arm: Allow building assembly in thumb mode

Windows on arm is thumb2 only.

Add the necessary "it" instructions before conditionally executed
instructions (which doesn't emit any extra instructions when not
building in thumb mode). The number of "it" instructions could
be reduced in some places by reordering the instructions, but keeping
them as they were originally to avoid any impact on existing
targets.

Remove redundant .arm directives; the assembler normally starts out
in that mode anyway, and for windows we shouldn't override the
default mode.

(cherry picked from commit 4a9b6e5193)
Signed-off-by: Hugo Beauzée-Luyssen <hugo@beauzee.fr>

modules/arm_neon/amplify.S

@@ -21,7 +21,6 @@
 #include "asm.S"
 
 	.syntax	unified
-	.arm
 #if HAVE_AS_FPU_DIRECTIVE
 	.fpu	neon
 #endif
@@ -33,6 +32,7 @@
 	.align 2
 function amplify_float_arm_neon
 	cmp		SIZE,	#0
+	it		eq
 	bxeq		lr
 #ifdef __ARM_PCS
 	vmov		s0,	r3	@ softfp

modules/arm_neon/deinterleave_chroma.S

@@ -52,6 +52,7 @@ function deinterleave_chroma_neon
 	sub		IPAD,	IPITCH,	WIDTH, lsl #1
 	sub		OPAD,	OPITCH,	WIDTH
 1:
+	ite		gt
 	movsgt		COUNT,	WIDTH
 	pople		{r4-r6,pc}
 2:

modules/arm_neon/i420_rgb.S

@@ -95,6 +95,7 @@ function i420_rgb_neon
 	/* round the width to be a multiple of 16 */
 	ands		OPAD, WIDTH, #15
 	sub			WIDTH, WIDTH, OPAD
+	it		ne
 	addne		WIDTH, WIDTH, #16
 
 	/* init constants (scale value by 64) */
@@ -115,10 +116,12 @@ function i420_rgb_neon
 	sub			YPAD,	YPITCH,	WIDTH
 
 loop_row:
+	it	gt
 	movsgt	COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
 	/* exit if all rows have been processed */
+	itt	le
 	vpople	{q4-q7}
 	pople	{r4-r8,r10-r11,pc}
 

modules/arm_neon/i420_rv16.S

@@ -98,6 +98,7 @@ function i420_rv16_neon
 	/* round the width to be a multiple of 16 */
 	ands		OPAD, WIDTH, #15
 	sub			WIDTH, WIDTH, OPAD
+	it		ne
 	addne		WIDTH, WIDTH, #16
 
 	/* init constants (scale value by 64) */
@@ -117,10 +118,12 @@ function i420_rv16_neon
 	sub			YPAD,	YPITCH,	WIDTH
 
 loop_row:
+	it	gt
 	movsgt	COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
 	/* exit if all rows have been processed */
+	itt	le
 	vpople	{q4-q7}
 	pople	{r4-r8,r10-r11,pc}
 

modules/arm_neon/i420_yuyv.S

@@ -49,9 +49,11 @@ function i420_yuyv_neon
 	sub		OPAD,	OPITCH,	WIDTH,	lsl #1
 	sub		YPAD,	YPITCH,	WIDTH
 1:
+	it		gt
 	movsgt		COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
+	it		le
 	pople		{r4-r8,r10-r11,pc}
 2:
 	pld		[U, #64]
@@ -86,9 +88,11 @@ function i420_uyvy_neon
 	sub		OPAD,	OPITCH,	WIDTH,	lsl #1
 	sub		YPAD,	YPITCH,	WIDTH
 1:
+	it		gt
 	movsgt		COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
+	it		le
 	pople		{r4-r8,r10-r11,pc}
 2:
 	pld		[U, #64]

modules/arm_neon/i422_yuyv.S

@@ -45,6 +45,7 @@ function i422_yuyv_neon
 	sub		OPAD,	OPAD,	WIDTH,	lsl #1
 	sub		YPAD,	YPAD,	WIDTH
 1:
+	ite		gt
 	movsgt		COUNT,	WIDTH
 	pople		{r4-r6,pc}
 2:
@@ -76,6 +77,7 @@ function i422_uyvy_neon
 	sub		OPAD,	OPAD,	WIDTH,	lsl #1
 	sub		YPAD,	YPAD,	WIDTH
 1:
+	ite		gt
 	movsgt		COUNT,	WIDTH
 	pople		{r4-r6,pc}
 2:

modules/arm_neon/nv12_rgb.S

@@ -91,6 +91,7 @@ function nv12_rgb_neon
 	/* round the width to be a multiple of 16 */
 	ands		OPAD, WIDTH, #15
 	sub			WIDTH, WIDTH, OPAD
+	it		ne
 	addne		WIDTH, WIDTH, #16
 
 	/* init constants (scale value by 64) */
@@ -111,10 +112,12 @@ function nv12_rgb_neon
 	sub			YPAD,	YPITCH,	WIDTH
 
 loop_row:
+	it	gt
 	movsgt	COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
 	/* exit if all rows have been processed */
+	itt	le
 	vpople	{q4-q7}
 	pople	{r4-r8,r10-r11,pc}
 

modules/arm_neon/nv21_rgb.S

@@ -91,6 +91,7 @@ function nv21_rgb_neon
 	/* round the width to be a multiple of 16 */
 	ands		OPAD, WIDTH, #15
 	sub			WIDTH, WIDTH, OPAD
+	it		ne
 	addne		WIDTH, WIDTH, #16
 
 	/* init constants (scale value by 64) */
@@ -111,10 +112,12 @@ function nv21_rgb_neon
 	sub			YPAD,	YPITCH,	WIDTH
 
 loop_row:
+	it	gt
 	movsgt	COUNT,	WIDTH
 	add		O2,	O1,	OPITCH
 	add		Y2,	Y1,	YPITCH
 	/* exit if all rows have been processed */
+	itt	le
 	vpople	{q4-q7}
 	pople	{r4-r8,r10-r11,pc}
 

modules/arm_neon/yuyv_i422.S

@@ -45,6 +45,7 @@ function yuyv_i422_neon
 	sub		YPAD,	YPAD,	WIDTH
 	sub		IPAD,	IPAD,	WIDTH,	lsl #1
 1:
+	ite		gt
 	movsgt		COUNT,	WIDTH
 	pople		{r4-r6,pc}
 2:
@@ -74,6 +75,7 @@ function uyvy_i422_neon
 	sub		YPAD,	YPAD,	WIDTH
 	sub		IPAD,	IPAD,	WIDTH,	lsl #1
 1:
+	ite		gt
 	movsgt		COUNT,	WIDTH
 	pople		{r4-r6,pc}
 2:

modules/video_filter/deinterlace/merge_arm.S

@@ -21,7 +21,6 @@
 #include "../arm_neon/asm.S"
 
 	.syntax	unified
-	.arm
 #if HAVE_AS_ARCH_DIRECTIVE
 	.arch	armv6
 #endif
@@ -67,6 +66,7 @@ function merge8_arm_neon
 	vst1.u8		{q0-q1},	[DEST,:128]!
 3:
 	cmp		SIZE,	#16
+	it		lo
 	bxlo		lr
 	vld1.u8		{q0},		[SRC1,:128]!
 	sub		SIZE,	SIZE,	#16
@@ -106,6 +106,7 @@ function merge16_arm_neon
 	vst1.u16	{q0-q1},	[DEST,:128]!
 3:
 	cmp		SIZE,	#16
+	it		lo
 	bxlo		lr
 	vld1.u16	{q0},		[SRC1,:128]!
 	sub		SIZE,	SIZE,	#16
@@ -131,6 +132,7 @@ function merge8_armv6
 	stm		DEST!,	{r4-r5}
 	uhadd8		r7,	r7,	lr
 	stm		DEST!,	{r6-r7}
+	it		eq
 	popeq		{r4-r9,pc}
 	b		1b
 
@@ -151,5 +153,6 @@ function merge16_armv6
 	stm		DEST!,	{r4-r5}
 	uhadd16		r7,	r7,	lr
 	stm		DEST!,	{r6-r7}
+	it		eq
 	popeq		{r4-r9,pc}
 	b		1b