Fix ARM code in light of plotter changes.

3 files changed, 218 insertions, 137 deletions

diff --git a/source/fitz/draw-scale-simple.c b/source/fitz/draw-scale-simple.c
index 0f267743..5d4c5b98 100644
--- a/source/fitz/draw-scale-simple.c
+++ b/source/fitz/draw-scale-simple.c
@@ -560,6 +560,10 @@ scale_row_to_temp2(unsigned char *dst, unsigned char *src, fz_weights *weights)
 __attribute__((naked));
 
 static void
+scale_row_to_temp3(unsigned char *dst, unsigned char *src, fz_weights *weights)
+__attribute__((naked));
+
+static void
 scale_row_to_temp4(unsigned char *dst, unsigned char *src, fz_weights *weights)
 __attribute__((naked));
 
@@ -704,6 +708,91 @@ scale_row_to_temp2(unsigned char *dst, unsigned char *src, fz_weights *weights)
 }
 
 static void
+scale_row_to_temp3(unsigned char *dst, unsigned char *src, fz_weights *weights)
+{
+	asm volatile(
+	ENTER_ARM
+	"stmfd	r13!,{r4-r11,r14}				\n"
+	"@ r0 = dst						\n"
+	"@ r1 = src						\n"
+	"@ r2 = weights						\n"
+	"ldr	r12,[r2],#4		@ r12= flip		\n"
+	"ldr	r3, [r2],#20		@ r3 = count r2 = &index\n"
+	"ldr	r4, [r2]		@ r4 = index[0]		\n"
+	"cmp	r12,#0			@ if (flip)		\n"
+	"beq	4f			@ {			\n"
+	"add	r2, r2, r4, LSL #1	@			\n"
+	"add	r2, r2, r4		@ r2 = &index[index[0]] \n"
+	"add	r0, r0, r3, LSL #1	@			\n"
+	"add	r0, r0, r3		@ dst += 3*count	\n"
+	"1:							\n"
+	"ldr	r4, [r2], #4		@ r4 = *contrib++	\n"
+	"ldr	r9, [r2], #4		@ r9 = len = *contrib++	\n"
+	"mov	r5, #128		@ r5 = r = 128		\n"
+	"mov	r6, #128		@ r6 = g = 128		\n"
+	"add	r7, r1, r4, LSL #1	@			\n"
+	"add	r4, r7, r4		@ r4 = min = &src[3*r4]	\n"
+	"mov	r7, #128		@ r7 = b = 128		\n"
+	"cmp	r9, #0			@ while (len-- > 0)	\n"
+	"beq	3f			@ {			\n"
+	"2:							\n"
+	"ldr	r14,[r2], #4		@ r14 = *contrib++	\n"
+	"ldrb	r8, [r4], #1		@ r8  = *min++		\n"
+	"ldrb	r11,[r4], #1		@ r11 = *min++		\n"
+	"ldrb	r12,[r4], #1		@ r12 = *min++		\n"
+	"subs	r9, r9, #1		@ r9 = len--		\n"
+	"mla	r5, r14,r8, r5		@ r += r8  * r14	\n"
+	"mla	r6, r14,r11,r6		@ g += r11 * r14	\n"
+	"mla	r7, r14,r12,r7		@ b += r12 * r14	\n"
+	"bgt	2b			@ }			\n"
+	"3:							\n"
+	"mov	r5, r5, lsr #8		@ r >>= 8		\n"
+	"mov	r6, r6, lsr #8		@ g >>= 8		\n"
+	"mov	r7, r7, lsr #8		@ b >>= 8		\n"
+	"strb	r5, [r0, #-3]!		@ *--dst=r		\n"
+	"strb	r6, [r0, #1]		@ *--dst=g		\n"
+	"strb	r7, [r0, #2]		@ *--dst=b		\n"
+	"subs	r3, r3, #1		@ i--			\n"
+	"bgt	1b			@ 			\n"
+	"ldmfd	r13!,{r4-r11,PC}	@ pop, return to thumb	\n"
+	"4:"
+	"add	r2, r2, r4, LSL #1	@			\n"
+	"add	r2, r2, r4		@ r2 = &index[index[0]] \n"
+	"5:"
+	"ldr	r4, [r2], #4		@ r4 = *contrib++	\n"
+	"ldr	r9, [r2], #4		@ r9 = len = *contrib++	\n"
+	"mov	r5, #128		@ r5 = r = 128		\n"
+	"mov	r6, #128		@ r6 = g = 128		\n"
+	"add	r7, r1, r4, LSL #1	@ r7 = min = &src[2*r4]	\n"
+	"add	r4, r7, r4		@ r4 = min = &src[3*r4]	\n"
+	"mov	r7, #128		@ r7 = b = 128		\n"
+	"cmp	r9, #0			@ while (len-- > 0)	\n"
+	"beq	7f			@ {			\n"
+	"6:							\n"
+	"ldr	r14,[r2], #4		@ r10 = *contrib++	\n"
+	"ldrb	r8, [r4], #1		@ r8  = *min++		\n"
+	"ldrb	r11,[r4], #1		@ r11 = *min++		\n"
+	"ldrb	r12,[r4], #1		@ r12 = *min++		\n"
+	"subs	r9, r9, #1		@ r9 = len--		\n"
+	"mla	r5, r14,r8, r5		@ r += r8  * r14	\n"
+	"mla	r6, r14,r11,r6		@ g += r11 * r14	\n"
+	"mla	r7, r14,r12,r7		@ b += r12 * r14	\n"
+	"bgt	6b			@ }			\n"
+	"7:							\n"
+	"mov	r5, r5, lsr #8		@ r >>= 8		\n"
+	"mov	r6, r6, lsr #8		@ g >>= 8		\n"
+	"mov	r7, r7, lsr #8		@ b >>= 8		\n"
+	"strb	r5, [r0], #1		@ *dst++=r		\n"
+	"strb	r6, [r0], #1		@ *dst++=g		\n"
+	"strb	r7, [r0], #1		@ *dst++=b		\n"
+	"subs	r3, r3, #1		@ i--			\n"
+	"bgt	5b			@ 			\n"
+	"ldmfd	r13!,{r4-r11,PC}	@ pop, return to thumb	\n"
+	ENTER_THUMB
+	);
+}
+
+static void
 scale_row_to_temp4(unsigned char *dst, unsigned char *src, fz_weights *weights)
 {
 	asm volatile(
@@ -943,6 +1032,59 @@ scale_row_to_temp2(unsigned char *dst, unsigned char *src, fz_weights *weights)
 }
 
 static void
+scale_row_to_temp3(unsigned char *dst, unsigned char *src, fz_weights *weights)
+{
+	int *contrib = &weights->index[weights->index[0]];
+	int len, i;
+	unsigned char *min;
+
+	assert(weights->n == 3);
+	if (weights->flip)
+	{
+		dst += 3*weights->count;
+		for (i=weights->count; i > 0; i--)
+		{
+			int c1 = 128;
+			int c2 = 128;
+			int c3 = 128;
+			min = &src[3 * *contrib++];
+			len = *contrib++;
+			while (len-- > 0)
+			{
+				int c = *contrib++;
+				c1 += *min++ * c;
+				c2 += *min++ * c;
+				c3 += *min++ * c;
+			}
+			*--dst = (unsigned char)(c3>>8);
+			*--dst = (unsigned char)(c2>>8);
+			*--dst = (unsigned char)(c1>>8);
+		}
+	}
+	else
+	{
+		for (i=weights->count; i > 0; i--)
+		{
+			int c1 = 128;
+			int c2 = 128;
+			int c3 = 128;
+			min = &src[3 * *contrib++];
+			len = *contrib++;
+			while (len-- > 0)
+			{
+				int c = *contrib++;
+				c1 += *min++ * c;
+				c2 += *min++ * c;
+				c3 += *min++ * c;
+			}
+			*dst++ = (unsigned char)(c1>>8);
+			*dst++ = (unsigned char)(c2>>8);
+			*dst++ = (unsigned char)(c3>>8);
+		}
+	}
+}
+
+static void
 scale_row_to_temp4(unsigned char *dst, unsigned char *src, fz_weights *weights)
 {
 	int *contrib = &weights->index[weights->index[0]];
@@ -1429,13 +1571,16 @@ fz_scale_pixmap_cached(fz_context *ctx, const fz_pixmap *src, float x, float y,
 		default:
 			row_scale = scale_row_to_temp;
 			break;
-		case 1: /* Image mask case */
+		case 1: /* Image mask case or Greyscale case */
 			row_scale = scale_row_to_temp1;
 			break;
 		case 2: /* Greyscale with alpha case */
 			row_scale = scale_row_to_temp2;
 			break;
-		case 4: /* RGBA */
+		case 3: /* RGB case */
+			row_scale = scale_row_to_temp3;
+			break;
+		case 4: /* RGBA or CMYK case */
 			row_scale = scale_row_to_temp4;
 			break;
 		}
diff --git a/source/fitz/halftone.c b/source/fitz/halftone.c
index 60e3c182..abe6d026 100644
--- a/source/fitz/halftone.c
+++ b/source/fitz/halftone.c
@@ -178,40 +178,38 @@ do_threshold_1(const unsigned char * restrict ht_line, const unsigned char * res
 	"1:							\n"
 	"ldr	r14,[r1], #4		@ r14= pixmap[0..3]	\n"
 	"ldr	r5, [r1], #4		@ r5 = pixmap[4..7]	\n"
-	"ldr	r6, [r1], #4		@ r6 = pixmap[8..11]	\n"
-	"ldr	r7, [r1], #4		@ r7 = pixmap[12..15]	\n"
 	"ldrb	r4, [r0], #8		@ r0 = ht_line += 8	\n"
-	"eors	r14,r14,r5, ROR #8	@ if (white)		\n"
-	"teqeq	r6, r7, ROR #8		@			\n"
+	"adds   r14,r14,#1		@ set eq iff r14=-1	\n"
+	"addeqs	r5, r5, #1		@ set eq iff r14=r5=-1	\n"
 	"beq	9b			@	white		\n"
-	"ldrb	r5, [r1, #-16]		@ r5 = pixmap[0]	\n"
+	"ldrb	r5, [r1, #-8]		@ r5 = pixmap[0]	\n"
 	"ldrb	r6, [r0, #-7]		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1, #-14]		@ r7 = pixmap[2]	\n"
+	"ldrb	r7, [r1, #-7]		@ r7 = pixmap[1]	\n"
 	"mov	r14,#0			@ r14= h = 0		\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x80		@	h |= 0x80	\n"
 	"ldrb	r4, [r0, #-6]		@ r4 = ht_line[2]	\n"
-	"ldrb	r5, [r1, #-12]		@ r5 = pixmap[4]	\n"
+	"ldrb	r5, [r1, #-6]		@ r5 = pixmap[2]	\n"
 	"cmp	r7, r6			@ if (r7 < r6)		\n"
 	"orrlt	r14,r14,#0x40		@	h |= 0x40	\n"
 	"ldrb	r6, [r0, #-5]		@ r6 = ht_line[3]	\n"
-	"ldrb	r7, [r1, #-10]		@ r7 = pixmap[6]	\n"
+	"ldrb	r7, [r1, #-5]		@ r7 = pixmap[3]	\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x20		@	h |= 0x20	\n"
 	"ldrb	r4, [r0, #-4]		@ r4 = ht_line[4]	\n"
-	"ldrb	r5, [r1, #-8]		@ r5 = pixmap[8]	\n"
+	"ldrb	r5, [r1, #-4]		@ r5 = pixmap[4]	\n"
 	"cmp	r7, r6			@ if (r7 < r6)		\n"
 	"orrlt	r14,r14,#0x10		@	h |= 0x10	\n"
 	"ldrb	r6, [r0, #-3]		@ r6 = ht_line[5]	\n"
-	"ldrb	r7, [r1, #-6]		@ r7 = pixmap[10]	\n"
+	"ldrb	r7, [r1, #-3]		@ r7 = pixmap[5]	\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x08		@	h |= 0x08	\n"
 	"ldrb	r4, [r0, #-2]		@ r4 = ht_line[6]	\n"
-	"ldrb	r5, [r1, #-4]		@ r5 = pixmap[12]	\n"
+	"ldrb	r5, [r1, #-2]		@ r5 = pixmap[6]	\n"
 	"cmp	r7, r6			@ if (r7 < r6)		\n"
 	"orrlt	r14,r14,#0x04		@	h |= 0x04	\n"
 	"ldrb	r6, [r0, #-1]		@ r6 = ht_line[7]	\n"
-	"ldrb	r7, [r1, #-2]		@ r7 = pixmap[14]	\n"
+	"ldrb	r7, [r1, #-1]		@ r7 = pixmap[7]	\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x02		@	h |= 0x02	\n"
 	"cmp	r7, r6			@ if (r7 < r6)		\n"
@@ -226,43 +224,43 @@ do_threshold_1(const unsigned char * restrict ht_line, const unsigned char * res
 	"adds	r3, r3, #7		@ w += 7		\n"
 	"ble	4f			@ if (w >= 0) {		\n"
 	"ldrb	r4, [r0], #1		@ r4 = ht_line[0]	\n"
-	"ldrb	r5, [r1], #2		@ r5 = pixmap[0]	\n"
+	"ldrb	r5, [r1], #1		@ r5 = pixmap[0]	\n"
 	"mov	r14, #0			@ r14= h = 0		\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x80		@	h |= 0x80	\n"
 	"cmp	r3, #1			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[1]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[2]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[1]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x40		@	h |= 0x40	\n"
 	"cmp	r3, #2			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[2]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[4]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[2]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x20		@	h |= 0x20	\n"
 	"cmp	r3, #3			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[3]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[6]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[3]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x10		@	h |= 0x10	\n"
 	"cmp	r3, #4			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[4]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[8]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[4]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x08		@	h |= 0x08	\n"
 	"cmp	r3, #5			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[5]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[10]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[5]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x04		@	h |= 0x04	\n"
 	"cmp	r3, #6			@			\n"
 	"ldrgtb	r4, [r0], #1		@ r6 = ht_line[6]	\n"
-	"ldrgtb	r5, [r1], #2		@ r7 = pixmap[12]	\n"
+	"ldrgtb	r5, [r1], #1		@ r7 = pixmap[6]	\n"
 	"ble	3f			@			\n"
 	"cmp	r5, r4			@ if (r5 < r4)		\n"
 	"orrlt	r14,r14,#0x02		@	h |= 0x02	\n"
@@ -344,7 +342,6 @@ static void
 do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * restrict pixmap, unsigned char *restrict out, int w, int ht_len)
 __attribute__((naked));
 
-#ifdef ARCH_UNALIGNED_OK
 static void
 do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * restrict pixmap, unsigned char *restrict out, int w, int ht_len)
 {
@@ -371,123 +368,40 @@ do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * res
 	"beq	2f			@ }			\n"
 	"blt	3f			@			\n"
 	"1:							\n"
-	"ldr	r5, [r1], #5		@ r5 = pixmap[0..3]	\n"
-	"ldr	r7, [r1], #5		@ r7 = pixmap[5..8]	\n"
+	"ldr	r5, [r1], #4		@ r5 = pixmap[0..3]	\n"
+	"ldr	r7, [r1], #4		@ r7 = pixmap[4..7]	\n"
 	"add	r0, r0, #8		@ r0 = ht_line += 8	\n"
 	"mov	r14,#0			@ r14= h = 0		\n"
 	"orrs	r5, r5, r7		@ if (r5 | r7 == 0)	\n"
 	"beq	9b			@	white		\n"
 	"ldrb	r4, [r0, #-8]		@ r4 = ht_line[0]	\n"
-	"ldrb	r5, [r1, #-10]		@ r5 = pixmap[0]	\n"
+	"ldrb	r5, [r1, #-8]		@ r5 = pixmap[0]	\n"
 	"ldrb	r6, [r0, #-7]		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1, #-9]		@ r7 = pixmap[1]	\n"
+	"ldrb	r7, [r1, #-7]		@ r7 = pixmap[1]	\n"
 	"cmp	r4, r5			@ if (r4 < r5)		\n"
 	"orrle	r14,r14,#0x80		@	h |= 0x80	\n"
 	"ldrb	r4, [r0, #-6]		@ r4 = ht_line[2]	\n"
-	"ldrb	r5, [r1, #-8]		@ r5 = pixmap[2]	\n"
+	"ldrb	r5, [r1, #-6]		@ r5 = pixmap[2]	\n"
 	"cmp	r6, r7			@ if (r6 < r7)		\n"
 	"orrle	r14,r14,#0x40		@	h |= 0x40	\n"
 	"ldrb	r6, [r0, #-5]		@ r6 = ht_line[3]	\n"
-	"ldrb	r7, [r1, #-7]		@ r7 = pixmap[3]	\n"
+	"ldrb	r7, [r1, #-5]		@ r7 = pixmap[3]	\n"
 	"cmp	r4, r5			@ if (r4 < r5)		\n"
 	"orrle	r14,r14,#0x20		@	h |= 0x20	\n"
 	"ldrb	r4, [r0, #-4]		@ r4 = ht_line[4]	\n"
-	"ldrb	r5, [r1, #-5]		@ r5 = pixmap[5]	\n"
+	"ldrb	r5, [r1, #-4]		@ r5 = pixmap[4]	\n"
 	"cmp	r6, r7			@ if (r6 < r7)		\n"
 	"orrle	r14,r14,#0x10		@	h |= 0x10	\n"
 	"ldrb	r6, [r0, #-3]		@ r6 = ht_line[5]	\n"
-	"ldrb	r7, [r1, #-4]		@ r7 = pixmap[6]	\n"
+	"ldrb	r7, [r1, #-3]		@ r7 = pixmap[5]	\n"
 	"cmp	r4, r5			@ if (r4 < r5)		\n"
 	"orrle	r14,r14,#0x08		@	h |= 0x08	\n"
 	"ldrb	r4, [r0, #-2]		@ r4 = ht_line[6]	\n"
-	"ldrb	r5, [r1, #-3]		@ r5 = pixmap[7]	\n"
+	"ldrb	r5, [r1, #-2]		@ r5 = pixmap[6]	\n"
 	"cmp	r6, r7			@ if (r6 < r7)		\n"
 	"orrle	r14,r14,#0x04		@	h |= 0x04	\n"
 	"ldrb	r6, [r0, #-1]		@ r6 = ht_line[7]	\n"
-	"ldrb	r7, [r1, #-2]		@ r7 = pixmap[8]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x02		@	h |= 0x02	\n"
-	"cmp	r6, r7			@ if (r7 < r6)		\n"
-	"orrle	r14,r14,#0x01		@	h |= 0x01	\n"
-	"subs	r12,r12,#2		@ r12 = l -= 2		\n"
-	"strb	r14,[r2], #1		@ *out++ = h		\n"
-	"moveq	r12,r9			@ if(l==0) l = ht_len	\n"
-	"subeq	r0, r0, r9, LSL #2	@          ht_line -= l	\n"
-	"subs	r3, r3, #2		@ w -= 2		\n"
-	"bgt	1b			@ }			\n"
-	"blt	3f			@			\n"
-	"2:							\n"
-	"ldrb	r4, [r0], #1		@ r4 = ht_line[0]	\n"
-	"ldrb	r5, [r1], #1		@ r5 = pixmap[0]	\n"
-	"mov	r14, #0			@ r14= h = 0		\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1], #1		@ r7 = pixmap[1]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x80		@	h |= 0x80	\n"
-	"ldrb	r4, [r0], #1		@ r6 = ht_line[2]	\n"
-	"ldrb	r5, [r1], #1		@ r7 = pixmap[2]	\n"
-	"cmp	r6, r7			@ if (r6 < r7)		\n"
-	"orrle	r14,r14,#0x40		@	h |= 0x40	\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1], #2		@ r7 = pixmap[2]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x20		@	h |= 0x20	\n"
-	"cmp	r6, r7			@ if (r6 < r7)		\n"
-	"orrle	r14,r14,#0x10		@	h |= 0x10	\n"
-	"strb	r14,[r2]		@ *out = h		\n"
-	"3:							\n"
-	"ldmfd	r13!,{r4-r7,r9,PC}	@ pop, return to thumb	\n"
-	ENTER_THUMB
-	);
-}
-#else
-/* Vanilla version, should work on all ARMs */
-static void
-do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * restrict pixmap, unsigned char *restrict out, int w, int ht_len)
-{
-	asm volatile(
-	ENTER_ARM
-	// Store one more reg that required to keep double stack alignment
-	"stmfd	r13!,{r4-r7,r9,r14}				\n"
-	"@ r0 = ht_line						\n"
-	"@ r1 = pixmap						\n"
-	"@ r2 = out						\n"
-	"@ r3 = w						\n"
-	"@ <> = ht_len						\n"
-	"ldr	r9, [r13,#6*4]		@ r9 = ht_len		\n"
-	"subs	r3, r3, #1		@ r3 = w -= 1		\n"
-	"ble	2f			@ while (w > 0) {	\n"
-	"mov	r12,r9			@ r12= l = ht_len	\n"
-	"1:							\n"
-	"mov	r14,#0			@ r14= h = 0		\n"
-	"ldrb	r4, [r0], #1		@ r4 = ht_line[0]	\n"
-	"ldrb	r5, [r1], #1		@ r5 = pixmap[0]	\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1], #1		@ r7 = pixmap[1]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x80		@	h |= 0x80	\n"
-	"ldrb	r4, [r0], #1		@ r4 = ht_line[2]	\n"
-	"ldrb	r5, [r1], #1		@ r5 = pixmap[2]	\n"
-	"cmp	r6, r7			@ if (r6 < r7)		\n"
-	"orrle	r14,r14,#0x40		@	h |= 0x40	\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[3]	\n"
-	"ldrb	r7, [r1], #2		@ r7 = pixmap[3]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x20		@	h |= 0x20	\n"
-	"ldrb	r4, [r0], #1		@ r4 = ht_line[4]	\n"
-	"ldrb	r5, [r1], #1		@ r5 = pixmap[4]	\n"
-	"cmp	r6, r7			@ if (r6 < r7)		\n"
-	"orrle	r14,r14,#0x10		@	h |= 0x10	\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[5]	\n"
-	"ldrb	r7, [r1], #1		@ r7 = pixmap[6]	\n"
-	"cmp	r4, r5			@ if (r4 < r5)		\n"
-	"orrle	r14,r14,#0x08		@	h |= 0x08	\n"
-	"ldrb	r4, [r0], #1		@ r4 = ht_line[6]	\n"
-	"ldrb	r5, [r1], #1		@ r5 = pixmap[7]	\n"
-	"cmp	r6, r7			@ if (r6 < r7)		\n"
-	"orrle	r14,r14,#0x04		@	h |= 0x04	\n"
-	"ldrb	r6, [r0], #1		@ r6 = ht_line[7]	\n"
-	"ldrb	r7, [r1], #2		@ r7 = pixmap[8]	\n"
+	"ldrb	r7, [r1, #-1]		@ r7 = pixmap[7]	\n"
 	"cmp	r4, r5			@ if (r4 < r5)		\n"
 	"orrle	r14,r14,#0x02		@	h |= 0x02	\n"
 	"cmp	r6, r7			@ if (r7 < r6)		\n"
@@ -512,7 +426,7 @@ do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * res
 	"cmp	r6, r7			@ if (r6 < r7)		\n"
 	"orrle	r14,r14,#0x40		@	h |= 0x40	\n"
 	"ldrb	r6, [r0], #1		@ r6 = ht_line[1]	\n"
-	"ldrb	r7, [r1], #2		@ r7 = pixmap[2]	\n"
+	"ldrb	r7, [r1], #1		@ r7 = pixmap[3]	\n"
 	"cmp	r4, r5			@ if (r4 < r5)		\n"
 	"orrle	r14,r14,#0x20		@	h |= 0x20	\n"
 	"cmp	r6, r7			@ if (r6 < r7)		\n"
@@ -523,7 +437,6 @@ do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * res
 	ENTER_THUMB
 	);
 }
-#endif /* UNALIGNED */
 #else
 static void do_threshold_4(const unsigned char * restrict ht_line, const unsigned char * restrict pixmap, unsigned char * restrict out, int w, int ht_len)
 {
diff --git a/source/fitz/pixmap.c b/source/fitz/pixmap.c
index e0d24265..3385021f 100644
--- a/source/fitz/pixmap.c
+++ b/source/fitz/pixmap.c
@@ -201,11 +201,11 @@ fz_pixmap_samples(fz_context *ctx, fz_pixmap *pix)
 */
 #ifdef ARCH_ARM
 static void
-clear_cmyk_bitmap(unsigned char *samples, int c, int value)
+clear_cmyka_bitmap_ARM(uint32_t *samples, int c, int value)
 __attribute__((naked));
 
 static void
-clear_cmyk_bitmap(unsigned char *samples, int c, int value)
+clear_cmyka_bitmap_ARM(uint32_t *samples, int c, int value)
 {
 	asm volatile(
 	ENTER_ARM
@@ -251,7 +251,8 @@ clear_cmyk_bitmap(unsigned char *samples, int c, int value)
 	ENTER_THUMB
 	);
 }
-#else
+#endif
+
 static void
 clear_cmyk_bitmap(unsigned char *samples, int w, int h, int stride, int value, int alpha)
 {
@@ -264,6 +265,10 @@ clear_cmyk_bitmap(unsigned char *samples, int w, int h, int stride, int value, i
 		stride -= w*5;
 		if (stride == 0)
 		{
+#ifdef ARCH_ARM
+			clear_cmyka_bitmap_ARM(s, c, alpha);
+			return;
+#else
 			/* We can do it all fast (except for maybe a few stragglers) */
 			union
 			{
@@ -306,6 +311,7 @@ clear_cmyk_bitmap(unsigned char *samples, int w, int h, int stride, int value, i
 				}
 			}
 			c += 3;
+#endif
 		}
 		t = (unsigned char *)s;
 		w = c;
@@ -327,27 +333,45 @@ clear_cmyk_bitmap(unsigned char *samples, int w, int h, int stride, int value, i
 	else
 	{
 		stride -= w*4;
-		if ((stride & 3)== 0)
+		if ((stride & 3) == 0)
 		{
-			/* We can do it all fast */
-			union
+			if (stride == 0)
 			{
-				uint8_t bytes[4];
-				uint32_t word;
-			} d;
-
-			d.word = 0;
-			d.bytes[3] = value;
-			w *= h;
-
+				w *= h;
+				h = 1;
+			}
+			w *= 4;
+			if (value == 0)
 			{
-				const uint32_t a0 = d.word;
-				while (w > 0)
+				while (h--)
 				{
-					*s++ = a0;
-					w--;
+					memset(s, 0, w);
+					s += (stride>>2);
+				}
+			}
+			else
+			{
+				/* We can do it all fast */
+				union
+				{
+					uint8_t bytes[4];
+					uint32_t word;
+				} d;
+
+				d.word = 0;
+				d.bytes[3] = value;
+				{
+					const uint32_t a0 = d.word;
+					while (h--)
+					{
+						int ww = w;
+						while (ww--)
+						{
+							*s++ = a0;
+						}
+						s += (stride>>2);
+					}
 				}
-				s += (stride>>2);
 			}
 		}
 		else
@@ -369,7 +393,6 @@ clear_cmyk_bitmap(unsigned char *samples, int w, int h, int stride, int value, i
 		}
 	}
 }
-#endif
 
 void
 fz_clear_pixmap(fz_context *ctx, fz_pixmap *pix)