[X86][Btver2] Fix MicroOps counts for DPPS/YMM memory folded instructions

This was due to a misunderstanding over what llvm calls a micro-op (retirement unit) is actually called a macro-op on the AMD/Jaguar target. Folded loads don't affect num macro ops.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@328320 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ScheduleBtVer2.td b/lib/Target/X86/X86ScheduleBtVer2.td
index 47a4d67..3b67d47 100644 (file)
--- a/lib/Target/X86/X86ScheduleBtVer2.td
+++ b/lib/Target/X86/X86ScheduleBtVer2.td
@@ -398,7 +398,7 @@ def : InstRW<[JWriteDPPS], (instrs DPPSrri, VDPPSrri)>;
 def JWriteDPPSLd: SchedWriteRes<[JLAGU, JFPU0, JFPU1]> {
   let Latency = 16;
   let ResourceCycles = [1, 3, 3];
-  let NumMicroOps = 6;
+  let NumMicroOps = 5;
 }
 def : InstRW<[JWriteDPPSLd], (instrs DPPSrmi, VDPPSrmi)>;
 
@@ -508,7 +508,7 @@ def : InstRW<[JWriteFLogicY], (instrs VORPDYrr, VORPSYrr,
 def JWriteFLogicYLd: SchedWriteRes<[JLAGU, JFPU01, JFPX]> {
   let Latency = 6;
   let ResourceCycles = [1, 2, 2];
-  let NumMicroOps = 3;
+  let NumMicroOps = 2;
 }
 def : InstRW<[JWriteFLogicYLd], (instrs VORPDYrm, VORPSYrm,
                                         VXORPDYrm, VXORPSYrm,
@@ -525,7 +525,7 @@ def : InstRW<[JWriteVDPPSY], (instrs VDPPSYrri)>;
 def JWriteVDPPSYLd: SchedWriteRes<[JLAGU, JFPU1, JFPU0]> {
   let Latency = 17;
   let ResourceCycles = [1, 6, 6];
-  let NumMicroOps = 11;
+  let NumMicroOps = 10;
 }
 def : InstRW<[JWriteVDPPSYLd, ReadAfterLd], (instrs VDPPSYrmi)>;
 

diff --git a/test/tools/llvm-mca/X86/BtVer2/resources-avx1.s b/test/tools/llvm-mca/X86/BtVer2/resources-avx1.s
index 44e071e..1ce8f44 100644 (file)
--- a/test/tools/llvm-mca/X86/BtVer2/resources-avx1.s
+++ b/test/tools/llvm-mca/X86/BtVer2/resources-avx1.s
@@ -1087,7 +1087,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -     6.00    -     1.00    -      -      -      -      -      -            vblendvps       %ymm3, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00    -      -      -      -      -      -            vbroadcastf128  (%rax), %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -     4.00   1.00    -      -      -      -      -      -            vbroadcastsd    (%rax), %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            vbroadcastss    (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00   1.00    -     1.00    -      -      -      -      -      -            vbroadcastss    (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     4.00    -     1.00    -      -      -      -      -      -            vbroadcastss    (%rax), %ymm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vcmppd  $0, %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            vcmppd  $0, (%rax), %xmm1, %xmm2
@@ -1120,7 +1120,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -            vcvtpd2ps       %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -     1.00    -      -      -            vcvtpd2psx      (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -     4.00    -     1.00    -      -      -     2.00    -      -      -            vcvtpd2ps       %ymm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     4.00    -     1.00   1.00    -      -     2.00    -      -      -            vcvtpd2psy      (%rax), %xmm2
+# CHECK-NEXT:  -      -      -     4.00    -      -     1.00   1.00    -      -     2.00    -      -      -            vcvtpd2psy      (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -            vcvtps2dq       %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -     1.00    -      -      -            vcvtps2dq       (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     2.00    -      -      -            vcvtps2dq       %ymm0, %ymm2
@@ -1151,7 +1151,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -     1.00    -      -      -            vcvtss2si       (%rax), %rcx
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -            vcvttpd2dq      %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -     1.00    -      -      -            vcvttpd2dqx     (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     4.00    -      -     1.00    -      -      -     2.00    -      -      -            vcvttpd2dq      %ymm0, %xmm2
+# CHECK-NEXT:  -      -      -      -     4.00    -     1.00    -      -      -     2.00    -      -      -            vcvttpd2dq      %ymm0, %xmm2
 # CHECK-NEXT:  -      -      -     4.00    -      -     1.00   1.00    -      -     2.00    -      -      -            vcvttpd2dqy     (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -            vcvttps2dq      %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -     1.00    -      -      -            vcvttps2dq      (%rax), %xmm2
@@ -1180,7 +1180,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -     6.00   6.00    -      -      -      -      -      -      -            vdpps   $22, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -     6.00   6.00   1.00    -      -      -      -      -      -            vdpps   $22, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vextractf128    $1, %ymm0, %xmm2
-# CHECK-NEXT:  -      -      -      -     1.00   1.00    -      -      -     1.00    -      -      -      -            vextractf128    $1, %ymm0, (%rax)
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -     1.00    -      -      -      -            vextractf128    $1, %ymm0, (%rax)
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vextractps      $1, %xmm0, %ecx
 # CHECK-NEXT:  -      -      -     1.00    -      -     1.00   1.00    -     1.00    -      -      -      -            vextractps      $1, %xmm0, (%rax)
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vhaddpd %xmm0, %xmm1, %xmm2
@@ -1204,7 +1204,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vinsertps       $1, %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            vinsertps       $1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vlddqu  (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vlddqu  (%rax), %ymm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vlddqu  (%rax), %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -      -      -      -      -            vldmxcsr        (%rax)
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -     1.00    -     0.50   0.50    -            vmaskmovdqu     %xmm0, %xmm1
 # CHECK-NEXT:  -      -      -      -     2.00    -     1.00   1.00    -      -      -      -      -      -            vmaskmovpd      (%rax), %xmm0, %xmm2
@@ -1346,19 +1346,19 @@ vzeroupper
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            vorps   (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     2.00    -     2.00    -      -      -      -      -      -      -      -            vorps   %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     2.00    -     2.00    -     1.00    -      -      -      -      -      -            vorps   (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpabsb  %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpabsb  %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpabsb  (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpabsd  %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpabsd  (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpabsw  %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpabsw  (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpackssdw       %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpabsw  %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpabsw  (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpackssdw       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpackssdw       (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpacksswb       %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpacksswb       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpacksswb       (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpackusdw       %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpackusdw       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpackusdw       (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpackuswb       %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpackuswb       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpackuswb       (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpaddb  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpaddb  (%rax), %xmm1, %xmm2
@@ -1387,7 +1387,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpavgw  %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpavgw  (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     2.00   2.00    -            vpblendvb       %xmm3, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     2.00   2.00    -            vpblendvb       %xmm3, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     2.00   2.00    -            vpblendvb       %xmm3, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpblendw        $11, %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpblendw        $11, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -     1.00          vpclmulqdq      $11, %xmm0, %xmm1, %xmm2
@@ -1409,17 +1409,17 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpcmpgtw        %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpcmpgtw        (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vperm2f128      $1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -      -            vperm2f128      $1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -            vpermilpd       $1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -      -            vpermilpd       $1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00    -      -      -      -      -      -            vperm2f128      $1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vpermilpd       $1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00    -      -      -      -      -      -            vpermilpd       $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     4.00    -      -      -      -      -      -      -      -            vpermilpd       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     4.00   1.00    -      -      -      -      -      -            vpermilpd       (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -            vpermilpd       $1, %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -     2.00   1.00    -      -      -      -      -      -            vpermilpd       $1, (%rax), %ymm2
 # CHECK-NEXT:  -      -      -      -      -     6.00    -      -      -      -      -      -      -      -            vpermilpd       %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -     6.00   1.00    -      -      -      -      -      -            vpermilpd       (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -            vpermilps       $1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -      -            vpermilps       $1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vpermilps       $1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00    -      -      -      -      -      -            vpermilps       $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     4.00    -      -      -      -      -      -      -      -            vpermilps       %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     4.00   1.00    -      -      -      -      -      -            vpermilps       (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -            vpermilps       $1, %ymm0, %ymm2
@@ -1429,7 +1429,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpextrb $1, %xmm0, %ecx
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -     1.00    -     0.50   0.50    -            vpextrb $1, %xmm0, (%rax)
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpextrd $1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -     1.00    -     0.50   0.50    -            vpextrd $1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -     1.00    -     0.50   0.50    -            vpextrd $1, %xmm0, (%rax)
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpextrq $1, %xmm0, %rcx
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -     1.00    -     0.50   0.50    -            vpextrq $1, %xmm0, (%rax)
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpextrw $1, %xmm0, %ecx
@@ -1445,28 +1445,28 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vphsubd %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vphsubd (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vphsubsw        %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vphsubsw        (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vphsubw %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vphsubw (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpinsrb $1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpinsrb $1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpinsrd $1, %eax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpinsrd $1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpinsrq $1, %rax, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpinsrq $1, (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpinsrw $1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vphsubsw        (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vphsubw %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vphsubw (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpinsrb $1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpinsrb $1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpinsrd $1, %eax, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpinsrd $1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpinsrq $1, %rax, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpinsrq $1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpinsrw $1, %eax, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpinsrw $1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -     1.00          vpmaddubsw      %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     1.00          vpmaddubsw      (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -     1.00          vpmaddwd        %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     1.00          vpmaddwd        (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpmaxsb %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpmaxsb %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpmaxsb (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpmaxsd %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpmaxsd %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpmaxsd (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpmaxsw %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpmaxsw %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpmaxsw (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpmaxub %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpmaxub %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpmaxub (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpmaxud %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpmaxud (%rax), %xmm1, %xmm2
@@ -1531,7 +1531,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -      -     4.00   1.00    -      -      -      -      -      -            vpshufb (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpshufd $1, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpshufd $1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpshufhw        $1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpshufhw        $1, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpshufhw        $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpshuflw        $1, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     0.50   0.50    -            vpshuflw        $1, (%rax), %xmm2
@@ -1589,7 +1589,7 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -      -     2.00   2.00   1.00    -      -      -      -      -      -            vptest  (%rax), %ymm1
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpunpckhbw      %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpunpckhbw      (%rax), %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpunpckhdq      %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.50   0.50    -            vpunpckhdq      %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpunpckhdq      (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     0.50   0.50    -            vpunpckhqdq     %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -            vpunpckhqdq     (%rax), %xmm1, %xmm2
@@ -1630,13 +1630,13 @@ vzeroupper
 # CHECK-NEXT:  -      -      -      -     1.00    -     1.00    -      -      -      -      -      -      -            vrsqrtss        %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -     1.00    -     1.00   1.00    -      -      -      -      -      -            vrsqrtss        (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vshufpd $1, %xmm0, %xmm1, %xmm2
-# CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            vshufpd $1, (%rax), %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00   1.00    -     1.00    -      -      -      -      -      -            vshufpd $1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -            vshufpd $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -     2.00    -     1.00    -      -      -      -      -      -            vshufpd $1, (%rax), %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -     1.00    -      -     1.00    -      -      -      -      -      -      -            vshufps $1, %xmm0, %xmm1, %xmm2
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -            vshufps $1, %xmm0, %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -     1.00    -      -     1.00   1.00    -      -      -      -      -      -            vshufps $1, (%rax), %xmm1, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     2.00    -      -      -      -      -      -      -      -            vshufps $1, %ymm0, %ymm1, %ymm2
-# CHECK-NEXT:  -      -      -      -      -     2.00    -     1.00    -      -      -      -      -      -            vshufps $1, (%rax), %ymm1, %ymm2
+# CHECK-NEXT:  -      -      -      -      -      -     2.00   1.00    -      -      -      -      -      -            vshufps $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     21.00   -     1.00    -      -      -      -      -      -      -            vsqrtpd %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -     21.00   -     1.00   1.00    -      -      -      -      -      -            vsqrtpd (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -     54.00   -     1.00    -      -      -      -      -      -      -            vsqrtpd %ymm0, %ymm2

diff --git a/test/tools/llvm-mca/X86/BtVer2/resources-sse41.s b/test/tools/llvm-mca/X86/BtVer2/resources-sse41.s
index fb3a9cd..7c04155 100644 (file)
--- a/test/tools/llvm-mca/X86/BtVer2/resources-sse41.s
+++ b/test/tools/llvm-mca/X86/BtVer2/resources-sse41.s
@@ -168,18 +168,18 @@ roundss     $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -     0.99   0.01   0.99   0.01    -      -      -      -      -      -      -            blendps $11, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -     0.01   0.99   0.01   0.99   1.00    -      -      -      -      -      -            blendps $11, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -     3.94   0.06   0.99   0.01    -      -      -      -      -      -      -            blendvpd        %xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.06   3.94   0.01   0.99   1.00    -      -      -      -      -      -            blendvpd        %xmm0, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.06   3.94   1.00    -      -      -      -      -      -      -      -            blendvps        %xmm0, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     3.94   0.06   0.01   0.99   1.00    -      -      -      -      -      -            blendvps        %xmm0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -     0.06   3.94   1.00    -     1.00    -      -      -      -      -      -            blendvpd        %xmm0, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -     4.00    -     0.99   0.01    -      -      -      -      -      -      -            blendvps        %xmm0, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -     4.00   0.01   0.99   1.00    -      -      -      -      -      -            blendvps        %xmm0, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -            dppd    $22, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     3.00   3.00   1.00    -      -      -      -      -      -            dppd    $22, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     3.00   3.00    -      -      -      -      -      -      -            dpps    $22, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     3.00   3.00   1.00    -      -      -      -      -      -            dpps    $22, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -     0.99   0.01   0.99   0.01    -      -      -      -      -      -      -            extractps       $1, %xmm0, %ecx
-# CHECK-NEXT:  -      -      -     0.01   0.99    -     1.00   1.00    -     1.00    -      -      -      -            extractps       $1, %xmm0, (%rax)
-# CHECK-NEXT:  -      -      -     0.99   0.01   1.00    -      -      -      -      -      -      -      -            insertps        $1, %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -     0.01   0.99    -     1.00   1.00    -      -      -      -      -      -            insertps        $1, (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     0.99   0.01    -            movntdqa        (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -     1.00    -     1.00    -      -      -      -      -      -      -            extractps       $1, %xmm0, %ecx
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -     1.00    -      -      -      -            extractps       $1, %xmm0, (%rax)
+# CHECK-NEXT:  -      -      -      -     1.00    -     1.00    -      -      -      -      -      -      -            insertps        $1, %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -            insertps        $1, (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     0.01   0.99   1.00    -      -      -     0.99   0.01    -            movntdqa        (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -     2.00          mpsadbw $1, %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -      -     2.00          mpsadbw $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     0.99   0.01    -            packusdw        %xmm0, %xmm2
@@ -211,10 +211,10 @@ roundss     $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00    -            pmaxsd  (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -            pmaxud  %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00    -            pmaxud  (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -     1.00    -            pmaxuw  %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -     1.00    -      -            pmaxuw  (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pminsb  %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     1.00    -      -            pminsb  (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -            pmaxuw  %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00    -            pmaxuw  (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -            pminsb  %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pminsb  (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pminsd  %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pminsd  (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pminud  %xmm0, %xmm2
@@ -224,12 +224,12 @@ roundss     $1, (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -            pmovsxbd        %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00    -            pmovsxbd        (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -     1.00    -      -            pmovsxbq        %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     1.00    -      -            pmovsxbq        (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pmovsxbw        %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     1.00    -      -            pmovsxbw        (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pmovsxdq        %xmm0, %xmm2
-# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -     1.00    -      -            pmovsxdq        (%rax), %xmm2
-# CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -     1.00    -            pmovsxwd        %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00   1.00    -      -      -      -     1.00    -            pmovsxbq        (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -            pmovsxbw        %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pmovsxbw        (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -            pmovsxdq        %xmm0, %xmm2
+# CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pmovsxdq        (%rax), %xmm2
+# CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -            pmovsxwd        %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pmovsxwd        (%rax), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -     1.00    -      -            pmovsxwq        %xmm0, %xmm2
 # CHECK-NEXT:  -      -      -      -      -     1.00    -     1.00    -      -      -      -     1.00    -            pmovsxwq        (%rax), %xmm2