Skip to main content
  • Home
  • Development
  • Documentation
  • Donate
  • Operational login
  • Browse the archive

swh logo
SoftwareHeritage
Software
Heritage
Archive
Features
  • Search

  • Downloads

  • Save code now

  • Add forge now

  • Help

  • a3483c2
  • /
  • ntruhrss701
  • /
  • m4f
  • /
  • __jump32divsteps_mod3.S
Raw File Download
Permalinks

To reference or cite the objects present in the Software Heritage archive, permalinks based on SoftWare Hash IDentifiers (SWHIDs) must be used.
Select below a type of object currently browsed in order to display its associated SWHID and permalink.

  • content
  • directory
content badge Iframe embedding
swh:1:cnt:c9d9614234de92c2cb1a1c8f216282cbfa7424c8
directory badge Iframe embedding
swh:1:dir:a6444f08b9a843ba17dc39e96e03d8d90295f559
Citations

This interface enables to generate software citations, provided that the root directory of browsed objects contains a citation.cff or codemeta.json file.
Select below a type of object currently browsed in order to generate citations for them.

  • content
  • directory
Generate software citation in BibTex format (requires biblatex-software package)
Generating citation ...
Generate software citation in BibTex format (requires biblatex-software package)
Generating citation ...
__jump32divsteps_mod3.S
// void __gf_polymul_16x16_2x2_x2p2_mod3 (int *V, int *M, int *fh, int *gh);
.p2align 2,,3
.syntax unified
.text
.global __gf_polymul_16x16_2x2_x2p2_mod3
.type __gf_polymul_16x16_2x2_x2p2_mod3, %function
__gf_polymul_16x16_2x2_x2p2_mod3:
	push.w {r4-r12,lr}
	vpush.w {s16-s17}
	vmov.w s0, s1, r2, r3
	add.w r1, #32
	mov.w r6, #0
	ldr.w r5, [r1], #4
	ldr.w r4, [r2, #12]
	ldr.w lr, [r2, #8]
	ldr.w r12, [r2, #4]
	ldr.w r11, [r2], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r2, #-28]
	ldr.w lr, [r2, #-24]
	ldr.w r4, [r2, #-20]
	ldr.w r9, [r1, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r1, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r1, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r2, r3
	mov.w r6, #0
	ldr.w r5, [r1], #4
	ldr.w r4, [r2, #12]
	ldr.w lr, [r2, #8]
	ldr.w r12, [r2, #4]
	ldr.w r11, [r2], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r2, #-28]
	ldr.w lr, [r2, #-24]
	ldr.w r4, [r2, #-20]
	ldr.w r9, [r1, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r1, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r1, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r2, r3, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r2
	add.w r11, r7, r3
	add.w r12, r8, r4
	add.w lr, r9, r5
	ubfx.w r2, r12, #24, #8
	add.w lr, r2, lr, LSL #8
	ubfx.w r2, r11, #24, #8
	add.w r12, r2, r12, LSL #8
	ubfx.w r2, r10, #24, #8
	add.w r11, r2, r11, LSL #8
	vmov.w r2, r3, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r2, r6
	add.w r3, r7
	add.w r4, r8
	add.w r5, r9
	ubfx.w r6, r5, #24, #8
	add.w r10, r6, r10, LSL #8
	ldr.w r6, [r1, #-64]
	ldr.w r7, [r1, #-60]
	ldr.w r8, [r1, #-56]
	ldr.w r9, [r1, #-52]
	add.w r6, r6, r2, LSL #8
	add.w r7, r7, r3, LSL #8
	add.w r7, r7, r2, LSR #24
	add.w r8, r8, r4, LSL #8
	add.w r8, r8, r3, LSR #24
	add.w r9, r9, r5, LSL #8
	add.w r9, r9, r4, LSR #24
	mov.w r4, #0x03030303
	bic.w r5, r6, r4 // top 3b < 8
	and.w r6, r6, r4 // bot 2b < 4
	add.w r6, r6, r5, LSR #2 // range <=10
	bic.w r5, r6, r4 // top 3b < 3
	and.w r6, r6, r4 // bot 2b < 4
	add.w r6, r6, r5, LSR #2 // range <=5
	usub8.w r5, r6, r4 // >= 3 ?
	sel.w r6, r5, r6 // select
	and.w r5, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r5, LSR #4 // range < 31
	bic.w r5, r7, r4 // top 3b < 8
	and.w r7, r7, r4 // bot 2b < 4
	add.w r7, r7, r5, LSR #2 // range <=10
	bic.w r5, r7, r4 // top 3b < 3
	and.w r7, r7, r4 // bot 2b < 4
	add.w r7, r7, r5, LSR #2 // range <=5
	usub8.w r5, r7, r4 // >= 3 ?
	sel.w r7, r5, r7 // select
	and.w r5, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r5, LSR #4 // range < 31
	bic.w r5, r8, r4 // top 3b < 8
	and.w r8, r8, r4 // bot 2b < 4
	add.w r8, r8, r5, LSR #2 // range <=10
	bic.w r5, r8, r4 // top 3b < 3
	and.w r8, r8, r4 // bot 2b < 4
	add.w r8, r8, r5, LSR #2 // range <=5
	usub8.w r5, r8, r4 // >= 3 ?
	sel.w r8, r5, r8 // select
	and.w r5, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r5, LSR #4 // range < 31
	bic.w r5, r9, r4 // top 3b < 8
	and.w r9, r9, r4 // bot 2b < 4
	add.w r9, r9, r5, LSR #2 // range <=10
	bic.w r5, r9, r4 // top 3b < 3
	and.w r9, r9, r4 // bot 2b < 4
	add.w r9, r9, r5, LSR #2 // range <=5
	usub8.w r5, r9, r4 // >= 3 ?
	sel.w r9, r5, r9 // select
	and.w r5, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r5, LSR #4 // range < 31
	bic.w r5, r10, r4 // top 3b < 8
	and.w r10, r10, r4 // bot 2b < 4
	add.w r10, r10, r5, LSR #2 // range <=10
	bic.w r5, r10, r4 // top 3b < 3
	and.w r10, r10, r4 // bot 2b < 4
	add.w r10, r10, r5, LSR #2 // range <=5
	usub8.w r5, r10, r4 // >= 3 ?
	sel.w r10, r5, r10 // select
	and.w r5, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r5, LSR #4 // range < 31
	bic.w r5, r11, r4 // top 3b < 8
	and.w r11, r11, r4 // bot 2b < 4
	add.w r11, r11, r5, LSR #2 // range <=10
	bic.w r5, r11, r4 // top 3b < 3
	and.w r11, r11, r4 // bot 2b < 4
	add.w r11, r11, r5, LSR #2 // range <=5
	usub8.w r5, r11, r4 // >= 3 ?
	sel.w r11, r5, r11 // select
	and.w r5, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r5, LSR #4 // range < 31
	bic.w r5, r12, r4 // top 3b < 8
	and.w r12, r12, r4 // bot 2b < 4
	add.w r12, r12, r5, LSR #2 // range <=10
	bic.w r5, r12, r4 // top 3b < 3
	and.w r12, r12, r4 // bot 2b < 4
	add.w r12, r12, r5, LSR #2 // range <=5
	usub8.w r5, r12, r4 // >= 3 ?
	sel.w r12, r5, r12 // select
	bic.w r5, lr, r4 // top 3b < 8
	and.w lr, lr, r4 // bot 2b < 4
	add.w lr, lr, r5, LSR #2 // range <=10
	bic.w r5, lr, r4 // top 3b < 3
	and.w lr, lr, r4 // bot 2b < 4
	add.w lr, lr, r5, LSR #2 // range <=5
	usub8.w r5, lr, r4 // >= 3 ?
	sel.w lr, r5, lr // select
	str.w r7, [r0, #4]
	str.w r8, [r0, #8]
	str.w r9, [r0, #12]
	str.w r10, [r0, #16]
	str.w r11, [r0, #20]
	str.w r12, [r0, #24]
	str.w lr, [r0, #28]
	str.w r6, [r0], #32
	vmov.w r2, r3, s0, s1
	mov.w r6, #0
	ldr.w r5, [r1], #4
	ldr.w r4, [r2, #12]
	ldr.w lr, [r2, #8]
	ldr.w r12, [r2, #4]
	ldr.w r11, [r2], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r2, #-28]
	ldr.w lr, [r2, #-24]
	ldr.w r4, [r2, #-20]
	ldr.w r9, [r1, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r1, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r1, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r2, r3
	mov.w r6, #0
	ldr.w r5, [r1], #4
	ldr.w r4, [r2, #12]
	ldr.w lr, [r2, #8]
	ldr.w r12, [r2, #4]
	ldr.w r11, [r2], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r1], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r2, #-28]
	ldr.w lr, [r2, #-24]
	ldr.w r4, [r2, #-20]
	ldr.w r9, [r1, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r1, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r1, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r2, r3, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r2
	add.w r11, r7, r3
	add.w r12, r8, r4
	add.w lr, r9, r5
	vmov.w r2, r3, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r6, r6, r2
	add.w r7, r7, r3
	add.w r8, r8, r4
	add.w r9, r9, r5
	ldr.w r2, [r1, #-80]
	ldr.w r3, [r1, #-76]
	ldr.w r4, [r1, #-72]
	ldr.w r5, [r1, #-68]
	add.w r6, r2
	add.w r7, r3
	add.w r8, r4
	add.w r9, r5
	mov.w r4, #0x03030303
	bic.w r5, r6, r4 // top 3b < 8
	and.w r6, r6, r4 // bot 2b < 4
	add.w r6, r6, r5, LSR #2 // range <=10
	bic.w r5, r6, r4 // top 3b < 3
	and.w r6, r6, r4 // bot 2b < 4
	add.w r6, r6, r5, LSR #2 // range <=5
	usub8.w r5, r6, r4 // >= 3 ?
	sel.w r6, r5, r6 // select
	and.w r5, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r5, LSR #4 // range < 31
	bic.w r5, r7, r4 // top 3b < 8
	and.w r7, r7, r4 // bot 2b < 4
	add.w r7, r7, r5, LSR #2 // range <=10
	bic.w r5, r7, r4 // top 3b < 3
	and.w r7, r7, r4 // bot 2b < 4
	add.w r7, r7, r5, LSR #2 // range <=5
	usub8.w r5, r7, r4 // >= 3 ?
	sel.w r7, r5, r7 // select
	and.w r5, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r5, LSR #4 // range < 31
	bic.w r5, r8, r4 // top 3b < 8
	and.w r8, r8, r4 // bot 2b < 4
	add.w r8, r8, r5, LSR #2 // range <=10
	bic.w r5, r8, r4 // top 3b < 3
	and.w r8, r8, r4 // bot 2b < 4
	add.w r8, r8, r5, LSR #2 // range <=5
	usub8.w r5, r8, r4 // >= 3 ?
	sel.w r8, r5, r8 // select
	and.w r5, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r5, LSR #4 // range < 31
	bic.w r5, r9, r4 // top 3b < 8
	and.w r9, r9, r4 // bot 2b < 4
	add.w r9, r9, r5, LSR #2 // range <=10
	bic.w r5, r9, r4 // top 3b < 3
	and.w r9, r9, r4 // bot 2b < 4
	add.w r9, r9, r5, LSR #2 // range <=5
	usub8.w r5, r9, r4 // >= 3 ?
	sel.w r9, r5, r9 // select
	and.w r5, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r5, LSR #4 // range < 31
	bic.w r5, r10, r4 // top 3b < 8
	and.w r10, r10, r4 // bot 2b < 4
	add.w r10, r10, r5, LSR #2 // range <=10
	bic.w r5, r10, r4 // top 3b < 3
	and.w r10, r10, r4 // bot 2b < 4
	add.w r10, r10, r5, LSR #2 // range <=5
	usub8.w r5, r10, r4 // >= 3 ?
	sel.w r10, r5, r10 // select
	and.w r5, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r5, LSR #4 // range < 31
	bic.w r5, r11, r4 // top 3b < 8
	and.w r11, r11, r4 // bot 2b < 4
	add.w r11, r11, r5, LSR #2 // range <=10
	bic.w r5, r11, r4 // top 3b < 3
	and.w r11, r11, r4 // bot 2b < 4
	add.w r11, r11, r5, LSR #2 // range <=5
	usub8.w r5, r11, r4 // >= 3 ?
	sel.w r11, r5, r11 // select
	and.w r5, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r5, LSR #4 // range < 31
	bic.w r5, r12, r4 // top 3b < 8
	and.w r12, r12, r4 // bot 2b < 4
	add.w r12, r12, r5, LSR #2 // range <=10
	bic.w r5, r12, r4 // top 3b < 3
	and.w r12, r12, r4 // bot 2b < 4
	add.w r12, r12, r5, LSR #2 // range <=5
	usub8.w r5, r12, r4 // >= 3 ?
	sel.w r12, r5, r12 // select
	bic.w r5, lr, r4 // top 3b < 8
	and.w lr, lr, r4 // bot 2b < 4
	add.w lr, lr, r5, LSR #2 // range <=10
	bic.w r5, lr, r4 // top 3b < 3
	and.w lr, lr, r4 // bot 2b < 4
	add.w lr, lr, r5, LSR #2 // range <=5
	usub8.w r5, lr, r4 // >= 3 ?
	sel.w lr, r5, lr // select
	str.w r7, [r0, #4]
	str.w r8, [r0, #8]
	str.w r9, [r0, #12]
	str.w r10, [r0, #16]
	str.w r11, [r0, #20]
	str.w r12, [r0, #24]
	str.w lr, [r0, #28]
	str.w r6, [r0], #32
	vpop.w {s16-s17}
	pop.w {r4-r12,pc}
// void __gf_polymul_16x16_2x2_x_2x2_mod3 (int *M,int *M1,int *M2);
.p2align 2,,3
.syntax unified
.text
.global __gf_polymul_16x16_2x2_x_2x2_mod3
.type __gf_polymul_16x16_2x2_x_2x2_mod3, %function
__gf_polymul_16x16_2x2_x_2x2_mod3:
	push.w {r3-r12,lr}
	vpush.w {s16-s17}
	mov.w r3, #0x03030303
	vmov.w s0, s1, r1, r2
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r1, r2, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r1, LSL #8
	add.w r11, r7, r2, LSL #8
	add.w r12, r8, r4, LSL #8
	add.w lr, r9, r5, LSL #8
	add.w r11, r11, r1, LSR #24
	add.w r12, r12, r2, LSR #24
	add.w lr, lr, r4, LSR #24
	vmov.w r1, r2, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r10, r10, r5, LSR #24
	add.w r6, r6, r1, LSL #8
	add.w r7, r7, r2, LSL #8
	add.w r8, r8, r4, LSL #8
	add.w r9, r9, r5, LSL #8
	add.w r7, r7, r1, LSR #24
	add.w r8, r8, r2, LSR #24
	add.w r9, r9, r4, LSR #24
	bic.w r2, r6, r3 // top 3b < 8
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=10
	bic.w r2, r6, r3 // top 3b < 3
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=5
	usub8.w r2, r6, r3 // >= 3 ?
	sel.w r6, r2, r6 // select
	and.w r2, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r2, LSR #4 // range < 31
	bic.w r2, r7, r3 // top 3b < 8
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=10
	bic.w r2, r7, r3 // top 3b < 3
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=5
	usub8.w r2, r7, r3 // >= 3 ?
	sel.w r7, r2, r7 // select
	and.w r2, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r2, LSR #4 // range < 31
	bic.w r2, r8, r3 // top 3b < 8
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=10
	bic.w r2, r8, r3 // top 3b < 3
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=5
	usub8.w r2, r8, r3 // >= 3 ?
	sel.w r8, r2, r8 // select
	and.w r2, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r2, LSR #4 // range < 31
	bic.w r2, r9, r3 // top 3b < 8
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=10
	bic.w r2, r9, r3 // top 3b < 3
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=5
	usub8.w r2, r9, r3 // >= 3 ?
	sel.w r9, r2, r9 // select
	and.w r2, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r2, LSR #4 // range < 31
	bic.w r2, r10, r3 // top 3b < 8
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=10
	bic.w r2, r10, r3 // top 3b < 3
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=5
	usub8.w r2, r10, r3 // >= 3 ?
	sel.w r10, r2, r10 // select
	and.w r2, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r2, LSR #4 // range < 31
	bic.w r2, r11, r3 // top 3b < 8
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=10
	bic.w r2, r11, r3 // top 3b < 3
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=5
	usub8.w r2, r11, r3 // >= 3 ?
	sel.w r11, r2, r11 // select
	and.w r2, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r2, LSR #4 // range < 31
	bic.w r2, r12, r3 // top 3b < 8
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=10
	bic.w r2, r12, r3 // top 3b < 3
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=5
	usub8.w r2, r12, r3 // >= 3 ?
	sel.w r12, r2, r12 // select
	bic.w r2, lr, r3 // top 3b < 8
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=10
	bic.w r2, lr, r3 // top 3b < 3
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=5
	usub8.w r2, lr, r3 // >= 3 ?
	sel.w lr, r2, lr // select
	str r7, [r0, #4]
	str r8, [r0, #8]
	str r9, [r0, #12]
	str r10, [r0, #16]
	str r11, [r0, #20]
	str r12, [r0, #24]
	str lr, [r0, #28]
	str r6, [r0], #32
	vmov.w r1, r2, s0, s1
	add.w r1, #16
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r1, r2, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r1, LSL #8
	add.w r11, r7, r2, LSL #8
	add.w r12, r8, r4, LSL #8
	add.w lr, r9, r5, LSL #8
	add.w r11, r11, r1, LSR #24
	add.w r12, r12, r2, LSR #24
	add.w lr, lr, r4, LSR #24
	vmov.w r1, r2, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r10, r10, r5, LSR #24
	add.w r6, r6, r1, LSL #8
	add.w r7, r7, r2, LSL #8
	add.w r8, r8, r4, LSL #8
	add.w r9, r9, r5, LSL #8
	add.w r7, r7, r1, LSR #24
	add.w r8, r8, r2, LSR #24
	add.w r9, r9, r4, LSR #24
	bic.w r2, r6, r3 // top 3b < 8
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=10
	bic.w r2, r6, r3 // top 3b < 3
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=5
	usub8.w r2, r6, r3 // >= 3 ?
	sel.w r6, r2, r6 // select
	and.w r2, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r2, LSR #4 // range < 31
	bic.w r2, r7, r3 // top 3b < 8
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=10
	bic.w r2, r7, r3 // top 3b < 3
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=5
	usub8.w r2, r7, r3 // >= 3 ?
	sel.w r7, r2, r7 // select
	and.w r2, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r2, LSR #4 // range < 31
	bic.w r2, r8, r3 // top 3b < 8
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=10
	bic.w r2, r8, r3 // top 3b < 3
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=5
	usub8.w r2, r8, r3 // >= 3 ?
	sel.w r8, r2, r8 // select
	and.w r2, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r2, LSR #4 // range < 31
	bic.w r2, r9, r3 // top 3b < 8
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=10
	bic.w r2, r9, r3 // top 3b < 3
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=5
	usub8.w r2, r9, r3 // >= 3 ?
	sel.w r9, r2, r9 // select
	and.w r2, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r2, LSR #4 // range < 31
	bic.w r2, r10, r3 // top 3b < 8
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=10
	bic.w r2, r10, r3 // top 3b < 3
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=5
	usub8.w r2, r10, r3 // >= 3 ?
	sel.w r10, r2, r10 // select
	and.w r2, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r2, LSR #4 // range < 31
	bic.w r2, r11, r3 // top 3b < 8
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=10
	bic.w r2, r11, r3 // top 3b < 3
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=5
	usub8.w r2, r11, r3 // >= 3 ?
	sel.w r11, r2, r11 // select
	and.w r2, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r2, LSR #4 // range < 31
	bic.w r2, r12, r3 // top 3b < 8
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=10
	bic.w r2, r12, r3 // top 3b < 3
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=5
	usub8.w r2, r12, r3 // >= 3 ?
	sel.w r12, r2, r12 // select
	bic.w r2, lr, r3 // top 3b < 8
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=10
	bic.w r2, lr, r3 // top 3b < 3
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=5
	usub8.w r2, lr, r3 // >= 3 ?
	sel.w lr, r2, lr // select
	str r7, [r0, #4]
	str r8, [r0, #8]
	str r9, [r0, #12]
	str r10, [r0, #16]
	str r11, [r0, #20]
	str r12, [r0, #24]
	str lr, [r0, #28]
	str r6, [r0], #32
	vmov.w r1, r2, s0, s1
	add.w r2, #32
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r1, r2, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r1, LSL #8
	add.w r11, r7, r2, LSL #8
	add.w r12, r8, r4, LSL #8
	add.w lr, r9, r5, LSL #8
	add.w r11, r11, r1, LSR #24
	add.w r12, r12, r2, LSR #24
	add.w lr, lr, r4, LSR #24
	vmov.w r1, r2, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r10, r10, r5, LSR #24
	add.w r6, r6, r1, LSL #8
	add.w r7, r7, r2, LSL #8
	add.w r8, r8, r4, LSL #8
	add.w r9, r9, r5, LSL #8
	add.w r7, r7, r1, LSR #24
	add.w r8, r8, r2, LSR #24
	add.w r9, r9, r4, LSR #24
	bic.w r2, r6, r3 // top 3b < 8
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=10
	bic.w r2, r6, r3 // top 3b < 3
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=5
	usub8.w r2, r6, r3 // >= 3 ?
	sel.w r6, r2, r6 // select
	and.w r2, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r2, LSR #4 // range < 31
	bic.w r2, r7, r3 // top 3b < 8
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=10
	bic.w r2, r7, r3 // top 3b < 3
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=5
	usub8.w r2, r7, r3 // >= 3 ?
	sel.w r7, r2, r7 // select
	and.w r2, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r2, LSR #4 // range < 31
	bic.w r2, r8, r3 // top 3b < 8
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=10
	bic.w r2, r8, r3 // top 3b < 3
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=5
	usub8.w r2, r8, r3 // >= 3 ?
	sel.w r8, r2, r8 // select
	and.w r2, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r2, LSR #4 // range < 31
	bic.w r2, r9, r3 // top 3b < 8
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=10
	bic.w r2, r9, r3 // top 3b < 3
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=5
	usub8.w r2, r9, r3 // >= 3 ?
	sel.w r9, r2, r9 // select
	and.w r2, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r2, LSR #4 // range < 31
	bic.w r2, r10, r3 // top 3b < 8
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=10
	bic.w r2, r10, r3 // top 3b < 3
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=5
	usub8.w r2, r10, r3 // >= 3 ?
	sel.w r10, r2, r10 // select
	and.w r2, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r2, LSR #4 // range < 31
	bic.w r2, r11, r3 // top 3b < 8
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=10
	bic.w r2, r11, r3 // top 3b < 3
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=5
	usub8.w r2, r11, r3 // >= 3 ?
	sel.w r11, r2, r11 // select
	and.w r2, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r2, LSR #4 // range < 31
	bic.w r2, r12, r3 // top 3b < 8
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=10
	bic.w r2, r12, r3 // top 3b < 3
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=5
	usub8.w r2, r12, r3 // >= 3 ?
	sel.w r12, r2, r12 // select
	bic.w r2, lr, r3 // top 3b < 8
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=10
	bic.w r2, lr, r3 // top 3b < 3
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=5
	usub8.w r2, lr, r3 // >= 3 ?
	sel.w lr, r2, lr // select
	str r7, [r0, #4]
	str r8, [r0, #8]
	str r9, [r0, #12]
	str r10, [r0, #16]
	str r11, [r0, #20]
	str r12, [r0, #24]
	str lr, [r0, #28]
	str r6, [r0], #32
	vmov.w r1, r2, s0, s1
	add.w r1, #16
	add.w r2, #32
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s2, s3, r6, r7
	vmov.w s4, s5, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s6, s7, r5, r6
	vmov.w s8, s9, r7, r8
	mov.w r6, #0
	ldr.w r5, [r2], #4
	ldr.w r4, [r1, #12]
	ldr.w lr, [r1, #8]
	ldr.w r12, [r1, #4]
	ldr.w r11, [r1], #32
	umull.w r7, r8, r12, r5
	umull.w r9, r10, r4, r5
	umlal.w r6, r7, r11, r5
	umlal.w r8, r9, lr, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, lr, r5
	umlal.w r8, r9, r12, r5
	umlal.w r7, r8, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r12, r5
	umlal.w r8, r9, r11, r5
	ldr.w r5, [r2], #4
	umlal.w r9, r10, r11, r5
	vmov.w s10, s11, r6, r7
	vmov.w s12, s13, r8, r9
	mov.w r6, #0
	mov.w r5, r10
	ldr.w r12, [r1, #-28]
	ldr.w lr, [r1, #-24]
	ldr.w r4, [r1, #-20]
	ldr.w r9, [r2, #-4]
	umull.w r7, r8, r4, r9
	umlal.w r6, r7, lr, r9
	umlal.w r5, r6, r12, r9
	ldr.w r9, [r2, #-8]
	umlal.w r6, r7, r4, r9
	umlal.w r5, r6, lr, r9
	ldr.w r9, [r2, #-12]
	umlal.w r5, r6, r4, r9
	vmov.w s14, s15, r5, r6
	vmov.w s16, s17, r7, r8
	vmov.w r1, r2, s6, s7
	vmov.w r4, r5, s8, s9
	vmov.w r6, r7, s14, s15
	vmov.w r8, r9, s16, s17
	add.w r10, r6, r1, LSL #8
	add.w r11, r7, r2, LSL #8
	add.w r12, r8, r4, LSL #8
	add.w lr, r9, r5, LSL #8
	add.w r11, r11, r1, LSR #24
	add.w r12, r12, r2, LSR #24
	add.w lr, lr, r4, LSR #24
	vmov.w r1, r2, s2, s3
	vmov.w r4, r5, s4, s5
	vmov.w r6, r7, s10, s11
	vmov.w r8, r9, s12, s13
	add.w r10, r10, r5, LSR #24
	add.w r6, r6, r1, LSL #8
	add.w r7, r7, r2, LSL #8
	add.w r8, r8, r4, LSL #8
	add.w r9, r9, r5, LSL #8
	add.w r7, r7, r1, LSR #24
	add.w r8, r8, r2, LSR #24
	add.w r9, r9, r4, LSR #24
	bic.w r2, r6, r3 // top 3b < 8
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=10
	bic.w r2, r6, r3 // top 3b < 3
	and.w r6, r6, r3 // bot 2b < 4
	add.w r6, r6, r2, LSR #2 // range <=5
	usub8.w r2, r6, r3 // >= 3 ?
	sel.w r6, r2, r6 // select
	and.w r2, r7, #0xF0F0F0F0 // top 4b < 16
	and.w r7, r7, #0x0F0F0F0F // bot 4b < 16
	add.w r7, r7, r2, LSR #4 // range < 31
	bic.w r2, r7, r3 // top 3b < 8
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=10
	bic.w r2, r7, r3 // top 3b < 3
	and.w r7, r7, r3 // bot 2b < 4
	add.w r7, r7, r2, LSR #2 // range <=5
	usub8.w r2, r7, r3 // >= 3 ?
	sel.w r7, r2, r7 // select
	and.w r2, r8, #0xF0F0F0F0 // top 4b < 16
	and.w r8, r8, #0x0F0F0F0F // bot 4b < 16
	add.w r8, r8, r2, LSR #4 // range < 31
	bic.w r2, r8, r3 // top 3b < 8
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=10
	bic.w r2, r8, r3 // top 3b < 3
	and.w r8, r8, r3 // bot 2b < 4
	add.w r8, r8, r2, LSR #2 // range <=5
	usub8.w r2, r8, r3 // >= 3 ?
	sel.w r8, r2, r8 // select
	and.w r2, r9, #0xF0F0F0F0 // top 4b < 16
	and.w r9, r9, #0x0F0F0F0F // bot 4b < 16
	add.w r9, r9, r2, LSR #4 // range < 31
	bic.w r2, r9, r3 // top 3b < 8
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=10
	bic.w r2, r9, r3 // top 3b < 3
	and.w r9, r9, r3 // bot 2b < 4
	add.w r9, r9, r2, LSR #2 // range <=5
	usub8.w r2, r9, r3 // >= 3 ?
	sel.w r9, r2, r9 // select
	and.w r2, r10, #0xF0F0F0F0 // top 4b < 16
	and.w r10, r10, #0x0F0F0F0F // bot 4b < 16
	add.w r10, r10, r2, LSR #4 // range < 31
	bic.w r2, r10, r3 // top 3b < 8
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=10
	bic.w r2, r10, r3 // top 3b < 3
	and.w r10, r10, r3 // bot 2b < 4
	add.w r10, r10, r2, LSR #2 // range <=5
	usub8.w r2, r10, r3 // >= 3 ?
	sel.w r10, r2, r10 // select
	and.w r2, r11, #0xF0F0F0F0 // top 4b < 16
	and.w r11, r11, #0x0F0F0F0F // bot 4b < 16
	add.w r11, r11, r2, LSR #4 // range < 31
	bic.w r2, r11, r3 // top 3b < 8
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=10
	bic.w r2, r11, r3 // top 3b < 3
	and.w r11, r11, r3 // bot 2b < 4
	add.w r11, r11, r2, LSR #2 // range <=5
	usub8.w r2, r11, r3 // >= 3 ?
	sel.w r11, r2, r11 // select
	and.w r2, r12, #0xF0F0F0F0 // top 4b < 16
	and.w r12, r12, #0x0F0F0F0F // bot 4b < 16
	add.w r12, r12, r2, LSR #4 // range < 31
	bic.w r2, r12, r3 // top 3b < 8
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=10
	bic.w r2, r12, r3 // top 3b < 3
	and.w r12, r12, r3 // bot 2b < 4
	add.w r12, r12, r2, LSR #2 // range <=5
	usub8.w r2, r12, r3 // >= 3 ?
	sel.w r12, r2, r12 // select
	bic.w r2, lr, r3 // top 3b < 8
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=10
	bic.w r2, lr, r3 // top 3b < 3
	and.w lr, lr, r3 // bot 2b < 4
	add.w lr, lr, r2, LSR #2 // range <=5
	usub8.w r2, lr, r3 // >= 3 ?
	sel.w lr, r2, lr // select
	str r7, [r0, #4]
	str r8, [r0, #8]
	str r9, [r0, #12]
	str r10, [r0, #16]
	str r11, [r0, #20]
	str r12, [r0, #24]
	str lr, [r0, #28]
	str r6, [r0], #32
	vpop.w {s16-s17}
	pop.w {r3-r12,pc}

Software Heritage — Copyright (C) 2015–2025, The Software Heritage developers. License: GNU AGPLv3+.
The source code of Software Heritage itself is available on our development forge.
The source code files archived by Software Heritage are available under their own copyright and licenses.
Terms of use: Archive access, API— Contact— JavaScript license information— Web API

back to top