Commit 02f03a6f authored by Craig Topper's avatar Craig Topper
Browse files

[X86] Match vpmullq latency to uops.info. Correct port usage for 512-bit memory form 

uops.info says these should be 15 cycle instructions. Uops.info also shows the 512-bit form uses port 0 and 5 for both register and memory. We had memory using 0 and 1.

Differential Revision: https://reviews.llvm.org/D75549
parent 1bedb234

llvm/lib/Target/X86/X86SchedSkylakeServer.td

+14 −8
Original line number Diff line number Diff line
@@ -1963,14 +1963,14 @@ def SKXWriteResGroup171 : SchedWriteRes<[SKXPort06,SKXPort0156]> { 
def: InstRW<[SKXWriteResGroup171], (instrs LOOPE, LOOPNE)>;



def SKXWriteResGroup174 : SchedWriteRes<[SKXPort01]> {

  let Latency = 12;

  let Latency = 15;

  let NumMicroOps = 3;

  let ResourceCycles = [3];

}

def: InstRW<[SKXWriteResGroup174], (instregex "VPMULLQ(Z128|Z256)rr")>;



def SKXWriteResGroup174z : SchedWriteRes<[SKXPort05]> {

  let Latency = 12;

  let Latency = 15;

  let NumMicroOps = 3;

  let ResourceCycles = [3];

}
@@ -2131,8 +2131,8 @@ def SKXWriteResGroup202 : SchedWriteRes<[SKXPort0,SKXPort1,SKXPort5,SKXPort6,SKX 
}

def: InstRW<[SKXWriteResGroup202], (instrs XCH_F)>;



def SKXWriteResGroup205 : SchedWriteRes<[SKXPort23,SKXPort015]> {

  let Latency = 18;

def SKXWriteResGroup205 : SchedWriteRes<[SKXPort23,SKXPort01]> {

  let Latency = 21;

  let NumMicroOps = 4;

  let ResourceCycles = [1,3];

}
@@ -2159,13 +2159,19 @@ def SKXWriteResGroup209 : SchedWriteRes<[SKXPort0,SKXPort23,SKXFPDivider]> { 
}

def : SchedAlias<WriteFDiv64Ld,  SKXWriteResGroup209>; // TODO - convert to ZnWriteResFpuPair



def SKXWriteResGroup211 : SchedWriteRes<[SKXPort23,SKXPort015]> {

  let Latency = 19;

def SKXWriteResGroup211 : SchedWriteRes<[SKXPort23,SKXPort01]> {

  let Latency = 22;

  let NumMicroOps = 4;

  let ResourceCycles = [1,3];

}

def: InstRW<[SKXWriteResGroup211], (instregex "VPMULLQZ256rm(b?)")>;



def SKXWriteResGroup211_1 : SchedWriteRes<[SKXPort23,SKXPort05]> {

  let Latency = 22;

  let NumMicroOps = 4;

  let ResourceCycles = [1,3];

}

def: InstRW<[SKXWriteResGroup211], (instregex "VPMULLQZ256rm(b?)",

                                              "VPMULLQZrm(b?)")>;

def: InstRW<[SKXWriteResGroup211_1], (instregex "VPMULLQZrm(b?)")>;



def SKXWriteResGroup215 : SchedWriteRes<[SKXPort0]> {

  let Latency = 20;

llvm/test/tools/llvm-mca/X86/SkylakeServer/resources-avx512dq.s

+10 −10
Original line number Diff line number Diff line
@@ -252,12 +252,12 @@ vxorps (%rax){1to16}, %zmm17, %zmm19 {z}{k1} 
# CHECK-NEXT:  2      9     1.00    *                   vfpclassss	$171, (%rax), %k1

# CHECK-NEXT:  1      4     1.00                        vfpclassss	$171, %xmm16, %k1 {%k2}

# CHECK-NEXT:  2      9     1.00    *                   vfpclassss	$171, (%rax), %k1 {%k2}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%zmm16, %zmm17, %zmm19

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  3      12    1.50                        vpmullq	%zmm16, %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%zmm16, %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%zmm16, %zmm17, %zmm19

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  3      15    1.50                        vpmullq	%zmm16, %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%zmm16, %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  1      1     0.50                        vxorpd	%zmm16, %zmm17, %zmm19

# CHECK-NEXT:  2      8     0.50    *                   vxorpd	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  2      8     0.50    *                   vxorpd	(%rax){1to8}, %zmm17, %zmm19
@@ -291,7 +291,7 @@ vxorps (%rax){1to16}, %zmm17, %zmm19 {z}{k1} 


# CHECK:      Resource pressure per iteration:

# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]

# CHECK-NEXT:  -      -     55.50  7.00   37.50  37.50   -     81.50   -      -

# CHECK-NEXT:  -      -     57.00  4.00   37.50  37.50   -     83.00   -      -



# CHECK:      Resource pressure by instruction:

# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
@@ -397,11 +397,11 @@ vxorps (%rax){1to16}, %zmm17, %zmm19 {z}{k1} 
# CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vfpclassss	$171, %xmm16, %k1 {%k2}

# CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vfpclassss	$171, (%rax), %k1 {%k2}

# CHECK-NEXT:  -      -     1.50    -      -      -      -     1.50    -      -     vpmullq	%zmm16, %zmm17, %zmm19

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  -      -     1.50    -     0.50   0.50    -     1.50    -      -     vpmullq	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  -      -     1.50    -      -      -      -     1.50    -      -     vpmullq	%zmm16, %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  -      -     1.50    -     0.50   0.50    -     1.50    -      -     vpmullq	(%rax), %zmm17, %zmm19 {%k1}

# CHECK-NEXT:  -      -     1.50    -      -      -      -     1.50    -      -     vpmullq	%zmm16, %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.50    -     0.50   0.50    -     1.50    -      -     vpmullq	(%rax), %zmm17, %zmm19 {%k1} {z}

# CHECK-NEXT:  -      -     0.50    -      -      -      -     0.50    -      -     vxorpd	%zmm16, %zmm17, %zmm19

# CHECK-NEXT:  -      -     0.50    -     0.50   0.50    -     0.50    -      -     vxorpd	(%rax), %zmm17, %zmm19

# CHECK-NEXT:  -      -     0.50    -     0.50   0.50    -     0.50    -      -     vxorpd	(%rax){1to8}, %zmm17, %zmm19

llvm/test/tools/llvm-mca/X86/SkylakeServer/resources-avx512dqvl.s

+19 −19
Original line number Diff line number Diff line
@@ -344,18 +344,18 @@ vxorps (%rax){1to8}, %ymm17, %ymm19 {z}{k1} 
# CHECK-NEXT:  1      4     1.00                        vfpclassps	$171, %ymm16, %k1 {%k2}

# CHECK-NEXT:  2      11    1.00    *                   vfpclasspsy	$171, (%rax), %k1 {%k2}

# CHECK-NEXT:  2      11    1.00    *                   vfpclassps	$171, (%rax){1to8}, %k1 {%k2}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%xmm16, %xmm17, %xmm19

# CHECK-NEXT:  4      18    1.00    *                   vpmullq	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  3      12    1.50                        vpmullq	%xmm16, %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  4      18    1.00    *                   vpmullq	(%rax), %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%xmm16, %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  4      18    1.00    *                   vpmullq	(%rax), %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%ymm16, %ymm17, %ymm19

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %ymm17, %ymm19

# CHECK-NEXT:  3      12    1.50                        vpmullq	%ymm16, %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  3      12    1.50                        vpmullq	%ymm16, %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  4      19    1.00    *                   vpmullq	(%rax), %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%xmm16, %xmm17, %xmm19

# CHECK-NEXT:  4      21    1.50    *                   vpmullq	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  3      15    1.50                        vpmullq	%xmm16, %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  4      21    1.50    *                   vpmullq	(%rax), %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%xmm16, %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  4      21    1.50    *                   vpmullq	(%rax), %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%ymm16, %ymm17, %ymm19

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %ymm17, %ymm19

# CHECK-NEXT:  3      15    1.50                        vpmullq	%ymm16, %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  3      15    1.50                        vpmullq	%ymm16, %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  4      22    1.50    *                   vpmullq	(%rax), %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  1      1     0.33                        vxorpd	%xmm16, %xmm17, %xmm19

# CHECK-NEXT:  2      7     0.50    *                   vxorpd	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  2      7     0.50    *                   vxorpd	(%rax){1to2}, %xmm17, %xmm19
@@ -407,7 +407,7 @@ vxorps (%rax){1to8}, %ymm17, %ymm19 {z}{k1} 


# CHECK:      Resource pressure per iteration:

# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]

# CHECK-NEXT:  -      -     64.00  64.00  59.00  59.00   -     76.00   -      -

# CHECK-NEXT:  -      -     67.00  67.00  59.00  59.00   -     70.00   -      -



# CHECK:      Resource pressure by instruction:

# CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
@@ -544,17 +544,17 @@ vxorps (%rax){1to8}, %ymm17, %ymm19 {z}{k1} 
# CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vfpclasspsy	$171, (%rax), %k1 {%k2}

# CHECK-NEXT:  -      -      -      -     0.50   0.50    -     1.00    -      -     vfpclassps	$171, (%rax){1to8}, %k1 {%k2}

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%xmm16, %xmm17, %xmm19

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%xmm16, %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %xmm17, %xmm19 {%k1}

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%xmm16, %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %xmm17, %xmm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%ymm16, %ymm17, %ymm19

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %ymm17, %ymm19

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %ymm17, %ymm19

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%ymm16, %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %ymm17, %ymm19 {%k1}

# CHECK-NEXT:  -      -     1.50   1.50    -      -      -      -      -      -     vpmullq	%ymm16, %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.00   1.00   0.50   0.50    -     1.00    -      -     vpmullq	(%rax), %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  -      -     1.50   1.50   0.50   0.50    -      -      -      -     vpmullq	(%rax), %ymm17, %ymm19 {%k1} {z}

# CHECK-NEXT:  -      -     0.33   0.33    -      -      -     0.33    -      -     vxorpd	%xmm16, %xmm17, %xmm19

# CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd	(%rax), %xmm17, %xmm19

# CHECK-NEXT:  -      -     0.33   0.33   0.50   0.50    -     0.33    -      -     vxorpd	(%rax){1to2}, %xmm17, %xmm19