yuzu/mbedtls
1
0
Fork 0
mirror of https://github.com/yuzu-emu/mbedtls.git synced 2025-05-09 23:52:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix BN ASM for PowerPC

Browse source
This commit is contained in:
Manuel Pégourié-Gonnard 2014-01-07 19:16:48 +01:00
parent 8b1b103418
commit 02d800c151
1 changed files with 141 additions and 100 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

217
include/polarssl/bn_mul.h
View file

@ -279,62 +279,83 @@
#if defined(__MACH__) && defined(__APPLE__) #if defined(__MACH__) && defined(__APPLE__)
#define MULADDC_INIT \ #define MULADDC_INIT \
asm( "ld r3, %0 " :: "m" (s)); \ asm( \
asm( "ld r4, %0 " :: "m" (d)); \ " \
asm( "ld r5, %0 " :: "m" (c)); \ ld r3, %3; \
asm( "ld r6, %0 " :: "m" (b)); \ ld r4, %4; \
asm( "addi r3, r3, -8 " ); \ ld r5, %5; \
asm( "addi r4, r4, -8 " ); \ ld r6, %6; \
asm( "addic r5, r5, 0 " ); addi r3, r3, -8; \
addi r4, r4, -8; \
addic r5, r5, 0; \
"
#define MULADDC_CORE \ #define MULADDC_CORE \
asm( "ldu r7, 8(r3) " ); \ " \
asm( "mulld r8, r7, r6 " ); \ ldu r7, 8(r3); \
asm( "mulhdu r9, r7, r6 " ); \ mulld r8, r7, r6; \
asm( "adde r8, r8, r5 " ); \ mulhdu r9, r7, r6; \
asm( "ld r7, 8(r4) " ); \ adde r8, r8, r5; \
asm( "addze r5, r9 " ); \ ld r7, 8(r4); \
asm( "addc r8, r8, r7 " ); \ addze r5, r9; \
asm( "stdu r8, 8(r4) " ); addc r8, r8, r7; \
stdu r8, 8(r4); \
"
#define MULADDC_STOP \ #define MULADDC_STOP \
asm( "addze r5, r5 " ); \ " \
asm( "addi r4, r4, 8 " ); \ addze r5, r5; \
asm( "addi r3, r3, 8 " ); \ addi r4, r4, 8; \
asm( "std r5, %0 " : "=m" (c)); \ addi r3, r3, 8; \
asm( "std r4, %0 " : "=m" (d)); \ std r5, %0; \
asm( "std r3, %0 " : "=m" (s) :: \ std r4, %1; \
"r3", "r4", "r5", "r6", "r7", "r8", "r9" ); std r3, %2; \
" \
: "=m" (c), "=m" (d), "=m" (s) \
: "m" (s), "m" (d), "m" (c), "m" (b) \
: "r3", "r4", "r5", "r6", "r7", "r8", "r9" \
);
#else #else
#define MULADDC_INIT \ #define MULADDC_INIT \
asm( "ld %%r3, %0 " :: "m" (s)); \ asm( \
asm( "ld %%r4, %0 " :: "m" (d)); \ " \
asm( "ld %%r5, %0 " :: "m" (c)); \ ld %%r3, %3; \
asm( "ld %%r6, %0 " :: "m" (b)); \ ld %%r4, %4; \
asm( "addi %r3, %r3, -8 " ); \ ld %%r5, %5; \
asm( "addi %r4, %r4, -8 " ); \ ld %%r6, %6; \
asm( "addic %r5, %r5, 0 " ); addi %%r3, %%r3, -8; \
addi %%r4, %%r4, -8; \
addic %%r5, %%r5, 0; \
"
#define MULADDC_CORE \ #define MULADDC_CORE \
asm( "ldu %r7, 8(%r3) " ); \ " \
asm( "mulld %r8, %r7, %r6 " ); \ ldu %%r7, 8(%%r3); \
asm( "mulhdu %r9, %r7, %r6 " ); \ mulld %%r8, %%r7, %%r6; \
asm( "adde %r8, %r8, %r5 " ); \ mulhdu %%r9, %%r7, %%r6; \
asm( "ld %r7, 8(%r4) " ); \ adde %%r8, %%r8, %%r5; \
asm( "addze %r5, %r9 " ); \ ld %%r7, 8(%%r4); \
asm( "addc %r8, %r8, %r7 " ); \ addze %%r5, %%r9; \
asm( "stdu %r8, 8(%r4) " ); addc %%r8, %%r8, %%r7; \
stdu %%r8, 8(%%r4); \
"
#define MULADDC_STOP \ #define MULADDC_STOP \
asm( "addze %r5, %r5 " ); \ " \
asm( "addi %r4, %r4, 8 " ); \ addze %%r5, %%r5; \
asm( "addi %r3, %r3, 8 " ); \ addi %%r4, %%r4, 8; \
asm( "std %%r5, %0 " : "=m" (c)); \ addi %%r3, %%r3, 8; \
asm( "std %%r4, %0 " : "=m" (d)); \ std %%r5, %0; \
asm( "std %%r3, %0 " : "=m" (s) :: \ std %%r4, %1; \
"r3", "r4", "r5", "r6", "r7", "r8", "r9" ); std %%r3, %2; \
" \
: "=m" (c), "=m" (d), "=m" (s) \
: "m" (s), "m" (d), "m" (c), "m" (b) \
: "r3", "r4", "r5", "r6", "r7", "r8", "r9" \
);
#endif #endif
@ -343,62 +364,82 @@
#if defined(__MACH__) && defined(__APPLE__) #if defined(__MACH__) && defined(__APPLE__)
#define MULADDC_INIT \ #define MULADDC_INIT \
asm( "lwz r3, %0 " :: "m" (s)); \ asm( \
asm( "lwz r4, %0 " :: "m" (d)); \ " \
asm( "lwz r5, %0 " :: "m" (c)); \ lwz r3, %3; \
asm( "lwz r6, %0 " :: "m" (b)); \ lwz r4, %4; \
asm( "addi r3, r3, -4 " ); \ lwz r5, %5; \
asm( "addi r4, r4, -4 " ); \ lwz r6, %6; \
asm( "addic r5, r5, 0 " ); addi r3, r3, -4; \
addi r4, r4, -4; \
addic r5, r5, 0; \
"
#define MULADDC_CORE \ #define MULADDC_CORE \
asm( "lwzu r7, 4(r3) " ); \ " \
asm( "mullw r8, r7, r6 " ); \ lwzu r7, 4(r3); \
asm( "mulhwu r9, r7, r6 " ); \ mullw r8, r7, r6; \
asm( "adde r8, r8, r5 " ); \ mulhwu r9, r7, r6; \
asm( "lwz r7, 4(r4) " ); \ adde r8, r8, r5; \
asm( "addze r5, r9 " ); \ lwz r7, 4(r4); \
asm( "addc r8, r8, r7 " ); \ addze r5, r9; \
asm( "stwu r8, 4(r4) " ); addc r8, r8, r7; \
stwu r8, 4(r4); \
"
#define MULADDC_STOP \ #define MULADDC_STOP \
asm( "addze r5, r5 " ); \ " \
asm( "addi r4, r4, 4 " ); \ addze r5, r5; \
asm( "addi r3, r3, 4 " ); \ addi r4, r4, 4; \
asm( "stw r5, %0 " : "=m" (c)); \ addi r3, r3, 4; \
asm( "stw r4, %0 " : "=m" (d)); \ stw r5, %0; \
asm( "stw r3, %0 " : "=m" (s) :: \ stw r4, %1; \
"r3", "r4", "r5", "r6", "r7", "r8", "r9" ); stw r3, %2; \
" \
: "=m" (c), "=m" (d), "=m" (s) \
: "m" (s), "m" (d), "m" (c), "m" (b) \
: "r3", "r4", "r5", "r6", "r7", "r8", "r9" \
);
#else #else
#define MULADDC_INIT \ #define MULADDC_INIT \
asm( "lwz %%r3, %0 " :: "m" (s)); \ asm( \
asm( "lwz %%r4, %0 " :: "m" (d)); \ " \
asm( "lwz %%r5, %0 " :: "m" (c)); \ lwz %%r3, %3; \
asm( "lwz %%r6, %0 " :: "m" (b)); \ lwz %%r4, %4; \
asm( "addi %r3, %r3, -4 " ); \ lwz %%r5, %5; \
asm( "addi %r4, %r4, -4 " ); \ lwz %%r6, %6; \
asm( "addic %r5, %r5, 0 " ); addi %%r3, %%r3, -4; \
addi %%r4, %%r4, -4; \
addic %%r5, %%r5, 0; \
"
#define MULADDC_CORE \ #define MULADDC_CORE \
asm( "lwzu %r7, 4(%r3) " ); \ " \
asm( "mullw %r8, %r7, %r6 " ); \ lwzu %%r7, 4(%%r3); \
asm( "mulhwu %r9, %r7, %r6 " ); \ mullw %%r8, %%r7, %%r6; \
asm( "adde %r8, %r8, %r5 " ); \ mulhwu %%r9, %%r7, %%r6; \
asm( "lwz %r7, 4(%r4) " ); \ adde %%r8, %%r8, %%r5; \
asm( "addze %r5, %r9 " ); \ lwz %%r7, 4(%%r4); \
asm( "addc %r8, %r8, %r7 " ); \ addze %%r5, %%r9; \
asm( "stwu %r8, 4(%r4) " ); addc %%r8, %%r8, %%r7; \
stwu %%r8, 4(%%r4); \
"
#define MULADDC_STOP \ #define MULADDC_STOP \
asm( "addze %r5, %r5 " ); \ " \
asm( "addi %r4, %r4, 4 " ); \ addze %%r5, %%r5; \
asm( "addi %r3, %r3, 4 " ); \ addi %%r4, %%r4, 4; \
asm( "stw %%r5, %0 " : "=m" (c)); \ addi %%r3, %%r3, 4; \
asm( "stw %%r4, %0 " : "=m" (d)); \ stw %%r5, %0; \
asm( "stw %%r3, %0 " : "=m" (s) :: \ stw %%r4, %1; \
"r3", "r4", "r5", "r6", "r7", "r8", "r9" ); stw %%r3, %2; \
" \
: "=m" (c), "=m" (d), "=m" (s) \
: "m" (s), "m" (d), "m" (c), "m" (b) \
: "r3", "r4", "r5", "r6", "r7", "r8", "r9" \
);
#endif #endif