM(NUM))
+#define ADD_W_CARRY(TO1, TO2, FROM) addl FROM, TO1; adcl IMM(0), TO2;
+#define SUB_IMM(TO, NUM) subl IMM(NUM), TO
+#define ADD2_IMM(TO, FROM, NUM) leal NUM(FROM), TO
+#define ADD3_IMM(TO, FROM, NUM) leal NUM(TO,FROM,1), TO
+#define MUL(REG) mull REG
+
+#define SHL_IMM(REG, SHIFT) shll IMM(SHIFT), REG
+#define SHR_IMM(REG, SHIFT) shrl IMM(SHIFT), REG
+#define SHL2_3(TO, FROM) leal 0(,FROM,8), TO
+
+#define XOR(TO, FROM) xorl FROM, TO
+#define AND(TO, FROM) andl FROM, TO
+#define OR(TO, FROM) orl FROM, TO
+#define NOT(REG) notl REG
+#define ZEROIZE(REG) XOR(REG, REG)
+
+#define ROTL_IMM(REG, NUM) roll IMM(NUM), REG
+#define ROTR_IMM(REG, NUM) rorl IMM(NUM), REG
+#define BSWAP(REG) bswapl REG
+
+#endif
diff --git a/src/utils/asm_ia32/info.txt b/src/utils/asm_ia32/info.txt
new file mode 100644
index 000000000..8f2ada914
--- /dev/null
+++ b/src/utils/asm_ia32/info.txt
@@ -0,0 +1,16 @@
+realname "Assembler Macros (IA-32)"
+
+load_on required
+
+<add>
+asm_macr.h
+</add>
+
+<arch>
+ia32
+</arch>
+
+<cc>
+gcc
+icc
+</cc>