85 lines
5.3 KiB
ArmAsm
85 lines
5.3 KiB
ArmAsm
# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
|
|
# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=sapphirerapids -instruction-tables < %s | FileCheck %s
|
|
|
|
vpopcntb %zmm1, %zmm0
|
|
vpopcntb (%rdi), %zmm0
|
|
vpopcntb %zmm1, %zmm0 {%k1}
|
|
vpopcntb (%rdi), %zmm0 {%k1}
|
|
vpopcntb %zmm1, %zmm0 {%k1} {z}
|
|
vpopcntb (%rdi), %zmm0 {%k1} {z}
|
|
|
|
vpopcntw %zmm1, %zmm0
|
|
vpopcntw (%rdi), %zmm0
|
|
vpopcntw %zmm1, %zmm0 {%k1}
|
|
vpopcntw (%rdi), %zmm0 {%k1}
|
|
vpopcntw %zmm1, %zmm0 {%k1} {z}
|
|
vpopcntw (%rdi), %zmm0 {%k1} {z}
|
|
|
|
vpshufbitqmb %zmm16, %zmm17, %k2
|
|
vpshufbitqmb (%rdi), %zmm17, %k2
|
|
vpshufbitqmb %zmm16, %zmm17, %k2 {%k1}
|
|
vpshufbitqmb (%rdi), %zmm17, %k2 {%k1}
|
|
|
|
# CHECK: Instruction Info:
|
|
# CHECK-NEXT: [1]: #uOps
|
|
# CHECK-NEXT: [2]: Latency
|
|
# CHECK-NEXT: [3]: RThroughput
|
|
# CHECK-NEXT: [4]: MayLoad
|
|
# CHECK-NEXT: [5]: MayStore
|
|
# CHECK-NEXT: [6]: HasSideEffects (U)
|
|
|
|
# CHECK: [1] [2] [3] [4] [5] [6] Instructions:
|
|
# CHECK-NEXT: 1 3 1.00 vpopcntb %zmm1, %zmm0
|
|
# CHECK-NEXT: 2 11 1.00 * vpopcntb (%rdi), %zmm0
|
|
# CHECK-NEXT: 1 5 1.00 vpopcntb %zmm1, %zmm0 {%k1}
|
|
# CHECK-NEXT: 2 13 1.00 * vpopcntb (%rdi), %zmm0 {%k1}
|
|
# CHECK-NEXT: 1 5 1.00 vpopcntb %zmm1, %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: 2 13 1.00 * vpopcntb (%rdi), %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: 1 3 1.00 vpopcntw %zmm1, %zmm0
|
|
# CHECK-NEXT: 2 11 1.00 * vpopcntw (%rdi), %zmm0
|
|
# CHECK-NEXT: 1 5 1.00 vpopcntw %zmm1, %zmm0 {%k1}
|
|
# CHECK-NEXT: 2 13 1.00 * vpopcntw (%rdi), %zmm0 {%k1}
|
|
# CHECK-NEXT: 1 5 1.00 vpopcntw %zmm1, %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: 2 13 1.00 * vpopcntw (%rdi), %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: 2 6 1.00 vpshufbitqmb %zmm16, %zmm17, %k2
|
|
# CHECK-NEXT: 3 6 1.00 * vpshufbitqmb (%rdi), %zmm17, %k2
|
|
# CHECK-NEXT: 2 8 1.00 vpshufbitqmb %zmm16, %zmm17, %k2 {%k1}
|
|
# CHECK-NEXT: 3 8 1.00 * vpshufbitqmb (%rdi), %zmm17, %k2 {%k1}
|
|
|
|
# CHECK: Resources:
|
|
# CHECK-NEXT: [0] - SPRPort00
|
|
# CHECK-NEXT: [1] - SPRPort01
|
|
# CHECK-NEXT: [2] - SPRPort02
|
|
# CHECK-NEXT: [3] - SPRPort03
|
|
# CHECK-NEXT: [4] - SPRPort04
|
|
# CHECK-NEXT: [5] - SPRPort05
|
|
# CHECK-NEXT: [6] - SPRPort06
|
|
# CHECK-NEXT: [7] - SPRPort07
|
|
# CHECK-NEXT: [8] - SPRPort08
|
|
# CHECK-NEXT: [9] - SPRPort09
|
|
# CHECK-NEXT: [10] - SPRPort10
|
|
# CHECK-NEXT: [11] - SPRPort11
|
|
# CHECK-NEXT: [12] - SPRPortInvalid
|
|
|
|
# CHECK: Resource pressure per iteration:
|
|
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12]
|
|
# CHECK-NEXT: 4.00 - 2.67 2.67 - 16.00 - - - - - 2.67 -
|
|
|
|
# CHECK: Resource pressure by instruction:
|
|
# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] Instructions:
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntb %zmm1, %zmm0
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntb (%rdi), %zmm0
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntb %zmm1, %zmm0 {%k1}
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntb (%rdi), %zmm0 {%k1}
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntb %zmm1, %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntb (%rdi), %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntw %zmm1, %zmm0
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntw (%rdi), %zmm0
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntw %zmm1, %zmm0 {%k1}
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntw (%rdi), %zmm0 {%k1}
|
|
# CHECK-NEXT: - - - - - 1.00 - - - - - - - vpopcntw %zmm1, %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: - - 0.33 0.33 - 1.00 - - - - - 0.33 - vpopcntw (%rdi), %zmm0 {%k1} {z}
|
|
# CHECK-NEXT: 1.00 - - - - 1.00 - - - - - - - vpshufbitqmb %zmm16, %zmm17, %k2
|
|
# CHECK-NEXT: 1.00 - 0.33 0.33 - 1.00 - - - - - 0.33 - vpshufbitqmb (%rdi), %zmm17, %k2
|
|
# CHECK-NEXT: 1.00 - - - - 1.00 - - - - - - - vpshufbitqmb %zmm16, %zmm17, %k2 {%k1}
|
|
# CHECK-NEXT: 1.00 - 0.33 0.33 - 1.00 - - - - - 0.33 - vpshufbitqmb (%rdi), %zmm17, %k2 {%k1}
|