/* Eight binary32 additions per short-vector instruction, in retained batches. */ .syntax unified .arch armv6 .fpu vfpv2 .text .align 2 .global shortvec_add .type shortvec_add, %function shortvec_add: cmp r3, #0 bxeq lr push {r4, r5, r6, lr} mov r5, r2 mov r6, r3 vmrs r4, fpscr vpush {d8-d15} bic r3, r4, #0x00370000 orr r3, r3, #0x00070000 vmsr fpscr, r3 .global shortvec_active_begin shortvec_active_begin: 1: vldmia r0!, {s8-s15} vldmia r1!, {s16-s23} vadd.f32 s24, s8, s16 vstmia r5!, {s24-s31} subs r6, r6, #1 bne 1b .global shortvec_active_end shortvec_active_end: vmsr fpscr, r4 vpop {d8-d15} pop {r4, r5, r6, pc} .size shortvec_add, .-shortvec_add .section .note.GNU-stack,"",%progbits