x86: optimize {,V}EXTRACT{F,I}{128,32x{4,8},64x{2,4}} with immediate 0

They, too, are equivalent to simple moves, which are up to 3 bytes
shorter to encode (and maybe also cheaper to execute).

8 files changed, 222 insertions, 0 deletions

diff --git a/gas/testsuite/gas/i386/optimize-1.d b/gas/testsuite/gas/i386/optimize-1.d
index c2802b894b9..fffa655d330 100644
--- a/gas/testsuite/gas/i386/optimize-1.d
+++ b/gas/testsuite/gas/i386/optimize-1.d
@@ -170,6 +170,26 @@ Disassembly of section .text:
  +[a-f0-9]+:	f3 .*	movss  %xmm1,\(%edx\)
  +[a-f0-9]+:	c5 .*	vmovd  %xmm1,%edx
  +[a-f0-9]+:	c5 .*	vmovss %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovupd %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovups %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovupd %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
  +[a-f0-9]+:	0f ba e0 0f          	bt     \$0xf,%eax
  +[a-f0-9]+:	66 0f ba e0 10       	bt     \$0x10,%ax
  +[a-f0-9]+:	0f ba f8 0f          	btc    \$0xf,%eax
diff --git a/gas/testsuite/gas/i386/optimize-1.s b/gas/testsuite/gas/i386/optimize-1.s
index 40546e8c157..afdeed4c423 100644
--- a/gas/testsuite/gas/i386/optimize-1.s
+++ b/gas/testsuite/gas/i386/optimize-1.s
@@ -199,6 +199,31 @@ _start:
 	vextractps	$0, %xmm1, %edx
 	vextractps	$0, %xmm1, (%edx)
 
+	vextractf128 $0, %ymm1, %xmm2
+	vextractf128 $0, %ymm1, (%edx)
+	vextracti128 $0, %ymm1, %xmm2
+	vextracti128 $0, %ymm1, (%edx)
+
+	vextractf32x4 $0, %ymm1, %xmm2
+	vextractf32x4 $0, %ymm1, (%edx)
+	vextracti32x4 $0, %ymm1, %xmm2
+	vextracti32x4 $0, %ymm1, (%edx)
+
+	vextractf64x2 $0, %ymm1, %xmm2
+	vextractf64x2 $0, %ymm1, (%edx)
+	vextracti64x2 $0, %ymm1, %xmm2
+	vextracti64x2 $0, %ymm1, (%edx)
+
+	vextractf32x8 $0, %zmm1, %ymm2
+	vextractf32x8 $0, %zmm1, (%edx)
+	vextracti32x8 $0, %zmm1, %ymm2
+	vextracti32x8 $0, %zmm1, (%edx)
+
+	vextractf64x4 $0, %zmm1, %ymm2
+	vextractf64x4 $0, %zmm1, (%edx)
+	vextracti64x4 $0, %zmm1, %ymm2
+	vextracti64x4 $0, %zmm1, (%edx)
+
 	bt	$15, %ax
 	bt	$16, %ax
 	btc	$15, %ax
diff --git a/gas/testsuite/gas/i386/optimize-1a.d b/gas/testsuite/gas/i386/optimize-1a.d
index 563567f3f30..aff37681a00 100644
--- a/gas/testsuite/gas/i386/optimize-1a.d
+++ b/gas/testsuite/gas/i386/optimize-1a.d
@@ -171,6 +171,26 @@ Disassembly of section .text:
  +[a-f0-9]+:	f3 .*	movss  %xmm1,\(%edx\)
  +[a-f0-9]+:	c5 .*	vmovd  %xmm1,%edx
  +[a-f0-9]+:	c5 .*	vmovss %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovupd %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovups %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovupd %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
  +[a-f0-9]+:	0f ba e0 0f          	bt     \$0xf,%eax
  +[a-f0-9]+:	66 0f ba e0 10       	bt     \$0x10,%ax
  +[a-f0-9]+:	0f ba f8 0f          	btc    \$0xf,%eax
diff --git a/gas/testsuite/gas/i386/optimize-4.d b/gas/testsuite/gas/i386/optimize-4.d
index 2449bc55ad9..fbc142a424c 100644
--- a/gas/testsuite/gas/i386/optimize-4.d
+++ b/gas/testsuite/gas/i386/optimize-4.d
@@ -170,6 +170,26 @@ Disassembly of section .text:
  +[a-f0-9]+:	f3 .*	movss  %xmm1,\(%edx\)
  +[a-f0-9]+:	c5 .*	vmovd  %xmm1,%edx
  +[a-f0-9]+:	c5 .*	vmovss %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovupd %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovups %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovupd %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
  +[a-f0-9]+:	0f ba e0 0f          	bt     \$0xf,%eax
  +[a-f0-9]+:	66 0f ba e0 10       	bt     \$0x10,%ax
  +[a-f0-9]+:	0f ba f8 0f          	btc    \$0xf,%eax
diff --git a/gas/testsuite/gas/i386/optimize-5.d b/gas/testsuite/gas/i386/optimize-5.d
index d60d8421cbf..a40022a3420 100644
--- a/gas/testsuite/gas/i386/optimize-5.d
+++ b/gas/testsuite/gas/i386/optimize-5.d
@@ -170,6 +170,26 @@ Disassembly of section .text:
  +[a-f0-9]+:	f3 .*	movss  %xmm1,\(%edx\)
  +[a-f0-9]+:	c5 .*	vmovd  %xmm1,%edx
  +[a-f0-9]+:	c5 .*	vmovss %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovups %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovupd %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %xmm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovaps %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovups %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovapd %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovupd %ymm1,\(%edx\)
+ +[a-f0-9]+:	c5 .*	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 .*	vmovdqu %ymm1,\(%edx\)
  +[a-f0-9]+:	0f ba e0 0f          	bt     \$0xf,%eax
  +[a-f0-9]+:	66 0f ba e0 10       	bt     \$0x10,%ax
  +[a-f0-9]+:	0f ba f8 0f          	btc    \$0xf,%eax
diff --git a/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.d b/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.d
new file mode 100644
index 00000000000..739a2fbb54a
--- /dev/null
+++ b/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.d
@@ -0,0 +1,59 @@
+#as: -O
+#objdump: -drw
+#name: x86-64 VEXTRACT{F,I}<nn> optimized encoding with -msse2avx
+
+.*: +file format .*
+
+
+Disassembly of section .text:
+
+0+ <vextract_128>:
+ +[a-f0-9]+:	c5 f8 29 ca          	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 f8 11 0a          	vmovups %xmm1,\(%rdx\)
+ +[a-f0-9]+:	c5 f9 7f ca          	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 fa 7f 0a          	vmovdqu %xmm1,\(%rdx\)
+
+0+[a-f0-9]+ <vextract_NNxM_XMM>:
+ +[a-f0-9]+:	c5 f8 29 ca          	vmovaps %xmm1,%xmm2
+ +[a-f0-9]+:	c5 f8 11 0a          	vmovups %xmm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 7c 08 29 ca    	vmovaps %xmm17,%xmm2
+ +[a-f0-9]+:	62 e1 7c 08 11 0a    	vmovups %xmm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 7c 08 11 0a    	vmovups %xmm1,\(%r18\)
+ +[a-f0-9]+:	c5 f9 29 ca          	vmovapd %xmm1,%xmm2
+ +[a-f0-9]+:	c5 f9 11 0a          	vmovupd %xmm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 fd 08 29 ca    	vmovapd %xmm17,%xmm2
+ +[a-f0-9]+:	62 e1 fd 08 11 0a    	vmovupd %xmm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 fd 08 11 0a    	vmovupd %xmm1,\(%r18\)
+ +[a-f0-9]+:	c5 f9 7f ca          	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 fa 7f 0a          	vmovdqu %xmm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 7d 08 7f ca    	vmovdqa32 %xmm17,%xmm2
+ +[a-f0-9]+:	62 e1 7e 08 7f 0a    	vmovdqu32 %xmm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 7e 08 7f 0a    	vmovdqu32 %xmm1,\(%r18\)
+ +[a-f0-9]+:	c5 f9 7f ca          	vmovdqa %xmm1,%xmm2
+ +[a-f0-9]+:	c5 fa 7f 0a          	vmovdqu %xmm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 fd 08 7f ca    	vmovdqa64 %xmm17,%xmm2
+ +[a-f0-9]+:	62 e1 fe 08 7f 0a    	vmovdqu64 %xmm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 fe 08 7f 0a    	vmovdqu64 %xmm1,\(%r18\)
+
+0+[a-f0-9]+ <vextract_NNxM_YMM>:
+ +[a-f0-9]+:	c5 fc 29 ca          	vmovaps %ymm1,%ymm2
+ +[a-f0-9]+:	c5 fc 11 0a          	vmovups %ymm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 7c 28 29 ca    	vmovaps %ymm17,%ymm2
+ +[a-f0-9]+:	62 e1 7c 28 11 0a    	vmovups %ymm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 7c 28 11 0a    	vmovups %ymm1,\(%r18\)
+ +[a-f0-9]+:	c5 fd 29 ca          	vmovapd %ymm1,%ymm2
+ +[a-f0-9]+:	c5 fd 11 0a          	vmovupd %ymm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 fd 28 29 ca    	vmovapd %ymm17,%ymm2
+ +[a-f0-9]+:	62 e1 fd 28 11 0a    	vmovupd %ymm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 fd 28 11 0a    	vmovupd %ymm1,\(%r18\)
+ +[a-f0-9]+:	c5 fd 7f ca          	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 fe 7f 0a          	vmovdqu %ymm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 7d 28 7f ca    	vmovdqa32 %ymm17,%ymm2
+ +[a-f0-9]+:	62 e1 7e 28 7f 0a    	vmovdqu32 %ymm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 7e 28 7f 0a    	vmovdqu32 %ymm1,\(%r18\)
+ +[a-f0-9]+:	c5 fd 7f ca          	vmovdqa %ymm1,%ymm2
+ +[a-f0-9]+:	c5 fe 7f 0a          	vmovdqu %ymm1,\(%rdx\)
+ +[a-f0-9]+:	62 e1 fd 28 7f ca    	vmovdqa64 %ymm17,%ymm2
+ +[a-f0-9]+:	62 e1 fe 28 7f 0a    	vmovdqu64 %ymm17,\(%rdx\)
+ +[a-f0-9]+:	62 f9 fe 28 7f 0a    	vmovdqu64 %ymm1,\(%r18\)
+#pass
diff --git a/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.s b/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.s
new file mode 100644
index 00000000000..5da8265d345
--- /dev/null
+++ b/gas/testsuite/gas/i386/x86-64-optimize-vextractNN.s
@@ -0,0 +1,57 @@
+	.text
+vextract_128:
+	vextractf128 $0, %ymm1, %xmm2
+	vextractf128 $0, %ymm1, (%rdx)
+
+	vextracti128 $0, %ymm1, %xmm2
+	vextracti128 $0, %ymm1, (%rdx)
+
+vextract_NNxM_XMM:
+	vextractf32x4 $0, %ymm1, %xmm2
+	vextractf32x4 $0, %ymm1, (%rdx)
+	vextractf32x4 $0, %ymm17, %xmm2
+	vextractf32x4 $0, %ymm17, (%rdx)
+	vextractf32x4 $0, %ymm1, (%r18)
+
+	vextractf64x2 $0, %ymm1, %xmm2
+	vextractf64x2 $0, %ymm1, (%rdx)
+	vextractf64x2 $0, %ymm17, %xmm2
+	vextractf64x2 $0, %ymm17, (%rdx)
+	vextractf64x2 $0, %ymm1, (%r18)
+
+	vextracti32x4 $0, %ymm1, %xmm2
+	vextracti32x4 $0, %ymm1, (%rdx)
+	vextracti32x4 $0, %ymm17, %xmm2
+	vextracti32x4 $0, %ymm17, (%rdx)
+	vextracti32x4 $0, %ymm1, (%r18)
+
+	vextracti64x2 $0, %ymm1, %xmm2
+	vextracti64x2 $0, %ymm1, (%rdx)
+	vextracti64x2 $0, %ymm17, %xmm2
+	vextracti64x2 $0, %ymm17, (%rdx)
+	vextracti64x2 $0, %ymm1, (%r18)
+
+vextract_NNxM_YMM:
+	vextractf32x8 $0, %zmm1, %ymm2
+	vextractf32x8 $0, %zmm1, (%rdx)
+	vextractf32x8 $0, %zmm17, %ymm2
+	vextractf32x8 $0, %zmm17, (%rdx)
+	vextractf32x8 $0, %zmm1, (%r18)
+
+	vextractf64x4 $0, %zmm1, %ymm2
+	vextractf64x4 $0, %zmm1, (%rdx)
+	vextractf64x4 $0, %zmm17, %ymm2
+	vextractf64x4 $0, %zmm17, (%rdx)
+	vextractf64x4 $0, %zmm1, (%r18)
+
+	vextracti32x8 $0, %zmm1, %ymm2
+	vextracti32x8 $0, %zmm1, (%rdx)
+	vextracti32x8 $0, %zmm17, %ymm2
+	vextracti32x8 $0, %zmm17, (%rdx)
+	vextracti32x8 $0, %zmm1, (%r18)
+
+	vextracti64x4 $0, %zmm1, %ymm2
+	vextracti64x4 $0, %zmm1, (%rdx)
+	vextracti64x4 $0, %zmm17, %ymm2
+	vextracti64x4 $0, %zmm17, (%rdx)
+	vextracti64x4 $0, %zmm1, (%r18)
diff --git a/gas/testsuite/gas/i386/x86-64.exp b/gas/testsuite/gas/i386/x86-64.exp
index 61fafd07a8c..ad713dc59ae 100644
--- a/gas/testsuite/gas/i386/x86-64.exp
+++ b/gas/testsuite/gas/i386/x86-64.exp
@@ -596,6 +596,7 @@ run_list_test "x86-64-optimize-pextr" "-O -aln"
 run_dump_test "x86-64-optimize-pextr"
 run_list_test "x86-64-optimize-extractps" "-O -aln"
 run_dump_test "x86-64-optimize-extractps"
+run_dump_test "x86-64-optimize-vextractNN"
 run_dump_test "x86-64-apx-ndd-optimize"
 run_dump_test "x86-64-align-branch-1a"
 run_dump_test "x86-64-align-branch-1b"