diff --git a/apps/codecs/demac/libdemac/vector_math16_armv5te.h b/apps/codecs/demac/libdemac/vector_math16_armv5te.h index 2940585a42..0a3679ce63 100644 --- a/apps/codecs/demac/libdemac/vector_math16_armv5te.h +++ b/apps/codecs/demac/libdemac/vector_math16_armv5te.h @@ -26,6 +26,13 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA #define FUSED_VECTOR_MATH +#define REPEAT_3(x) x x x +#if ORDER > 16 +#define REPEAT_MLA(x) x x x x x x x +#else +#define REPEAT_MLA(x) x x x +#endif + /* Calculate scalarproduct, then add a 2nd vector (fused for performance) * This version fetches data as 32 bit words, and *requires* v1 to be * 32 bit aligned. It also requires that f2 and s2 are either both 32 bit @@ -133,7 +140,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) ADDHALFREGS(r1, r2, r4) "stmia %[v1]!, {r0,r1} \n" - ".rept 3 \n" + REPEAT_3( "ldmia %[v1], {r1,r2} \n" "ldmia %[f2]!, {r3,r4} \n" "smlabb %[res], r1, r3, %[res] \n" @@ -144,7 +151,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) ADDHALFREGS(r0, r1, r3) ADDHALFREGS(r1, r2, r4) "stmia %[v1]!, {r0,r1} \n" - ".endr \n" + ) #if ORDER > 16 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" @@ -275,7 +282,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) SUBHALFREGS(r1, r2, r4) "stmia %[v1]!, {r0,r1} \n" - ".rept 3 \n" + REPEAT_3( "ldmia %[v1], {r1,r2} \n" "ldmia %[f2]!, {r3,r4} \n" "smlabb %[res], r1, r3, %[res] \n" @@ -286,7 +293,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) SUBHALFREGS(r0, r1, r3) SUBHALFREGS(r1, r2, r4) "stmia %[v1]!, {r0,r1} \n" - ".endr \n" + ) #if ORDER > 16 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" @@ -318,12 +325,6 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) int cnt = ORDER>>5; #endif -#if ORDER > 16 -#define MLA_BLOCKS "7" -#else -#define MLA_BLOCKS "3" -#endif - asm volatile ( #if ORDER > 32 "mov %[res], #0 \n" @@ -347,14 +348,14 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "smlabt %[res], r1, r2, %[res] \n" "smlatb %[res], r1, r3, %[res] \n" - ".rept " MLA_BLOCKS "\n" + REPEAT_MLA( "ldmia %[v1]!, {r0,r1} \n" "smlabt %[res], r0, r3, %[res] \n" "ldmia %[v2]!, {r2,r3} \n" "smlatb %[res], r0, r2, %[res] \n" "smlabt %[res], r1, r2, %[res] \n" "smlatb %[res], r1, r3, %[res] \n" - ".endr \n" + ) #if ORDER > 32 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" @@ -374,14 +375,14 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "smlabb %[res], r1, r3, %[res] \n" "smlatt %[res], r1, r3, %[res] \n" - ".rept " MLA_BLOCKS "\n" + REPEAT_MLA( "ldmia %[v1]!, {r0,r1} \n" "ldmia %[v2]!, {r2,r3} \n" "smlabb %[res], r0, r2, %[res] \n" "smlatt %[res], r0, r2, %[res] \n" "smlabb %[res], r1, r3, %[res] \n" "smlatt %[res], r1, r3, %[res] \n" - ".endr \n" + ) #if ORDER > 32 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" diff --git a/apps/codecs/demac/libdemac/vector_math16_armv6.h b/apps/codecs/demac/libdemac/vector_math16_armv6.h index 0ace6c5811..2ce62728cb 100644 --- a/apps/codecs/demac/libdemac/vector_math16_armv6.h +++ b/apps/codecs/demac/libdemac/vector_math16_armv6.h @@ -22,14 +22,14 @@ You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA -*/ +*/ #define FUSED_VECTOR_MATH #if ORDER > 16 -#define BLOCK_REPEAT "3" +#define REPEAT_BLOCK(x) x x x #else -#define BLOCK_REPEAT "1" +#define REPEAT_BLOCK(x) x #endif /* Calculate scalarproduct, then add a 2nd vector (fused for performance) @@ -77,7 +77,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "sadd16 r1, r1, r5 \n" "strd r0, [%[v1]], #8 \n" - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldmia %[s2]!, {r5,r6} \n" "pkhtb r4, r4, r2 \n" "pkhtb r2, r2, r3 \n" @@ -104,7 +104,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "sadd16 r0, r0, r6 \n" "sadd16 r1, r1, r5 \n" "strd r0, [%[v1]], #8 \n" - ".endr \n" + ) "ldmia %[s2]!, {r5,r6} \n" "pkhtb r4, r4, r2 \n" @@ -148,7 +148,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "sadd16 r1, r1, r7 \n" "strd r0, [%[v1]], #8 \n" - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "smlad %[res], r2, r4, %[res] \n" "ldrd r6, [%[s2]], #8 \n" "smlad %[res], r3, r5, %[res] \n" @@ -165,7 +165,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "sadd16 r0, r0, r6 \n" "sadd16 r1, r1, r7 \n" "strd r0, [%[v1]], #8 \n" - ".endr \n" + ) "smlad %[res], r2, r4, %[res] \n" "ldrd r6, [%[s2]], #8 \n" @@ -246,7 +246,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "ssub16 r1, r1, r5 \n" "strd r0, [%[v1]], #8 \n" - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldmia %[s2]!, {r5,r6} \n" "pkhtb r4, r4, r2 \n" "pkhtb r2, r2, r3 \n" @@ -273,7 +273,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "ssub16 r0, r0, r6 \n" "ssub16 r1, r1, r5 \n" "strd r0, [%[v1]], #8 \n" - ".endr \n" + ) "ldmia %[s2]!, {r5,r6} \n" "pkhtb r4, r4, r2 \n" @@ -317,7 +317,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "ssub16 r1, r1, r7 \n" "strd r0, [%[v1]], #8 \n" - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "smlad %[res], r2, r4, %[res] \n" "ldrd r6, [%[s2]], #8 \n" "smlad %[res], r3, r5, %[res] \n" @@ -334,7 +334,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "ssub16 r0, r0, r6 \n" "ssub16 r1, r1, r7 \n" "strd r0, [%[v1]], #8 \n" - ".endr \n" + ) "smlad %[res], r2, r4, %[res] \n" "ldrd r6, [%[s2]], #8 \n" @@ -400,7 +400,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) #else "smuadx %[res], r0, r3 \n" #endif - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "pkhtb r0, r6, r7 \n" "ldrd r2, [%[v1]], #8 \n" "smladx %[res], r1, r0, %[res] \n" @@ -413,8 +413,8 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "pkhtb r3, r5, r6 \n" "ldrd r4, [%[v2]], #8 \n" "smladx %[res], r0, r3, %[res] \n" - ".endr \n" - + ) + "pkhtb r0, r6, r7 \n" "ldrd r2, [%[v1]], #8 \n" "smladx %[res], r1, r0, %[res] \n" @@ -434,7 +434,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) #endif "b 99f \n" - + "20: \n" "ldrd r0, [%[v1]], #8 \n" "ldmia %[v2]!, {r5-r7} \n" @@ -446,7 +446,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) #else "smuad %[res], r0, r5 \n" #endif - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldrd r4, [%[v2]], #8 \n" "smlad %[res], r1, r6, %[res] \n" "ldrd r0, [%[v1]], #8 \n" @@ -455,7 +455,7 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "smlad %[res], r3, r4, %[res] \n" "ldrd r2, [%[v1]], #8 \n" "smlad %[res], r0, r5, %[res] \n" - ".endr \n" + ) #if ORDER > 32 "ldrd r4, [%[v2]], #8 \n" diff --git a/apps/codecs/demac/libdemac/vector_math16_cf.h b/apps/codecs/demac/libdemac/vector_math16_cf.h index 6e8216c9cc..4d77d3be31 100644 --- a/apps/codecs/demac/libdemac/vector_math16_cf.h +++ b/apps/codecs/demac/libdemac/vector_math16_cf.h @@ -28,6 +28,10 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA #define PREPARE_SCALARPRODUCT coldfire_set_macsr(0); /* signed integer mode */ +#define REPEAT_2(x) x x +#define REPEAT_3(x) x x x +#define REPEAT_7(x) x x x x x x x + /* Calculate scalarproduct, then add a 2nd vector (fused for performance) * This version fetches data as 32 bit words, and *recommends* v1 to be * 32 bit aligned. It also assumes that f2 and s2 are either both 32 bit @@ -64,7 +68,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "move.w (%[s2])+, %%d1 \n" "swap %%d1 \n" "1: \n" - ".rept 2 \n" + REPEAT_2( "movem.l (%[v1]), %%d6-%%d7/%%a0-%%a1 \n" "mac.w %%d0l, %%d6u, (%[f2])+, %%d0, %%acc0\n" "mac.w %%d0u, %%d6l, (%[s2])+, %%d2, %%acc0\n" @@ -82,7 +86,7 @@ static inline int32_t vector_sp_add(int16_t* v1, int16_t* f2, int16_t* s2) "move.l %%d6, (%[v1])+ \n" ADDHALFXREGS(%%a1, %%d1, %%d7) "move.l %%d7, (%[v1])+ \n" - ".endr \n" + ) #if ORDER > 16 "subq.l #1, %[res] \n" @@ -193,7 +197,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "move.w (%[s2])+, %%d1 \n" "swap %%d1 \n" "1: \n" - ".rept 2 \n" + REPEAT_2( "movem.l (%[v1]), %%d6-%%d7/%%a0-%%a1 \n" "mac.w %%d0l, %%d6u, (%[f2])+, %%d0, %%acc0\n" "mac.w %%d0u, %%d6l, (%[s2])+, %%d2, %%acc0\n" @@ -211,7 +215,7 @@ static inline int32_t vector_sp_sub(int16_t* v1, int16_t* f2, int16_t* s2) "move.l %%d6, (%[v1])+ \n" SUBHALFXREGS(%%a1, %%d1, %%d7) "move.l %%d7, (%[v1])+ \n" - ".endr \n" + ) #if ORDER > 16 "subq.l #1, %[res] \n" @@ -305,10 +309,10 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "move.l (%[v1])+, %%d0 \n" "move.w (%[v2])+, %%d1 \n" "1: \n" - ".rept 7 \n" + REPEAT_7( "mac.w %%d0u, %%d1l, (%[v2])+, %%d1, %%acc0\n" "mac.w %%d0l, %%d1u, (%[v1])+, %%d0, %%acc0\n" - ".endr \n" + ) "mac.w %%d0u, %%d1l, (%[v2])+, %%d1, %%acc0\n" #if ORDER > 16 @@ -324,12 +328,12 @@ static inline int32_t scalarproduct(int16_t* v1, int16_t* v2) "move.l (%[v1])+, %%d0 \n" "move.l (%[v2])+, %%d1 \n" "1: \n" - ".rept 3 \n" + REPEAT_3( "mac.w %%d0u, %%d1u, (%[v1])+, %%d2, %%acc0\n" "mac.w %%d0l, %%d1l, (%[v2])+, %%d1, %%acc0\n" "mac.w %%d2u, %%d1u, (%[v1])+, %%d0, %%acc0\n" "mac.w %%d2l, %%d1l, (%[v2])+, %%d1, %%acc0\n" - ".endr \n" + ) "mac.w %%d0u, %%d1u, (%[v1])+, %%d2, %%acc0\n" "mac.w %%d0l, %%d1l, (%[v2])+, %%d1, %%acc0\n" diff --git a/apps/codecs/demac/libdemac/vector_math32_armv4.h b/apps/codecs/demac/libdemac/vector_math32_armv4.h index 207fca3038..cb5fe9e0ee 100644 --- a/apps/codecs/demac/libdemac/vector_math32_armv4.h +++ b/apps/codecs/demac/libdemac/vector_math32_armv4.h @@ -27,11 +27,11 @@ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110, USA #define FUSED_VECTOR_MATH #if ORDER > 32 -#define BLOCK_REPEAT "8" +#define REPEAT_BLOCK(x) x x x x x x x x #elif ORDER > 16 -#define BLOCK_REPEAT "7" +#define REPEAT_BLOCK(x) x x x x x x x #else -#define BLOCK_REPEAT "3" +#define REPEAT_BLOCK(x) x x x #endif /* Calculate scalarproduct, then add a 2nd vector (fused for performance) */ @@ -60,7 +60,7 @@ static inline int32_t vector_sp_add(int32_t* v1, int32_t* f2, int32_t* s2) "add r3, r3, r7 \n" "stmia %[v1]!, {r0-r3} \n" #endif - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldmia %[v1], {r0-r3} \n" "ldmia %[f2]!, {r4-r7} \n" "mla %[res], r4, r0, %[res] \n" @@ -73,7 +73,7 @@ static inline int32_t vector_sp_add(int32_t* v1, int32_t* f2, int32_t* s2) "add r2, r2, r6 \n" "add r3, r3, r7 \n" "stmia %[v1]!, {r0-r3} \n" - ".endr \n" + ) #if ORDER > 32 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" @@ -120,7 +120,7 @@ static inline int32_t vector_sp_sub(int32_t* v1, int32_t* f2, int32_t* s2) "sub r3, r3, r7 \n" "stmia %[v1]!, {r0-r3} \n" #endif - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldmia %[v1], {r0-r3} \n" "ldmia %[f2]!, {r4-r7} \n" "mla %[res], r4, r0, %[res] \n" @@ -133,7 +133,7 @@ static inline int32_t vector_sp_sub(int32_t* v1, int32_t* f2, int32_t* s2) "sub r2, r2, r6 \n" "sub r3, r3, r7 \n" "stmia %[v1]!, {r0-r3} \n" - ".endr \n" + ) #if ORDER > 32 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n" @@ -173,14 +173,14 @@ static inline int32_t scalarproduct(int32_t* v1, int32_t* v2) "mla %[res], r6, r2, %[res] \n" "mla %[res], r7, r3, %[res] \n" #endif - ".rept " BLOCK_REPEAT "\n" + REPEAT_BLOCK( "ldmia %[v1]!, {r0-r3} \n" "ldmia %[v2]!, {r4-r7} \n" "mla %[res], r4, r0, %[res] \n" "mla %[res], r5, r1, %[res] \n" "mla %[res], r6, r2, %[res] \n" "mla %[res], r7, r3, %[res] \n" - ".endr \n" + ) #if ORDER > 32 "subs %[cnt], %[cnt], #1 \n" "bne 1b \n"