Fix ldn/stn multiple instructions. Fix testcases with unaligned data.

	sim/aarch64/
	* simulator.c (vec_load): Add M argument.  Rewrite to iterate over
	registers based on structure size.
	(LD4, LD3, LD2, LD1_2, LD1_3, LD1_4): Pass new arg to vec_load.
	(LD1_1): Replace with call to vec_load.
	(vec_store): Add new M argument.  Rewrite to iterate over registers
	based on structure size.
	(ST4, ST3, ST2, ST1_2, ST1_3, ST1_4): Pass new arg to vec_store.
	(ST1_1): Replace with call to vec_store.

	sim/testsuite/sim/aarch64/
	* fcvtz.s, fstur.s, ldn_single.s, ldnr.s, mla.s, mls.s, uzp.s: Align
	data.
	* sumulh.s: Delete unnecessary data alignment.
	* stn_single.s: Align data.  Fix unaligned ldr insns.  Adjust cmp
	arguments to match change.
	* ldn_multiple.s, stn_multiple.s: New.

12 files changed, 341 insertions, 14 deletions

diff --git a/sim/testsuite/sim/aarch64/ChangeLog b/sim/testsuite/sim/aarch64/ChangeLog
index cf0da6d..f4671da 100644
--- a/sim/testsuite/sim/aarch64/ChangeLog
+++ b/sim/testsuite/sim/aarch64/ChangeLog
@@ -1,3 +1,12 @@
+2017-04-22  Jim Wilson  <jim.wilson@linaro.org>
+
+	* fcvtz.s, fstur.s, ldn_single.s, ldnr.s, mla.s, mls.s, uzp.s: Align
+	data.
+	* sumulh.s: Delete unnecessary data alignment.
+	* stn_single.s: Align data.  Fix unaligned ldr insns.  Adjust cmp
+	arguments to match change.
+	* ldn_multiple.s, stn_multiple.s: New.
+
 2017-04-08  Jim Wilson  <jim.wilson@linaro.org>
 
 	* fcvtl.s: New.
diff --git a/sim/testsuite/sim/aarch64/fcvtz.s b/sim/testsuite/sim/aarch64/fcvtz.s
index 9bb6f9b..311fc2e 100644
--- a/sim/testsuite/sim/aarch64/fcvtz.s
+++ b/sim/testsuite/sim/aarch64/fcvtz.s
@@ -8,6 +8,7 @@
 # For 64-bit unsigned convert, test values 1.5, LONG_MAX, and ULONG_MAX.
 
 	.data
+	.align 4
 fm1p5:
 	.word	3217031168
 fimax:
diff --git a/sim/testsuite/sim/aarch64/fstur.s b/sim/testsuite/sim/aarch64/fstur.s
index 2206ae5..80e5c67 100644
--- a/sim/testsuite/sim/aarch64/fstur.s
+++ b/sim/testsuite/sim/aarch64/fstur.s
@@ -8,6 +8,7 @@
 .include "testutils.inc"
 
 	.data
+	.align 4
 fm1:
 	.word 3212836864
 fmax:
diff --git a/sim/testsuite/sim/aarch64/ldn_multiple.s b/sim/testsuite/sim/aarch64/ldn_multiple.s
new file mode 100644
index 0000000..285ef7e
--- /dev/null
+++ b/sim/testsuite/sim/aarch64/ldn_multiple.s
@@ -0,0 +1,136 @@
+# mach: aarch64
+
+# Check the load multiple structure instructions: ld1, ld2, ld3, ld4.
+# Check the addressing modes: no offset, post-index immediate offset,
+# post-index register offset.
+
+.include "testutils.inc"
+
+	.data
+	.align 4
+input:
+	.word 0x04030201
+	.word 0x08070605
+	.word 0x0c0b0a09
+	.word 0x100f0e0d
+	.word 0xfcfdfeff
+	.word 0xf8f9fafb
+	.word 0xf4f5f6f7
+	.word 0xf0f1f2f3
+
+	start
+	adrp x0, input
+	add x0, x0, :lo12:input
+
+	mov x2, x0
+	mov x3, #16
+	ld1 {v0.16b}, [x2], 16
+	ld1 {v1.8h}, [x2], x3
+	addv b4, v0.16b
+	addv b5, v1.16b
+	mov x4, v4.d[0]
+	cmp x4, #136
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #120
+	bne .Lfailure
+
+	mov x2, x0
+	mov x3, #16
+	ld2 {v0.8b, v1.8b}, [x2], x3
+	ld2 {v2.4h, v3.4h}, [x2], 16
+	addv b4, v0.8b
+	addv b5, v1.8b
+	addv b6, v2.8b
+	addv b7, v3.8b
+	mov x4, v4.d[0]
+	cmp x4, #64
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #72
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #196
+	bne .Lfailure
+	mov x7, v7.d[0]
+	cmp x7, #180
+	bne .Lfailure
+
+	mov x2, x0
+	ld3 {v0.2s, v1.2s, v2.2s}, [x2]
+	addv b4, v0.8b
+	addv b5, v1.8b
+	addv b6, v2.8b
+	mov x4, v4.d[0]
+	cmp x4, #68
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #16
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #16
+	bne .Lfailure
+
+	mov x2, x0
+	ld4 {v0.4h, v1.4h, v2.4h, v3.4h}, [x2]
+	addv b4, v0.8b
+	addv b5, v1.8b
+	addv b6, v2.8b
+	addv b7, v3.8b
+	mov x4, v4.d[0]
+	cmp x4, #0
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #0
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #0
+	bne .Lfailure
+	mov x7, v7.d[0]
+	cmp x7, #0
+	bne .Lfailure
+
+	mov x2, x0
+	ld1 {v0.4s, v1.4s}, [x2]
+	addv b4, v0.16b
+	addv b5, v1.16b
+	mov x4, v4.d[0]
+	cmp x4, #136
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #120
+	bne .Lfailure
+
+	mov x2, x0
+	ld1 {v0.1d, v1.1d, v2.1d}, [x2]
+	addv b4, v0.8b
+	addv b5, v1.8b
+	addv b6, v2.8b
+	mov x4, v4.d[0]
+	cmp x4, #36
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #100
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #220
+	bne .Lfailure
+
+	mov x2, x0
+	ld1 {v0.1d, v1.1d, v2.1d, v3.1d}, [x2]
+	addv b4, v0.8b
+	addv b5, v1.8b
+	addv b6, v2.8b
+	mov x4, v4.d[0]
+	cmp x4, #36
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #100
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #220
+	bne .Lfailure
+
+	pass
+.Lfailure:
+	fail
diff --git a/sim/testsuite/sim/aarch64/ldn_single.s b/sim/testsuite/sim/aarch64/ldn_single.s
index 4c460fb..9681520 100644
--- a/sim/testsuite/sim/aarch64/ldn_single.s
+++ b/sim/testsuite/sim/aarch64/ldn_single.s
@@ -7,6 +7,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input:
 	.word 0x04030201
 	.word 0x08070605
diff --git a/sim/testsuite/sim/aarch64/ldnr.s b/sim/testsuite/sim/aarch64/ldnr.s
index a4bfffa..7126c46 100644
--- a/sim/testsuite/sim/aarch64/ldnr.s
+++ b/sim/testsuite/sim/aarch64/ldnr.s
@@ -7,6 +7,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input:
 	.word 0x04030201
 	.word 0x08070605
diff --git a/sim/testsuite/sim/aarch64/mla.s b/sim/testsuite/sim/aarch64/mla.s
index e0065e7..e3ea836 100644
--- a/sim/testsuite/sim/aarch64/mla.s
+++ b/sim/testsuite/sim/aarch64/mla.s
@@ -4,6 +4,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input:
 	.word 0x04030201
 	.word 0x08070605
diff --git a/sim/testsuite/sim/aarch64/mls.s b/sim/testsuite/sim/aarch64/mls.s
index a34a1aa..5c9e225 100644
--- a/sim/testsuite/sim/aarch64/mls.s
+++ b/sim/testsuite/sim/aarch64/mls.s
@@ -4,6 +4,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input:
 	.word 0x04030201
 	.word 0x08070605
diff --git a/sim/testsuite/sim/aarch64/stn_multiple.s b/sim/testsuite/sim/aarch64/stn_multiple.s
new file mode 100644
index 0000000..1a3f24d
--- /dev/null
+++ b/sim/testsuite/sim/aarch64/stn_multiple.s
@@ -0,0 +1,171 @@
+# mach: aarch64
+
+# Check the store multiple structure instructions: st1, st2, st3, st4.
+# Check the addressing modes: no offset, post-index immediate offset,
+# post-index register offset.
+
+.include "testutils.inc"
+
+	.data
+	.align 4
+input:
+	.word 0x04030201
+	.word 0x08070605
+	.word 0x0c0b0a09
+	.word 0x100f0e0d
+	.word 0xfcfdfeff
+	.word 0xf8f9fafb
+	.word 0xf4f5f6f7
+	.word 0xf0f1f2f3
+output:
+	.zero 64
+
+	start
+	adrp x0, input
+	add x0, x0, :lo12:input
+	adrp x1, output
+	add x1, x1, :lo12:output
+
+	mov x2, x0
+	ldr q0, [x2], 16
+	ldr q1, [x2]
+	mov x2, x0
+	ldr q2, [x2], 16
+	ldr q3, [x2]
+
+	mov x2, x1
+	mov x3, #16
+	st1 {v0.16b}, [x2], 16
+	st1 {v1.8h}, [x2], x3
+	mov x2, x1
+	ldr q4, [x2], 16
+	ldr q5, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	mov x4, v4.d[0]
+	cmp x4, #136
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #120
+	bne .Lfailure
+
+	mov x2, x1
+	mov x3, #16
+	st2 {v0.8b, v1.8b}, [x2], 16
+	st2 {v2.4h, v3.4h}, [x2], x3
+	mov x2, x1
+	ldr q4, [x2], 16
+	ldr q5, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	mov x4, v4.d[0]
+	cmp x4, #0
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #0
+	bne .Lfailure
+
+	mov x2, x1
+	st3 {v0.4s, v1.4s, v2.4s}, [x2]
+	ldr q4, [x2], 16
+	ldr q5, [x2], 16
+	ldr q6, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	addv b6, v6.16b
+	mov x4, v4.d[0]
+	cmp x4, #36
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #0
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #100
+	bne .Lfailure
+
+	mov x2, x1
+	st4 {v0.2d, v1.2d, v2.2d, v3.2d}, [x2]
+	ldr q4, [x2], 16
+	ldr q5, [x2], 16
+	ldr q6, [x2], 16
+	ldr q7, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	addv b6, v6.16b
+	addv b7, v7.16b
+	mov x4, v4.d[0]
+	cmp x4, #0
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #0
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #0
+	bne .Lfailure
+	mov x7, v7.d[0]
+	cmp x7, #0
+	bne .Lfailure
+
+	pass
+
+	mov x2, x1
+	st1 {v0.2s, v1.2s}, [x2], 16
+	st1 {v2.1d, v3.1d}, [x2]
+	mov x2, x1
+	ldr q4, [x2], 16
+	ldr q5, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	mov x4, v4.d[0]
+	cmp x4, #0
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #0
+	bne .Lfailure
+
+	mov x2, x1
+	st1 {v0.2d, v1.2d, v2.2d}, [x2]
+	mov x2, x1
+	ldr q4, [x2], 16
+	ldr q5, [x2], 16
+	ldr q6, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	addv b6, v6.16b
+	mov x4, v4.d[0]
+	cmp x4, #136
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #120
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #136
+	bne .Lfailure
+
+	mov x2, x1
+	st1 {v0.2d, v1.2d, v2.2d, v3.2d}, [x2]
+	mov x2, x1
+	ldr q4, [x2], 16
+	ldr q5, [x2], 16
+	ldr q6, [x2], 16
+	ldr q7, [x2]
+	addv b4, v4.16b
+	addv b5, v5.16b
+	addv b6, v6.16b
+	addv b7, v7.16b
+	mov x4, v4.d[0]
+	cmp x4, #136
+	bne .Lfailure
+	mov x5, v5.d[0]
+	cmp x5, #120
+	bne .Lfailure
+	mov x6, v6.d[0]
+	cmp x6, #136
+	bne .Lfailure
+	mov x7, v7.d[0]
+	cmp x7, #120
+	bne .Lfailure
+
+	pass
+.Lfailure:
+	fail
diff --git a/sim/testsuite/sim/aarch64/stn_single.s b/sim/testsuite/sim/aarch64/stn_single.s
index 2bd19cf..a24b084 100644
--- a/sim/testsuite/sim/aarch64/stn_single.s
+++ b/sim/testsuite/sim/aarch64/stn_single.s
@@ -7,6 +7,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input:
 	.word 0x04030201
 	.word 0x08070605
@@ -26,10 +28,10 @@ output:
 	add x1, x1, :lo12:output
 
 	mov x2, x0
-	ldr q0, [x2], 8
+	ldr q0, [x2], 16
 	ldr q1, [x2]
 	mov x2, x0
-	ldr q2, [x2], 8
+	ldr q2, [x2], 16
 	ldr q3, [x2]
 
 	mov x2, x1
@@ -61,9 +63,9 @@ output:
 	addv b5, v5.16b
 	mov x5, v4.d[0]
 	mov x6, v5.d[0]
-	cmp x5, #136
+	cmp x5, #200
 	bne .Lfailure
-	cmp x6, #8
+	cmp x6, #72
 	bne .Lfailure
 
 	mov x2, x1
@@ -82,11 +84,11 @@ output:
 	mov x4, v4.d[0]
 	mov x5, v5.d[0]
 	mov x6, v6.d[0]
-	cmp x4, #88
+	cmp x4, #120
 	bne .Lfailure
-	cmp x5, #200
+	cmp x5, #8
 	bne .Lfailure
-	cmp x6, #248
+	cmp x6, #24
 	bne .Lfailure
 
 	mov x2, x1
@@ -108,13 +110,13 @@ output:
 	mov x5, v5.d[0]
 	mov x6, v6.d[0]
 	mov x7, v7.d[0]
-	cmp x4, #104
+	cmp x4, #168
 	bne .Lfailure
-	cmp x5, #168
+	cmp x5, #232
 	bne .Lfailure
-	cmp x6, #232
+	cmp x6, #40
 	bne .Lfailure
-	cmp x7, #40
+	cmp x7, #104
 	bne .Lfailure
 
 	pass
diff --git a/sim/testsuite/sim/aarch64/sumulh.s b/sim/testsuite/sim/aarch64/sumulh.s
index 17f1ecd..d75e0c6 100644
--- a/sim/testsuite/sim/aarch64/sumulh.s
+++ b/sim/testsuite/sim/aarch64/sumulh.s
@@ -6,9 +6,6 @@
 
 .include "testutils.inc"
 
-	.data
-	.align 4
-
 	start
 
 	mov x0, #-2
diff --git a/sim/testsuite/sim/aarch64/uzp.s b/sim/testsuite/sim/aarch64/uzp.s
index 55e2cd7..851005e 100644
--- a/sim/testsuite/sim/aarch64/uzp.s
+++ b/sim/testsuite/sim/aarch64/uzp.s
@@ -4,6 +4,8 @@
 
 .include "testutils.inc"
 
+	.data
+	.align 4
 input1:
 	.word 0x04030201
 	.word 0x08070605