1/* SPDX-License-Identifier: GPL-2.0 */
  2#include <linux/linkage.h>
  3#include <asm/processor.h>
  4#include <asm/ppc_asm.h>
  5#include <asm/reg.h>
  6#include <asm/asm-offsets.h>
  7#include <asm/cputable.h>
  8#include <asm/thread_info.h>
  9#include <asm/page.h>
 10#include <asm/ptrace.h>
 11#include <asm/export.h>
 12#include <asm/asm-compat.h>
 13
 14/*
 15 * Load state from memory into VMX registers including VSCR.
 16 * Assumes the caller has enabled VMX in the MSR.
 17 */
 18_GLOBAL(load_vr_state)
 19	li	r4,VRSTATE_VSCR
 20	lvx	v0,r4,r3
 21	mtvscr	v0
 22	REST_32VRS(0,r4,r3)
 23	blr
 24EXPORT_SYMBOL(load_vr_state)
 25_ASM_NOKPROBE_SYMBOL(load_vr_state); /* used by restore_math */
 26
 27/*
 28 * Store VMX state into memory, including VSCR.
 29 * Assumes the caller has enabled VMX in the MSR.
 30 */
 31_GLOBAL(store_vr_state)
 32	SAVE_32VRS(0, r4, r3)
 33	mfvscr	v0
 34	li	r4, VRSTATE_VSCR
 35	stvx	v0, r4, r3
 36	blr
 37EXPORT_SYMBOL(store_vr_state)
 38
 39/*
 
 40 * Disable VMX for the task which had it previously,
 41 * and save its vector registers in its thread_struct.
 42 * Enables the VMX for use in the kernel on return.
 43 * On SMP we know the VMX is free, since we give it up every
 44 * switch (ie, no lazy save of the vector registers).
 45 *
 46 * Note that on 32-bit this can only use registers that will be
 47 * restored by fast_exception_return, i.e. r3 - r6, r10 and r11.
 48 */
 49_GLOBAL(load_up_altivec)
 50	mfmsr	r5			/* grab the current MSR */
 51#ifdef CONFIG_PPC_BOOK3S_64
 52	/* interrupt doesn't set MSR[RI] and HPT can fault on current access */
 53	ori	r5,r5,MSR_RI
 54#endif
 55	oris	r5,r5,MSR_VEC@h
 56	MTMSRD(r5)			/* enable use of AltiVec now */
 57	isync
 58
 59	/*
 60	 * While userspace in general ignores VRSAVE, glibc uses it as a boolean
 61	 * to optimise userspace context save/restore. Whenever we take an
 62	 * altivec unavailable exception we must set VRSAVE to something non
 63	 * zero. Set it to all 1s. See also the programming note in the ISA.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 64	 */
 65	mfspr	r4,SPRN_VRSAVE
 66	cmpwi	0,r4,0
 67	bne+	1f
 68	li	r4,-1
 69	mtspr	SPRN_VRSAVE,r4
 701:
 71	/* enable use of VMX after return */
 72#ifdef CONFIG_PPC32
 73	addi	r5,r2,THREAD
 74	oris	r9,r9,MSR_VEC@h
 75#else
 76	ld	r4,PACACURRENT(r13)
 77	addi	r5,r4,THREAD		/* Get THREAD */
 78	oris	r12,r12,MSR_VEC@h
 79	std	r12,_MSR(r1)
 80#ifdef CONFIG_PPC_BOOK3S_64
 81	li	r4,0
 82	stb	r4,PACASRR_VALID(r13)
 83#endif
 84#endif
 85	li	r4,1
 86	stb	r4,THREAD_LOAD_VEC(r5)
 87	addi	r6,r5,THREAD_VRSTATE
 88	li	r10,VRSTATE_VSCR
 89	stw	r4,THREAD_USED_VR(r5)
 90	lvx	v0,r10,r6
 91	mtvscr	v0
 92	REST_32VRS(0,r4,r6)
 
 
 
 
 
 
 93	/* restore registers and return */
 94	blr
 95_ASM_NOKPROBE_SYMBOL(load_up_altivec)
 
 
 
 
 
 
 
 
 
 96
 97/*
 98 * save_altivec(tsk)
 99 * Save the vector registers to its thread_struct
 
 
100 */
101_GLOBAL(save_altivec)
 
 
 
 
 
 
 
102	addi	r3,r3,THREAD		/* want THREAD of task */
103	PPC_LL	r7,THREAD_VRSAVEAREA(r3)
104	PPC_LL	r5,PT_REGS(r3)
105	PPC_LCMPI	0,r7,0
106	bne	2f
107	addi	r7,r3,THREAD_VRSTATE
1082:	SAVE_32VRS(0,r4,r7)
109	mfvscr	v0
110	li	r4,VRSTATE_VSCR
111	stvx	v0,r4,r7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
112	blr
113
114#ifdef CONFIG_VSX
115
116#ifdef CONFIG_PPC32
117#error This asm code isn't ready for 32-bit kernels
118#endif
119
120/*
121 * load_up_vsx(unused, unused, tsk)
122 * Disable VSX for the task which had it previously,
123 * and save its vector registers in its thread_struct.
124 * Reuse the fp and vsx saves, but first check to see if they have
125 * been saved already.
126 */
127_GLOBAL(load_up_vsx)
128/* Load FP and VSX registers if they haven't been done yet */
129	andi.	r5,r12,MSR_FP
130	beql+	load_up_fpu		/* skip if already loaded */
131	andis.	r5,r12,MSR_VEC@h
132	beql+	load_up_altivec		/* skip if already loaded */
133
134#ifdef CONFIG_PPC_BOOK3S_64
135	/* interrupt doesn't set MSR[RI] and HPT can fault on current access */
136	li	r5,MSR_RI
137	mtmsrd	r5,1
138#endif
139
 
 
 
 
 
 
 
 
140	ld	r4,PACACURRENT(r13)
141	addi	r4,r4,THREAD		/* Get THREAD */
142	li	r6,1
143	stw	r6,THREAD_USED_VSR(r4) /* ... also set thread used vsr */
144	/* enable use of VSX after return */
145	oris	r12,r12,MSR_VSX@h
146	std	r12,_MSR(r1)
147	li	r4,0
148	stb	r4,PACASRR_VALID(r13)
149	b	fast_interrupt_return_srr
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
150
151#endif /* CONFIG_VSX */
152
153
154/*
155 * The routines below are in assembler so we can closely control the
156 * usage of floating-point registers.  These routines must be called
157 * with preempt disabled.
158 */
159	.data
160#ifdef CONFIG_PPC32
 
161fpzero:
162	.long	0
163fpone:
164	.long	0x3f800000	/* 1.0 in single-precision FP */
165fphalf:
166	.long	0x3f000000	/* 0.5 in single-precision FP */
167
168#define LDCONST(fr, name)	\
169	lis	r11,name@ha;	\
170	lfs	fr,name@l(r11)
171#else
172
 
173fpzero:
174	.quad	0
175fpone:
176	.quad	0x3ff0000000000000	/* 1.0 */
177fphalf:
178	.quad	0x3fe0000000000000	/* 0.5 */
179
180#define LDCONST(fr, name)		\
181	addis	r11,r2,name@toc@ha;	\
182	lfd	fr,name@toc@l(r11)
183#endif
 
184	.text
185/*
186 * Internal routine to enable floating point and set FPSCR to 0.
187 * Don't call it from C; it doesn't use the normal calling convention.
188 */
189SYM_FUNC_START_LOCAL(fpenable)
190#ifdef CONFIG_PPC32
191	stwu	r1,-64(r1)
192#else
193	stdu	r1,-64(r1)
194#endif
195	mfmsr	r10
196	ori	r11,r10,MSR_FP
197	mtmsr	r11
198	isync
199	stfd	fr0,24(r1)
200	stfd	fr1,16(r1)
201	stfd	fr31,8(r1)
202	LDCONST(fr1, fpzero)
203	mffs	fr31
204	MTFSF_L(fr1)
205	blr
206SYM_FUNC_END(fpenable)
207
208fpdisable:
209	mtlr	r12
210	MTFSF_L(fr31)
211	lfd	fr31,8(r1)
212	lfd	fr1,16(r1)
213	lfd	fr0,24(r1)
214	mtmsr	r10
215	isync
216	addi	r1,r1,64
217	blr
218
219/*
220 * Vector add, floating point.
221 */
222_GLOBAL(vaddfp)
223	mflr	r12
224	bl	fpenable
225	li	r0,4
226	mtctr	r0
227	li	r6,0
2281:	lfsx	fr0,r4,r6
229	lfsx	fr1,r5,r6
230	fadds	fr0,fr0,fr1
231	stfsx	fr0,r3,r6
232	addi	r6,r6,4
233	bdnz	1b
234	b	fpdisable
235
236/*
237 * Vector subtract, floating point.
238 */
239_GLOBAL(vsubfp)
240	mflr	r12
241	bl	fpenable
242	li	r0,4
243	mtctr	r0
244	li	r6,0
2451:	lfsx	fr0,r4,r6
246	lfsx	fr1,r5,r6
247	fsubs	fr0,fr0,fr1
248	stfsx	fr0,r3,r6
249	addi	r6,r6,4
250	bdnz	1b
251	b	fpdisable
252
253/*
254 * Vector multiply and add, floating point.
255 */
256_GLOBAL(vmaddfp)
257	mflr	r12
258	bl	fpenable
259	stfd	fr2,32(r1)
260	li	r0,4
261	mtctr	r0
262	li	r7,0
2631:	lfsx	fr0,r4,r7
264	lfsx	fr1,r5,r7
265	lfsx	fr2,r6,r7
266	fmadds	fr0,fr0,fr2,fr1
267	stfsx	fr0,r3,r7
268	addi	r7,r7,4
269	bdnz	1b
270	lfd	fr2,32(r1)
271	b	fpdisable
272
273/*
274 * Vector negative multiply and subtract, floating point.
275 */
276_GLOBAL(vnmsubfp)
277	mflr	r12
278	bl	fpenable
279	stfd	fr2,32(r1)
280	li	r0,4
281	mtctr	r0
282	li	r7,0
2831:	lfsx	fr0,r4,r7
284	lfsx	fr1,r5,r7
285	lfsx	fr2,r6,r7
286	fnmsubs	fr0,fr0,fr2,fr1
287	stfsx	fr0,r3,r7
288	addi	r7,r7,4
289	bdnz	1b
290	lfd	fr2,32(r1)
291	b	fpdisable
292
293/*
294 * Vector reciprocal estimate.  We just compute 1.0/x.
295 * r3 -> destination, r4 -> source.
296 */
297_GLOBAL(vrefp)
298	mflr	r12
299	bl	fpenable
300	li	r0,4
301	LDCONST(fr1, fpone)
302	mtctr	r0
303	li	r6,0
3041:	lfsx	fr0,r4,r6
305	fdivs	fr0,fr1,fr0
306	stfsx	fr0,r3,r6
307	addi	r6,r6,4
308	bdnz	1b
309	b	fpdisable
310
311/*
312 * Vector reciprocal square-root estimate, floating point.
313 * We use the frsqrte instruction for the initial estimate followed
314 * by 2 iterations of Newton-Raphson to get sufficient accuracy.
315 * r3 -> destination, r4 -> source.
316 */
317_GLOBAL(vrsqrtefp)
318	mflr	r12
319	bl	fpenable
320	stfd	fr2,32(r1)
321	stfd	fr3,40(r1)
322	stfd	fr4,48(r1)
323	stfd	fr5,56(r1)
324	li	r0,4
325	LDCONST(fr4, fpone)
326	LDCONST(fr5, fphalf)
327	mtctr	r0
328	li	r6,0
3291:	lfsx	fr0,r4,r6
330	frsqrte	fr1,fr0		/* r = frsqrte(s) */
331	fmuls	fr3,fr1,fr0	/* r * s */
332	fmuls	fr2,fr1,fr5	/* r * 0.5 */
333	fnmsubs	fr3,fr1,fr3,fr4	/* 1 - s * r * r */
334	fmadds	fr1,fr2,fr3,fr1	/* r = r + 0.5 * r * (1 - s * r * r) */
335	fmuls	fr3,fr1,fr0	/* r * s */
336	fmuls	fr2,fr1,fr5	/* r * 0.5 */
337	fnmsubs	fr3,fr1,fr3,fr4	/* 1 - s * r * r */
338	fmadds	fr1,fr2,fr3,fr1	/* r = r + 0.5 * r * (1 - s * r * r) */
339	stfsx	fr1,r3,r6
340	addi	r6,r6,4
341	bdnz	1b
342	lfd	fr5,56(r1)
343	lfd	fr4,48(r1)
344	lfd	fr3,40(r1)
345	lfd	fr2,32(r1)
346	b	fpdisable

 
 
  1#include <asm/processor.h>
  2#include <asm/ppc_asm.h>
  3#include <asm/reg.h>
  4#include <asm/asm-offsets.h>
  5#include <asm/cputable.h>
  6#include <asm/thread_info.h>
  7#include <asm/page.h>
  8#include <asm/ptrace.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  9
 10/*
 11 * load_up_altivec(unused, unused, tsk)
 12 * Disable VMX for the task which had it previously,
 13 * and save its vector registers in its thread_struct.
 14 * Enables the VMX for use in the kernel on return.
 15 * On SMP we know the VMX is free, since we give it up every
 16 * switch (ie, no lazy save of the vector registers).
 
 
 
 17 */
 18_GLOBAL(load_up_altivec)
 19	mfmsr	r5			/* grab the current MSR */
 
 
 
 
 20	oris	r5,r5,MSR_VEC@h
 21	MTMSRD(r5)			/* enable use of AltiVec now */
 22	isync
 23
 24/*
 25 * For SMP, we don't do lazy VMX switching because it just gets too
 26 * horrendously complex, especially when a task switches from one CPU
 27 * to another.  Instead we call giveup_altvec in switch_to.
 28 * VRSAVE isn't dealt with here, that is done in the normal context
 29 * switch code. Note that we could rely on vrsave value to eventually
 30 * avoid saving all of the VREGs here...
 31 */
 32#ifndef CONFIG_SMP
 33	LOAD_REG_ADDRBASE(r3, last_task_used_altivec)
 34	toreal(r3)
 35	PPC_LL	r4,ADDROFF(last_task_used_altivec)(r3)
 36	PPC_LCMPI	0,r4,0
 37	beq	1f
 38
 39	/* Save VMX state to last_task_used_altivec's THREAD struct */
 40	toreal(r4)
 41	addi	r4,r4,THREAD
 42	SAVE_32VRS(0,r5,r4)
 43	mfvscr	vr0
 44	li	r10,THREAD_VSCR
 45	stvx	vr0,r10,r4
 46	/* Disable VMX for last_task_used_altivec */
 47	PPC_LL	r5,PT_REGS(r4)
 48	toreal(r5)
 49	PPC_LL	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
 50	lis	r10,MSR_VEC@h
 51	andc	r4,r4,r10
 52	PPC_STL	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
 531:
 54#endif /* CONFIG_SMP */
 55
 56	/* Hack: if we get an altivec unavailable trap with VRSAVE
 57	 * set to all zeros, we assume this is a broken application
 58	 * that fails to set it properly, and thus we switch it to
 59	 * all 1's
 60	 */
 61	mfspr	r4,SPRN_VRSAVE
 62	cmpwi	0,r4,0
 63	bne+	1f
 64	li	r4,-1
 65	mtspr	SPRN_VRSAVE,r4
 661:
 67	/* enable use of VMX after return */
 68#ifdef CONFIG_PPC32
 69	mfspr	r5,SPRN_SPRG_THREAD		/* current task's THREAD (phys) */
 70	oris	r9,r9,MSR_VEC@h
 71#else
 72	ld	r4,PACACURRENT(r13)
 73	addi	r5,r4,THREAD		/* Get THREAD */
 74	oris	r12,r12,MSR_VEC@h
 75	std	r12,_MSR(r1)
 
 
 
 
 76#endif
 77	li	r4,1
 78	li	r10,THREAD_VSCR
 
 
 79	stw	r4,THREAD_USED_VR(r5)
 80	lvx	vr0,r10,r5
 81	mtvscr	vr0
 82	REST_32VRS(0,r4,r5)
 83#ifndef CONFIG_SMP
 84	/* Update last_task_used_altivec to 'current' */
 85	subi	r4,r5,THREAD		/* Back to 'current' */
 86	fromreal(r4)
 87	PPC_STL	r4,ADDROFF(last_task_used_altivec)(r3)
 88#endif /* CONFIG_SMP */
 89	/* restore registers and return */
 90	blr
 91
 92_GLOBAL(giveup_altivec_notask)
 93	mfmsr	r3
 94	andis.	r4,r3,MSR_VEC@h
 95	bnelr				/* Already enabled? */
 96	oris	r3,r3,MSR_VEC@h
 97	SYNC
 98	MTMSRD(r3)			/* enable use of VMX now */
 99	isync
100	blr
101
102/*
103 * giveup_altivec(tsk)
104 * Disable VMX for the task given as the argument,
105 * and save the vector registers in its thread_struct.
106 * Enables the VMX for use in the kernel on return.
107 */
108_GLOBAL(giveup_altivec)
109	mfmsr	r5
110	oris	r5,r5,MSR_VEC@h
111	SYNC
112	MTMSRD(r5)			/* enable use of VMX now */
113	isync
114	PPC_LCMPI	0,r3,0
115	beqlr				/* if no previous owner, done */
116	addi	r3,r3,THREAD		/* want THREAD of task */
 
117	PPC_LL	r5,PT_REGS(r3)
118	PPC_LCMPI	0,r5,0
119	SAVE_32VRS(0,r4,r3)
120	mfvscr	vr0
121	li	r4,THREAD_VSCR
122	stvx	vr0,r4,r3
123	beq	1f
124	PPC_LL	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
125#ifdef CONFIG_VSX
126BEGIN_FTR_SECTION
127	lis	r3,(MSR_VEC|MSR_VSX)@h
128FTR_SECTION_ELSE
129	lis	r3,MSR_VEC@h
130ALT_FTR_SECTION_END_IFSET(CPU_FTR_VSX)
131#else
132	lis	r3,MSR_VEC@h
133#endif
134	andc	r4,r4,r3		/* disable FP for previous task */
135	PPC_STL	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
1361:
137#ifndef CONFIG_SMP
138	li	r5,0
139	LOAD_REG_ADDRBASE(r4,last_task_used_altivec)
140	PPC_STL	r5,ADDROFF(last_task_used_altivec)(r4)
141#endif /* CONFIG_SMP */
142	blr
143
144#ifdef CONFIG_VSX
145
146#ifdef CONFIG_PPC32
147#error This asm code isn't ready for 32-bit kernels
148#endif
149
150/*
151 * load_up_vsx(unused, unused, tsk)
152 * Disable VSX for the task which had it previously,
153 * and save its vector registers in its thread_struct.
154 * Reuse the fp and vsx saves, but first check to see if they have
155 * been saved already.
156 */
157_GLOBAL(load_up_vsx)
158/* Load FP and VSX registers if they haven't been done yet */
159	andi.	r5,r12,MSR_FP
160	beql+	load_up_fpu		/* skip if already loaded */
161	andis.	r5,r12,MSR_VEC@h
162	beql+	load_up_altivec		/* skip if already loaded */
163
164#ifndef CONFIG_SMP
165	ld	r3,last_task_used_vsx@got(r2)
166	ld	r4,0(r3)
167	cmpdi	0,r4,0
168	beq	1f
169	/* Disable VSX for last_task_used_vsx */
170	addi	r4,r4,THREAD
171	ld	r5,PT_REGS(r4)
172	ld	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
173	lis	r6,MSR_VSX@h
174	andc	r6,r4,r6
175	std	r6,_MSR-STACK_FRAME_OVERHEAD(r5)
1761:
177#endif /* CONFIG_SMP */
178	ld	r4,PACACURRENT(r13)
179	addi	r4,r4,THREAD		/* Get THREAD */
180	li	r6,1
181	stw	r6,THREAD_USED_VSR(r4) /* ... also set thread used vsr */
182	/* enable use of VSX after return */
183	oris	r12,r12,MSR_VSX@h
184	std	r12,_MSR(r1)
185#ifndef CONFIG_SMP
186	/* Update last_task_used_vsx to 'current' */
187	ld	r4,PACACURRENT(r13)
188	std	r4,0(r3)
189#endif /* CONFIG_SMP */
190	b	fast_exception_return
191
192/*
193 * __giveup_vsx(tsk)
194 * Disable VSX for the task given as the argument.
195 * Does NOT save vsx registers.
196 * Enables the VSX for use in the kernel on return.
197 */
198_GLOBAL(__giveup_vsx)
199	mfmsr	r5
200	oris	r5,r5,MSR_VSX@h
201	mtmsrd	r5			/* enable use of VSX now */
202	isync
203
204	cmpdi	0,r3,0
205	beqlr-				/* if no previous owner, done */
206	addi	r3,r3,THREAD		/* want THREAD of task */
207	ld	r5,PT_REGS(r3)
208	cmpdi	0,r5,0
209	beq	1f
210	ld	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
211	lis	r3,MSR_VSX@h
212	andc	r4,r4,r3		/* disable VSX for previous task */
213	std	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
2141:
215#ifndef CONFIG_SMP
216	li	r5,0
217	ld	r4,last_task_used_vsx@got(r2)
218	std	r5,0(r4)
219#endif /* CONFIG_SMP */
220	blr
221
222#endif /* CONFIG_VSX */
223
224
225/*
226 * The routines below are in assembler so we can closely control the
227 * usage of floating-point registers.  These routines must be called
228 * with preempt disabled.
229 */
 
230#ifdef CONFIG_PPC32
231	.data
232fpzero:
233	.long	0
234fpone:
235	.long	0x3f800000	/* 1.0 in single-precision FP */
236fphalf:
237	.long	0x3f000000	/* 0.5 in single-precision FP */
238
239#define LDCONST(fr, name)	\
240	lis	r11,name@ha;	\
241	lfs	fr,name@l(r11)
242#else
243
244	.section ".toc","aw"
245fpzero:
246	.tc	FD_0_0[TC],0
247fpone:
248	.tc	FD_3ff00000_0[TC],0x3ff0000000000000	/* 1.0 */
249fphalf:
250	.tc	FD_3fe00000_0[TC],0x3fe0000000000000	/* 0.5 */
251
252#define LDCONST(fr, name)	\
253	lfd	fr,name@toc(r2)
 
254#endif
255
256	.text
257/*
258 * Internal routine to enable floating point and set FPSCR to 0.
259 * Don't call it from C; it doesn't use the normal calling convention.
260 */
261fpenable:
262#ifdef CONFIG_PPC32
263	stwu	r1,-64(r1)
264#else
265	stdu	r1,-64(r1)
266#endif
267	mfmsr	r10
268	ori	r11,r10,MSR_FP
269	mtmsr	r11
270	isync
271	stfd	fr0,24(r1)
272	stfd	fr1,16(r1)
273	stfd	fr31,8(r1)
274	LDCONST(fr1, fpzero)
275	mffs	fr31
276	MTFSF_L(fr1)
277	blr
 
278
279fpdisable:
280	mtlr	r12
281	MTFSF_L(fr31)
282	lfd	fr31,8(r1)
283	lfd	fr1,16(r1)
284	lfd	fr0,24(r1)
285	mtmsr	r10
286	isync
287	addi	r1,r1,64
288	blr
289
290/*
291 * Vector add, floating point.
292 */
293_GLOBAL(vaddfp)
294	mflr	r12
295	bl	fpenable
296	li	r0,4
297	mtctr	r0
298	li	r6,0
2991:	lfsx	fr0,r4,r6
300	lfsx	fr1,r5,r6
301	fadds	fr0,fr0,fr1
302	stfsx	fr0,r3,r6
303	addi	r6,r6,4
304	bdnz	1b
305	b	fpdisable
306
307/*
308 * Vector subtract, floating point.
309 */
310_GLOBAL(vsubfp)
311	mflr	r12
312	bl	fpenable
313	li	r0,4
314	mtctr	r0
315	li	r6,0
3161:	lfsx	fr0,r4,r6
317	lfsx	fr1,r5,r6
318	fsubs	fr0,fr0,fr1
319	stfsx	fr0,r3,r6
320	addi	r6,r6,4
321	bdnz	1b
322	b	fpdisable
323
324/*
325 * Vector multiply and add, floating point.
326 */
327_GLOBAL(vmaddfp)
328	mflr	r12
329	bl	fpenable
330	stfd	fr2,32(r1)
331	li	r0,4
332	mtctr	r0
333	li	r7,0
3341:	lfsx	fr0,r4,r7
335	lfsx	fr1,r5,r7
336	lfsx	fr2,r6,r7
337	fmadds	fr0,fr0,fr2,fr1
338	stfsx	fr0,r3,r7
339	addi	r7,r7,4
340	bdnz	1b
341	lfd	fr2,32(r1)
342	b	fpdisable
343
344/*
345 * Vector negative multiply and subtract, floating point.
346 */
347_GLOBAL(vnmsubfp)
348	mflr	r12
349	bl	fpenable
350	stfd	fr2,32(r1)
351	li	r0,4
352	mtctr	r0
353	li	r7,0
3541:	lfsx	fr0,r4,r7
355	lfsx	fr1,r5,r7
356	lfsx	fr2,r6,r7
357	fnmsubs	fr0,fr0,fr2,fr1
358	stfsx	fr0,r3,r7
359	addi	r7,r7,4
360	bdnz	1b
361	lfd	fr2,32(r1)
362	b	fpdisable
363
364/*
365 * Vector reciprocal estimate.  We just compute 1.0/x.
366 * r3 -> destination, r4 -> source.
367 */
368_GLOBAL(vrefp)
369	mflr	r12
370	bl	fpenable
371	li	r0,4
372	LDCONST(fr1, fpone)
373	mtctr	r0
374	li	r6,0
3751:	lfsx	fr0,r4,r6
376	fdivs	fr0,fr1,fr0
377	stfsx	fr0,r3,r6
378	addi	r6,r6,4
379	bdnz	1b
380	b	fpdisable
381
382/*
383 * Vector reciprocal square-root estimate, floating point.
384 * We use the frsqrte instruction for the initial estimate followed
385 * by 2 iterations of Newton-Raphson to get sufficient accuracy.
386 * r3 -> destination, r4 -> source.
387 */
388_GLOBAL(vrsqrtefp)
389	mflr	r12
390	bl	fpenable
391	stfd	fr2,32(r1)
392	stfd	fr3,40(r1)
393	stfd	fr4,48(r1)
394	stfd	fr5,56(r1)
395	li	r0,4
396	LDCONST(fr4, fpone)
397	LDCONST(fr5, fphalf)
398	mtctr	r0
399	li	r6,0
4001:	lfsx	fr0,r4,r6
401	frsqrte	fr1,fr0		/* r = frsqrte(s) */
402	fmuls	fr3,fr1,fr0	/* r * s */
403	fmuls	fr2,fr1,fr5	/* r * 0.5 */
404	fnmsubs	fr3,fr1,fr3,fr4	/* 1 - s * r * r */
405	fmadds	fr1,fr2,fr3,fr1	/* r = r + 0.5 * r * (1 - s * r * r) */
406	fmuls	fr3,fr1,fr0	/* r * s */
407	fmuls	fr2,fr1,fr5	/* r * 0.5 */
408	fnmsubs	fr3,fr1,fr3,fr4	/* 1 - s * r * r */
409	fmadds	fr1,fr2,fr3,fr1	/* r = r + 0.5 * r * (1 - s * r * r) */
410	stfsx	fr1,r3,r6
411	addi	r6,r6,4
412	bdnz	1b
413	lfd	fr5,56(r1)
414	lfd	fr4,48(r1)
415	lfd	fr3,40(r1)
416	lfd	fr2,32(r1)
417	b	fpdisable