diff mbox series

[1/4] x86: optimize pre-AVX512 {,V}PCMPEQQ with identical sources

Message ID 503caac8-8824-823a-81c2-762cba207cb6@suse.com
State Accepted
Headers
Series x86: some more optimization plus a new pseudo insn form |

Checks

Context Check Description
snail/binutils-gdb-check success Github commit url

Commit Message

Jan Beulich June 16, 2023, 7:30 a.m. UTC 
  The {,V}PCMPEQD alternative is 1 byte shorter in many cases.
---
It's not really clear whether the same would be worthwhile for AVX512
forms: Some could be expressed via KXNOR* (when no masking is in effect)
or KOR* (when masking is in effect), but others cannot. And while in
pre-AVX512 code these patterns are likely to be used to produce all-ones
idioms, this looks less likely in AVX512.
diff mbox series

Patch

--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -4563,6 +4563,23 @@  optimize_encoding (void)
 	i.types[j].bitfield.disp8
 	  = fits_in_disp8 (i.op[j].disps->X_add_number);
     }
+  else if (optimize_for_space
+	   && i.tm.base_opcode == 0x29
+	   && i.tm.opcode_space == SPACE_0F38
+	   && i.operands == i.reg_operands
+	   && i.op[0].regs == i.op[1].regs
+	   && (!i.tm.opcode_modifier.vex
+	       || !(i.op[0].regs->reg_flags & RegRex))
+	   && !is_evex_encoding (&i.tm))
+    {
+      /* Optimize: -Os:
+         pcmpeqq %xmmN, %xmmN          -> pcmpeqd %xmmN, %xmmN
+         vpcmpeqq %xmmN, %xmmN, %xmmM  -> vpcmpeqd %xmmN, %xmmN, %xmmM (N < 8)
+         vpcmpeqq %ymmN, %ymmN, %ymmM  -> vpcmpeqd %ymmN, %ymmN, %ymmM (N < 8)
+       */
+      i.tm.opcode_space = SPACE_0F;
+      i.tm.base_opcode = 0x76;
+    }
 }
 
 /* Return non-zero for load instruction.  */
--- a/gas/testsuite/gas/i386/optimize-2.d
+++ b/gas/testsuite/gas/i386/optimize-2.d
@@ -161,4 +161,7 @@  Disassembly of section .text:
  +[a-f0-9]+:	62 .*	vporq  \(%eax\)\{1to2\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxord \(%eax\)\{1to4\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxorq \(%eax\)\{1to4\},%ymm2,%ymm3
+ +[a-f0-9]+:	66 .*	pcmpeqd %xmm2,%xmm2
+ +[a-f0-9]+:	c5 .*	vpcmpeqd %xmm2,%xmm2,%xmm0
+ +[a-f0-9]+:	c5 .*	vpcmpeqd %ymm2,%ymm2,%ymm0
 #pass
--- a/gas/testsuite/gas/i386/optimize-2.s
+++ b/gas/testsuite/gas/i386/optimize-2.s
@@ -180,3 +180,7 @@  _start:
 	vporq		(%eax){1to2}, %xmm2, %xmm3
 	vpxord		(%eax){1to4}, %xmm2, %xmm3
 	vpxorq		(%eax){1to4}, %ymm2, %ymm3
+
+	pcmpeqq		%xmm2, %xmm2
+	vpcmpeqq	%xmm2, %xmm2, %xmm0
+	vpcmpeqq	%ymm2, %ymm2, %ymm0
--- a/gas/testsuite/gas/i386/optimize-2b.d
+++ b/gas/testsuite/gas/i386/optimize-2b.d
@@ -162,4 +162,7 @@  Disassembly of section .text:
  +[a-f0-9]+:	62 .*	vporq  \(%eax\)\{1to2\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxord \(%eax\)\{1to4\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxorq \(%eax\)\{1to4\},%ymm2,%ymm3
+ +[a-f0-9]+:	66 .*	pcmpeqq %xmm2,%xmm2
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %xmm2,%xmm2,%xmm0
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %ymm2,%ymm2,%ymm0
 #pass
--- a/gas/testsuite/gas/i386/x86-64-optimize-3.d
+++ b/gas/testsuite/gas/i386/x86-64-optimize-3.d
@@ -199,4 +199,10 @@  Disassembly of section .text:
  +[a-f0-9]+:	62 .*	vporq  \(%rax\)\{1to2\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxord \(%rax\)\{1to4\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxorq \(%rax\)\{1to4\},%ymm2,%ymm3
+ +[a-f0-9]+:	66 .*	pcmpeqd %xmm2,%xmm2
+ +[a-f0-9]+:	c5 .*	vpcmpeqd %xmm2,%xmm2,%xmm0
+ +[a-f0-9]+:	c5 .*	vpcmpeqd %ymm2,%ymm2,%ymm0
+ +[a-f0-9]+:	66 .*	pcmpeqd %xmm12,%xmm12
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %xmm12,%xmm12,%xmm0
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %ymm12,%ymm12,%ymm0
 #pass
--- a/gas/testsuite/gas/i386/x86-64-optimize-3.s
+++ b/gas/testsuite/gas/i386/x86-64-optimize-3.s
@@ -221,3 +221,11 @@  _start:
 	vporq		(%rax){1to2}, %xmm2, %xmm3
 	vpxord		(%rax){1to4}, %xmm2, %xmm3
 	vpxorq		(%rax){1to4}, %ymm2, %ymm3
+
+	pcmpeqq		%xmm2, %xmm2
+	vpcmpeqq	%xmm2, %xmm2, %xmm0
+	vpcmpeqq	%ymm2, %ymm2, %ymm0
+
+	pcmpeqq		%xmm12, %xmm12
+	vpcmpeqq	%xmm12, %xmm12, %xmm0
+	vpcmpeqq	%ymm12, %ymm12, %ymm0
--- a/gas/testsuite/gas/i386/x86-64-optimize-3b.d
+++ b/gas/testsuite/gas/i386/x86-64-optimize-3b.d
@@ -200,4 +200,10 @@  Disassembly of section .text:
  +[a-f0-9]+:	62 .*	vporq  \(%rax\)\{1to2\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxord \(%rax\)\{1to4\},%xmm2,%xmm3
  +[a-f0-9]+:	62 .*	vpxorq \(%rax\)\{1to4\},%ymm2,%ymm3
+ +[a-f0-9]+:	66 .*	pcmpeqq %xmm2,%xmm2
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %xmm2,%xmm2,%xmm0
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %ymm2,%ymm2,%ymm0
+ +[a-f0-9]+:	66 .*	pcmpeqq %xmm12,%xmm12
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %xmm12,%xmm12,%xmm0
+ +[a-f0-9]+:	c4 .*	vpcmpeqq %ymm12,%ymm12,%ymm0
 #pass
--- a/opcodes/i386-opc.tbl
+++ b/opcodes/i386-opc.tbl
@@ -1363,7 +1363,7 @@  pblendvb, 0x664c, AVX, Modrm|Vex128|Spac
 pblendvb, 0x660f3810, SSE4_1, Modrm|NoSuf, { Acc|Xmmword, RegXMM|Unspecified|BaseIndex, RegXMM }
 pblendvb, 0x660f3810, SSE4_1, Modrm|NoSuf, { RegXMM|Unspecified|BaseIndex, RegXMM }
 pblendw<sse41>, 0x660f3a0e, <sse41:cpu>, Modrm|<sse41:attr>|<sse41:vvvv>|NoSuf, { Imm8|Imm8S, RegXMM|Unspecified|BaseIndex, RegXMM }
-pcmpeqq<sse41>, 0x660f3829, <sse41:cpu>, Modrm|<sse41:attr>|<sse41:vvvv>|NoSuf, { RegXMM|Unspecified|BaseIndex, RegXMM }
+pcmpeqq<sse41>, 0x660f3829, <sse41:cpu>, Modrm|<sse41:attr>|<sse41:vvvv>|NoSuf|Optimize, { RegXMM|Unspecified|BaseIndex, RegXMM }
 pextr<bw><sse41>, 0x660f3a14 | <bw:opc>, <sse41:cpu>, RegMem|<sse41:attr>|NoSuf|IgnoreSize|NoRex64, { Imm8, RegXMM, Reg32|Reg64 }
 pextr<bw><sse41>, 0x660f3a14 | <bw:opc>, <sse41:cpu>, Modrm|<sse41:attr>|NoSuf, { Imm8, RegXMM, <bw:elem>|Unspecified|BaseIndex }
 pextrd<sse41>, 0x660f3a16, <sse41:cpu>, Modrm|<sse41:attr>|NoSuf|IgnoreSize, { Imm8, RegXMM, Reg32|Unspecified|BaseIndex }
@@ -1592,7 +1592,7 @@  vpblendvb, 0x664c, AVX|AVX2, Modrm|Vex|S
 vpblendw, 0x660e, AVX|AVX2, Modrm|Vex|Space0F3A|VexVVVV|VexWIG|CheckOperandSize|NoSuf, { Imm8|Imm8S, Unspecified|BaseIndex|RegXMM|RegYMM, RegXMM|RegYMM, RegXMM|RegYMM }
 vpcmpeq<bw>, 0x6674 | <bw:opc>, AVX|AVX2, Modrm|C|Vex|Space0F|VexVVVV|VexWIG|CheckOperandSize|NoSuf, { Unspecified|BaseIndex|RegXMM|RegYMM, RegXMM|RegYMM, RegXMM|RegYMM }
 vpcmpeqd, 0x6676, AVX|AVX2, Modrm|C|Vex|Space0F|VexVVVV|VexWIG|CheckOperandSize|NoSuf, { Unspecified|BaseIndex|RegXMM|RegYMM, RegXMM|RegYMM, RegXMM|RegYMM }
-vpcmpeqq, 0x6629, AVX|AVX2, Modrm|Vex|Space0F38|VexVVVV|VexWIG|CheckOperandSize|NoSuf, { Unspecified|BaseIndex|RegXMM|RegYMM, RegXMM|RegYMM, RegXMM|RegYMM }
+vpcmpeqq, 0x6629, AVX|AVX2, Modrm|Vex|Space0F38|VexVVVV|VexWIG|CheckOperandSize|NoSuf|Optimize, { Unspecified|BaseIndex|RegXMM|RegYMM, RegXMM|RegYMM, RegXMM|RegYMM }
 vpcmpestri, 0x6661, AVX|No64, Modrm|Vex|Space0F3A|NoSuf, { Imm8, Unspecified|BaseIndex|RegXMM, RegXMM }
 vpcmpestri, 0x6661, AVX|x64, Modrm|Vex|Space0F3A|IgnoreSize|No_bSuf|No_wSuf|No_sSuf, { Imm8, Xmmword|Unspecified|BaseIndex|RegXMM, RegXMM }
 vpcmpestrm, 0x6660, AVX|No64, Modrm|Vex|Space0F3A|NoSuf, { Imm8, Unspecified|BaseIndex|RegXMM, RegXMM }