[X86] Update the haswell and broadwell scheduler information for gather instructions

[craig.topper]

Summary

Broadwell was missing half the gather instructions. Both models
had some mixups in the resource costs and number of uops.

I've updated here based on what I think the original IACA source
says with some cross checking against the microcode.

I'm not sure about latency as the IACA source I have doesn't have
that information. So I'm using the latency from uops.info.

I plan to update Skylake models as well, but I'll do that in a
separate patch.

[RKSimon]

PR35676 discusses the missing Haswell/Broadwell gather schedules
PR35675 discusses poor Skylake (Client/Server) gather schedules

[andreadb]

The change looks pretty mechanical, and I trust that the new numbers are correct.

I noticed that gather instructions in tablegen are all associated with Sched<[WriteLoad]> (see multiclass avx2_gather in X86InstrSSE.td).
A quick grep of GATHER in directory Target/X86 shows how most (if not all?) avx2 processor models currently introduce InstRW definitions to override the scheduling class of gather instructions.
Basically, WriteLoad is clearly not a good choice for gather instructions. Presumably that is the reason why there is a FIXME comment on avx2_gather (in X86InstrSSE.td) that writes "Improve sheduling of gather instructions.".

Here is the idea: why don't we just add new writes for gather instructions (ideally a new write per each type of gather)?. I think that we shall consider this alternative (maybe in a follow up patch? I don't particularly mind..). Not sure if other people agree with this.

[RKSimon]

In D73844#1854361, @andreadb wrote:

Here is the idea: why don't we just add new writes for gather instructions (ideally a new write per each type of gather)?. I think that we shall consider this alternative (maybe in a follow up patch? I don't particularly mind..). Not sure if other people agree with this.

Adding gather (and broadcast) schedule classes is on my backlog - as long as the InstRW entries are correct, I'll replace them when I get aroung to doing it. So, staying with InstRW for this patch is fine by me.

[andreadb]

In D73844#1854616, @RKSimon wrote:

In D73844#1854361, @andreadb wrote:

Here is the idea: why don't we just add new writes for gather instructions (ideally a new write per each type of gather)?. I think that we shall consider this alternative (maybe in a follow up patch? I don't particularly mind..). Not sure if other people agree with this.

Adding gather (and broadcast) schedule classes is on my backlog - as long as the InstRW entries are correct, I'll replace them when I get aroung to doing it. So, staying with InstRW for this patch is fine by me.

Sounds like a good plan to me.

I don't have other comments. Assuming that numbers are correct, the patch looks good to me.

[craig.topper]

Fix mistake in uop count and resource cycles for VGATHERQPSrm and VPGATHERQDrm

llvm/lib/Target/X86/X86SchedBroadwell.td

 def BWWriteResGroup182 : SchedWriteRes<[BWPort0,BWPort1,BWPort23]> {
   let Latency = 29;
   let NumMicroOps = 3;
   let ResourceCycles = [1,1,1];
 }
 def: InstRW<[BWWriteResGroup182], (instregex "DIVR_FI(16|32)m")>;
 
 def BWWriteResGroup183_1 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 22;
+  let Latency = 17;
   let NumMicroOps = 7;
   let ResourceCycles = [1,3,2,1];
 }
-def: InstRW<[BWWriteResGroup183_1], (instrs VGATHERQPDrm)>;
+def: InstRW<[BWWriteResGroup183_1], (instrs VGATHERDPDrm, VPGATHERDQrm,
+                                            VGATHERQPDrm, VPGATHERQQrm)>;
 
 def BWWriteResGroup183_2 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 23;
+  let Latency = 18;
   let NumMicroOps = 9;
   let ResourceCycles = [1,3,4,1];
 }
-def: InstRW<[BWWriteResGroup183_2], (instrs VGATHERQPDYrm)>;
+def: InstRW<[BWWriteResGroup183_2], (instrs VGATHERDPDYrm, VPGATHERDQYrm,
+                                            VGATHERQPDYrm, VPGATHERQQYrm)>;
 
 def BWWriteResGroup183_3 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 24;
+  let Latency = 19;
   let NumMicroOps = 9;
   let ResourceCycles = [1,5,2,1];
 }
-def: InstRW<[BWWriteResGroup183_3], (instrs VGATHERQPSYrm)>;
+def: InstRW<[BWWriteResGroup183_3], (instrs VGATHERQPSrm, VPGATHERQDrm)>;
 
 def BWWriteResGroup183_4 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 25;
-  let NumMicroOps = 7;
-  let ResourceCycles = [1,3,2,1];
+  let Latency = 19;
+  let NumMicroOps = 10;
+  let ResourceCycles = [1,4,4,1];
 }
-def: InstRW<[BWWriteResGroup183_4], (instrs VGATHERDPDrm,
-                                            VGATHERDPSrm)>;
+def: InstRW<[BWWriteResGroup183_4], (instrs VGATHERDPSrm, VPGATHERDDrm,
+                                            VGATHERQPSYrm, VPGATHERQDYrm)>;
 
 def BWWriteResGroup183_5 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 26;
-  let NumMicroOps = 9;
-  let ResourceCycles = [1,5,2,1];
-}
-def: InstRW<[BWWriteResGroup183_5], (instrs VGATHERDPDYrm)>;
-
-def BWWriteResGroup183_6 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 26;
+  let Latency = 21;
   let NumMicroOps = 14;
   let ResourceCycles = [1,4,8,1];
 }
-def: InstRW<[BWWriteResGroup183_6], (instrs VGATHERDPSYrm)>;
-
-def BWWriteResGroup183_7 : SchedWriteRes<[BWPort4, BWPort5, BWPort23, BWPort0156]> {
-  let Latency = 27;
-  let NumMicroOps = 9;
-  let ResourceCycles = [1,5,2,1];
-}
-def: InstRW<[BWWriteResGroup183_7], (instrs VGATHERQPSrm)>;
+def: InstRW<[BWWriteResGroup183_5], (instrs VGATHERDPSYrm, VPGATHERDDYrm)>;
 
 def BWWriteResGroup185 : SchedWriteRes<[BWPort4,BWPort6,BWPort23,BWPort237,BWPort0156]> {
   let Latency = 29;
   let NumMicroOps = 27;
   let ResourceCycles = [1,5,1,1,19];
 }
 def: InstRW<[BWWriteResGroup185], (instrs XSAVE64)>;
 

llvm/lib/Target/X86/X86SchedHaswell.td

 
 def HWWriteResGroup183 : SchedWriteRes<[HWPort0,HWPort1,HWPort4,HWPort5,HWPort6,HWPort237,HWPort06,HWPort0156]> {
   let Latency = 115;
   let NumMicroOps = 100;
   let ResourceCycles = [9,9,11,8,1,11,21,30];
 }
 def: InstRW<[HWWriteResGroup183], (instrs FSTENVm)>;
 
-def HWWriteResGroup184 : SchedWriteRes<[HWPort0, HWPort5, HWPort15, HWPort015, HWPort06, HWPort23]> {
-  let Latency = 26;
+def HWWriteResGroup184 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 14;
   let NumMicroOps = 12;
-  let ResourceCycles = [2,2,1,3,2,2];
+  let ResourceCycles = [2,2,2,1,3,2];
 }
-def: InstRW<[HWWriteResGroup184], (instrs VGATHERDPDrm,
-                                          VPGATHERDQrm,
-                                          VPGATHERDDrm)>;
+def: InstRW<[HWWriteResGroup184], (instrs VGATHERDPDrm, VPGATHERDQrm)>;
 
-def HWWriteResGroup185 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 24;
-  let NumMicroOps = 22;
-  let ResourceCycles = [5,3,4,1,5,4];
+def HWWriteResGroup185 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 17;
+  let NumMicroOps = 20;
+  let ResourceCycles = [3,3,4,1,5,4];
 }
-def: InstRW<[HWWriteResGroup185], (instrs VGATHERQPDYrm,
-                                          VPGATHERQQYrm)>;
+def: InstRW<[HWWriteResGroup185], (instrs VGATHERDPDYrm, VPGATHERDQYrm)>;
 
-def HWWriteResGroup186 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 28;
-  let NumMicroOps = 22;
-  let ResourceCycles = [5,3,4,1,5,4];
-}
-def: InstRW<[HWWriteResGroup186], (instrs VPGATHERQDYrm)>;
-
-def HWWriteResGroup187 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 25;
-  let NumMicroOps = 22;
-  let ResourceCycles = [5,3,4,1,5,4];
-}
-def: InstRW<[HWWriteResGroup187], (instrs VPGATHERQDrm)>;
-
-def HWWriteResGroup188 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 27;
+def HWWriteResGroup186 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 16;
   let NumMicroOps = 20;
   let ResourceCycles = [3,3,4,1,5,4];
 }
-def: InstRW<[HWWriteResGroup188], (instrs VGATHERDPDYrm,
-                                          VPGATHERDQYrm)>;
+def: InstRW<[HWWriteResGroup186], (instrs VGATHERDPSrm, VPGATHERDDrm)>;
 
-def HWWriteResGroup189 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 27;
+def HWWriteResGroup187 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 22;
   let NumMicroOps = 34;
   let ResourceCycles = [5,3,8,1,9,8];
 }
-def: InstRW<[HWWriteResGroup189], (instrs VGATHERDPSYrm,
-                                          VPGATHERDDYrm)>;
+def: InstRW<[HWWriteResGroup187], (instrs VGATHERDPSYrm, VPGATHERDDYrm)>;
 
-def HWWriteResGroup190 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 23;
+def HWWriteResGroup188 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 15;
   let NumMicroOps = 14;
   let ResourceCycles = [3,3,2,1,3,2];
 }
-def: InstRW<[HWWriteResGroup190], (instrs VGATHERQPDrm,
-                                          VPGATHERQQrm)>;
+def: InstRW<[HWWriteResGroup188], (instrs VGATHERQPDrm, VPGATHERQQrm)>;
 
-def HWWriteResGroup191 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 28;
-  let NumMicroOps = 15;
-  let ResourceCycles = [3,3,2,1,4,2];
+def HWWriteResGroup189 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 17;
+  let NumMicroOps = 22;
+  let ResourceCycles = [5,3,4,1,5,4];
 }
-def: InstRW<[HWWriteResGroup191], (instrs VGATHERQPSYrm)>;
+def: InstRW<[HWWriteResGroup189], (instrs VGATHERQPDYrm, VPGATHERQQYrm,
+                                          VGATHERQPSYrm, VPGATHERQDYrm)>;
 
-def HWWriteResGroup192 : SchedWriteRes<[HWPort0, HWPort5, HWPort06, HWPort15, HWPort015, HWPort23]> {
-  let Latency = 25;
+def HWWriteResGroup190 : SchedWriteRes<[HWPort0,HWPort5,HWPort06,HWPort15,HWPort015,HWPort23]> {
+  let Latency = 16;
   let NumMicroOps = 15;
   let ResourceCycles = [3,3,2,1,4,2];
 }
-def: InstRW<[HWWriteResGroup192], (instrs VGATHERQPSrm,
-                                          VGATHERDPSrm)>;
+def: InstRW<[HWWriteResGroup190], (instrs VGATHERQPSrm, VPGATHERQDrm)>;
 
 def: InstRW<[WriteZero], (instrs CLC)>;
 
 
 // Intruction variants handled by the renamer. These might not need execution
 // ports in certain conditions.
 // See Agner's Fog "The microarchitecture of Intel, AMD and VIA CPUs",
 // section "Haswell and Broadwell Pipeline" > "Register allocation and

llvm/test/tools/llvm-mca/X86/Broadwell/resources-avx2.s

 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      6     0.50    *                   vbroadcasti128  (%rax), %ymm0
 # CHECK-NEXT:  1      3     1.00                        vbroadcastsd  %xmm0, %ymm0
 # CHECK-NEXT:  1      3     1.00                        vbroadcastss  %xmm0, %ymm0
 # CHECK-NEXT:  1      3     1.00                        vextracti128  $1, %ymm0, %xmm2
 # CHECK-NEXT:  2      1     1.00           *            vextracti128  $1, %ymm0, (%rax)
-# CHECK-NEXT:  7      25    3.00    *                   vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  9      26    5.00    *                   vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  7      25    3.00    *                   vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  14     26    4.00    *                   vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  7      22    3.00    *                   vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  9      23    3.00    *                   vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  9      27    5.00    *                   vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  9      24    5.00    *                   vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  7      17    3.00    *                   vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  9      18    3.00    *                   vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  10     19    4.00    *                   vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  14     21    4.00    *                   vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  7      17    3.00    *                   vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  9      18    3.00    *                   vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  9      19    5.00    *                   vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  10     19    4.00    *                   vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
 # CHECK-NEXT:  1      3     1.00                        vinserti128 $1, %xmm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      6     0.50    *                   vinserti128 $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      6     0.50    *                   vmovntdqa (%rax), %ymm0
 # CHECK-NEXT:  3      7     2.00                        vmpsadbw  $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      13    2.00    *                   vmpsadbw  $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpabsb  %ymm0, %ymm2
 # CHECK-NEXT:  2      7     0.50    *                   vpabsb  (%rax), %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpabsd  %ymm0, %ymm2

 # CHECK-NEXT:  1      3     1.00                        vpermd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      9     1.00    *                   vpermd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermpd $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      9     1.00    *                   vpermpd $1, (%rax), %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermps %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      9     1.00    *                   vpermps (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermq  $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      9     1.00    *                   vpermq  $1, (%rax), %ymm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  1      5     0.50    *                   vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  10     19    4.00    *                   vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  14     21    4.00    *                   vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  7      17    3.00    *                   vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  9      18    3.00    *                   vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  9      19    5.00    *                   vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  10     19    4.00    *                   vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  7      17    3.00    *                   vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  9      18    3.00    *                   vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      9     2.00    *                   vphaddd (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddsw  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      9     2.00    *                   vphaddsw  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddw %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      9     2.00    *                   vphaddw (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphsubd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      9     2.00    *                   vphsubd (%rax), %ymm1, %ymm2

 # CHECK-NEXT: [5]   - BWPort3
 # CHECK-NEXT: [6]   - BWPort4
 # CHECK-NEXT: [7]   - BWPort5
 # CHECK-NEXT: [8]   - BWPort6
 # CHECK-NEXT: [9]   - BWPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -      -     94.67  58.67  85.67  85.67  13.00  237.67 2.00   1.67
+# CHECK-NEXT:  -      -     96.67  60.67  99.67  99.67  21.00  266.67 4.00   1.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vbroadcasti128  (%rax), %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vbroadcastsd  %xmm0, %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vbroadcastss  %xmm0, %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vextracti128  $1, %ymm0, %xmm2
 # CHECK-NEXT:  -      -      -      -     0.33   0.33   1.00    -      -     0.33   vextracti128  $1, %ymm0, (%rax)
 # CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   3.25   0.25    -     vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   5.25   0.25    -     vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   3.25   0.25    -     vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   3.25   0.25    -     vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   4.25   0.25    -     vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     0.25   0.25   4.00   4.00   1.00   4.25   0.25    -     vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   3.25   0.25    -     vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   3.25   0.25    -     vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   5.25   0.25    -     vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   5.25   0.25    -     vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   4.25   0.25    -     vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vinserti128 $1, %xmm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vinserti128 $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vmovntdqa (%rax), %ymm0
 # CHECK-NEXT:  -      -     1.00    -      -      -      -     2.00    -      -     vmpsadbw  $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     1.00    -     0.50   0.50    -     2.00    -      -     vmpsadbw  $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     0.50    -      -     vpabsb  %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     0.50    -      -     vpabsb  (%rax), %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     0.50    -      -     vpabsd  %ymm0, %ymm2

 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermpd $1, %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermpd $1, (%rax), %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermps %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermps (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermq  $1, %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermq  $1, (%rax), %ymm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   4.25   0.25    -     vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   4.00   4.00   1.00   4.25   0.25    -     vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   3.25   0.25    -     vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   3.25   0.25    -     vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   5.25   0.25    -     vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   4.25   0.25    -     vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   1.00   1.00   1.00   3.25   0.25    -     vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     0.25   0.25   2.00   2.00   1.00   3.25   0.25    -     vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphaddd (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddsw  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphaddsw  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddw %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphaddw (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphsubd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphsubd (%rax), %ymm1, %ymm2

llvm/test/tools/llvm-mca/X86/Haswell/resources-avx2.s

 # CHECK-NEXT: [6]: HasSideEffects (U)
 
 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 # CHECK-NEXT:  1      7     0.50    *                   vbroadcasti128  (%rax), %ymm0
 # CHECK-NEXT:  1      3     1.00                        vbroadcastsd  %xmm0, %ymm0
 # CHECK-NEXT:  1      3     1.00                        vbroadcastss  %xmm0, %ymm0
 # CHECK-NEXT:  1      3     1.00                        vextracti128  $1, %ymm0, %xmm2
 # CHECK-NEXT:  2      1     1.00           *            vextracti128  $1, %ymm0, (%rax)
-# CHECK-NEXT:  12     26    2.67    *                   vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  20     27    4.00    *                   vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  15     25    3.67    *                   vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  34     27    6.50    *                   vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  14     23    3.33    *                   vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  22     24    5.00    *                   vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  15     25    3.67    *                   vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  15     28    3.67    *                   vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  12     14    2.67    *                   vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  20     17    4.00    *                   vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  20     16    4.00    *                   vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  34     22    6.50    *                   vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  14     15    3.33    *                   vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  22     17    5.00    *                   vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  15     16    3.67    *                   vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  22     17    5.00    *                   vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
 # CHECK-NEXT:  1      3     1.00                        vinserti128 $1, %xmm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      7     0.50    *                   vinserti128 $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      7     0.50    *                   vmovntdqa (%rax), %ymm0
 # CHECK-NEXT:  3      7     2.00                        vmpsadbw  $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      14    2.00    *                   vmpsadbw  $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpabsb  %ymm0, %ymm2
 # CHECK-NEXT:  2      8     0.50    *                   vpabsb  (%rax), %ymm2
 # CHECK-NEXT:  1      1     0.50                        vpabsd  %ymm0, %ymm2

 # CHECK-NEXT:  1      3     1.00                        vpermd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      10    1.00    *                   vpermd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermpd $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      10    1.00    *                   vpermpd $1, (%rax), %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermps %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  2      10    1.00    *                   vpermps (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  1      3     1.00                        vpermq  $1, %ymm0, %ymm2
 # CHECK-NEXT:  2      10    1.00    *                   vpermq  $1, (%rax), %ymm2
-# CHECK-NEXT:  12     26    2.67    *                   vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  34     27    6.50    *                   vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
-# CHECK-NEXT:  12     26    2.67    *                   vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  20     27    4.00    *                   vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  22     25    5.00    *                   vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  22     28    5.00    *                   vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
-# CHECK-NEXT:  14     23    3.33    *                   vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  22     24    5.00    *                   vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  20     16    4.00    *                   vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  34     22    6.50    *                   vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
+# CHECK-NEXT:  12     14    2.67    *                   vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  20     17    4.00    *                   vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
+# CHECK-NEXT:  15     16    3.67    *                   vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  22     17    5.00    *                   vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  14     15    3.33    *                   vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  22     17    5.00    *                   vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      10    2.00    *                   vphaddd (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddsw  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      10    2.00    *                   vphaddsw  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphaddw %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      10    2.00    *                   vphaddw (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  3      3     2.00                        vphsubd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  4      10    2.00    *                   vphsubd (%rax), %ymm1, %ymm2

 # CHECK-NEXT: [5]   - HWPort3
 # CHECK-NEXT: [6]   - HWPort4
 # CHECK-NEXT: [7]   - HWPort5
 # CHECK-NEXT: [8]   - HWPort6
 # CHECK-NEXT: [9]   - HWPort7
 
 # CHECK:      Resource pressure per iteration:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
-# CHECK-NEXT:  -      -     202.67 89.67  97.67  97.67  5.00   282.67 28.00  1.67
+# CHECK-NEXT:  -      -     206.67 90.67  99.67  99.67  5.00   284.67 30.00  1.67
 
 # CHECK:      Resource pressure by instruction:
 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vbroadcasti128  (%rax), %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vbroadcastsd  %xmm0, %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vbroadcastss  %xmm0, %ymm0
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vextracti128  $1, %ymm0, %xmm2
 # CHECK-NEXT:  -      -      -      -     0.33   0.33   1.00    -      -     0.33   vextracti128  $1, %ymm0, (%rax)
 # CHECK-NEXT:  -      -     4.00   1.50   1.00   1.00    -     3.50   1.00    -     vgatherdpd  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     6.67   2.17   2.00   2.00    -     5.17   2.00    -     vgatherdpd  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  -      -     5.33   1.83   1.00   1.00    -     4.83   1.00    -     vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     6.67   2.17   2.00   2.00    -     5.17   2.00    -     vgatherdps  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     12.00  3.50   4.00   4.00    -     6.50   4.00    -     vgatherdps  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -     5.00   1.50   1.00   1.00    -     4.50   1.00    -     vgatherqpd  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     8.67   2.17   2.00   2.00    -     5.17   2.00    -     vgatherqpd  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -     5.33   1.83   1.00   1.00    -     4.83   1.00    -     vgatherqps  %xmm0, (%rax,%xmm1,2), %xmm2
-# CHECK-NEXT:  -      -     5.33   1.83   1.00   1.00    -     4.83   1.00    -     vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
+# CHECK-NEXT:  -      -     8.67   2.17   2.00   2.00    -     5.17   2.00    -     vgatherqps  %xmm0, (%rax,%ymm1,2), %xmm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vinserti128 $1, %xmm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vinserti128 $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -      -      -      -     vmovntdqa (%rax), %ymm0
 # CHECK-NEXT:  -      -     1.00    -      -      -      -     2.00    -      -     vmpsadbw  $1, %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -     1.00    -     0.50   0.50    -     2.00    -      -     vmpsadbw  $1, (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     0.50    -      -     vpabsb  %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     0.50    -      -     vpabsb  (%rax), %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     0.50    -      -     vpabsd  %ymm0, %ymm2

 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermd  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermd  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermpd $1, %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermpd $1, (%rax), %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermps %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermps (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermq  $1, %ymm0, %ymm2
 # CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vpermq  $1, (%rax), %ymm2
-# CHECK-NEXT:  -      -     4.00   1.50   1.00   1.00    -     3.50   1.00    -     vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     6.67   2.17   2.00   2.00    -     5.17   2.00    -     vpgatherdd  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     12.00  3.50   4.00   4.00    -     6.50   4.00    -     vpgatherdd  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -     4.00   1.50   1.00   1.00    -     3.50   1.00    -     vpgatherdq  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     6.67   2.17   2.00   2.00    -     5.17   2.00    -     vpgatherdq  %ymm0, (%rax,%xmm1,2), %ymm2
-# CHECK-NEXT:  -      -     8.67   2.17   2.00   2.00    -     5.17   2.00    -     vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
+# CHECK-NEXT:  -      -     5.33   1.83   1.00   1.00    -     4.83   1.00    -     vpgatherqd  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     8.67   2.17   2.00   2.00    -     5.17   2.00    -     vpgatherqd  %xmm0, (%rax,%ymm1,2), %xmm2
 # CHECK-NEXT:  -      -     5.00   1.50   1.00   1.00    -     4.50   1.00    -     vpgatherqq  %xmm0, (%rax,%xmm1,2), %xmm2
 # CHECK-NEXT:  -      -     8.67   2.17   2.00   2.00    -     5.17   2.00    -     vpgatherqq  %ymm0, (%rax,%ymm1,2), %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddd %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphaddd (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddsw  %ymm0, %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50   0.50   0.50    -     2.50    -      -     vphaddsw  (%rax), %ymm1, %ymm2
 # CHECK-NEXT:  -      -      -     0.50    -      -      -     2.50    -      -     vphaddw %ymm0, %ymm1, %ymm2