From: Simon Pilgrim Date: Sat, 19 Nov 2022 11:35:31 +0000 (+0000) Subject: [MCA][X86] Add test coverage for BITALG instructions X-Git-Tag: upstream/17.0.6~27114 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=aae08b1d372a45b8bef95e86e5fe9110045eb78d;p=platform%2Fupstream%2Fllvm.git [MCA][X86] Add test coverage for BITALG instructions --- diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalg.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalg.s new file mode 100644 index 0000000..3b8b7c1 --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalg.s @@ -0,0 +1,80 @@ +# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py +# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -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 1 0.50 vpopcntb %zmm1, %zmm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %zmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntb %zmm1, %zmm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %zmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntb %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntw %zmm1, %zmm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %zmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntw %zmm1, %zmm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %zmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntw %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %zmm16, %zmm17, %k2 +# CHECK-NEXT: 2 12 1.00 * vpshufbitqmb (%rdi), %zmm17, %k2 +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %zmm16, %zmm17, %k2 {%k1} +# CHECK-NEXT: 2 12 1.00 * vpshufbitqmb (%rdi), %zmm17, %k2 {%k1} + +# CHECK: Resources: +# CHECK-NEXT: [0] - SBDivider +# CHECK-NEXT: [1] - SBFPDivider +# CHECK-NEXT: [2] - SBPort0 +# CHECK-NEXT: [3] - SBPort1 +# CHECK-NEXT: [4] - SBPort4 +# CHECK-NEXT: [5] - SBPort5 +# CHECK-NEXT: [6.0] - SBPort23 +# CHECK-NEXT: [6.1] - SBPort23 + +# CHECK: Resource pressure per iteration: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] +# CHECK-NEXT: - - 4.00 6.00 - 6.00 4.00 4.00 + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %zmm1, %zmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %zmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %zmm1, %zmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %zmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %zmm1, %zmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %zmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %zmm1, %zmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %zmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %zmm16, %zmm17, %k2 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %zmm17, %k2 +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %zmm16, %zmm17, %k2 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %zmm17, %k2 {%k1} diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalgvl.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalgvl.s new file mode 100644 index 0000000..ff20a55 --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bitalgvl.s @@ -0,0 +1,131 @@ +# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py +# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -instruction-tables < %s | FileCheck %s + +vpopcntb %xmm1, %xmm0 +vpopcntb (%rdi), %xmm0 +vpopcntb %xmm1, %xmm0 {%k1} +vpopcntb (%rdi), %xmm0 {%k1} +vpopcntb %xmm1, %xmm0 {%k1} {z} +vpopcntb (%rdi), %xmm0 {%k1} {z} + +vpopcntb %ymm1, %ymm0 +vpopcntb (%rdi), %ymm0 +vpopcntb %ymm1, %ymm0 {%k1} +vpopcntb (%rdi), %ymm0 {%k1} +vpopcntb %ymm1, %ymm0 {%k1} {z} +vpopcntb (%rdi), %ymm0 {%k1} {z} + +vpopcntw %xmm1, %xmm0 +vpopcntw (%rdi), %xmm0 +vpopcntw %xmm1, %xmm0 {%k1} +vpopcntw (%rdi), %xmm0 {%k1} +vpopcntw %xmm1, %xmm0 {%k1} {z} +vpopcntw (%rdi), %xmm0 {%k1} {z} + +vpopcntw %ymm1, %ymm0 +vpopcntw (%rdi), %ymm0 +vpopcntw %ymm1, %ymm0 {%k1} +vpopcntw (%rdi), %ymm0 {%k1} +vpopcntw %ymm1, %ymm0 {%k1} {z} +vpopcntw (%rdi), %ymm0 {%k1} {z} + +vpshufbitqmb %xmm16, %xmm17, %k2 +vpshufbitqmb (%rdi), %xmm17, %k2 +vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} + +vpshufbitqmb %ymm16, %ymm17, %k2 +vpshufbitqmb (%rdi), %ymm17, %k2 +vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +vpshufbitqmb (%rdi), %ymm17, %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 1 0.50 vpopcntb %xmm1, %xmm0 +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntb %xmm1, %xmm0 {%k1} +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntb %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 {%k1} +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %xmm16, %xmm17, %k2 +# CHECK-NEXT: 2 11 1.00 * vpshufbitqmb (%rdi), %xmm17, %k2 +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +# CHECK-NEXT: 2 11 1.00 * vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %ymm16, %ymm17, %k2 +# CHECK-NEXT: 2 12 1.00 * vpshufbitqmb (%rdi), %ymm17, %k2 +# CHECK-NEXT: 1 5 1.00 vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +# CHECK-NEXT: 2 12 1.00 * vpshufbitqmb (%rdi), %ymm17, %k2 {%k1} + +# CHECK: Resources: +# CHECK-NEXT: [0] - SBDivider +# CHECK-NEXT: [1] - SBFPDivider +# CHECK-NEXT: [2] - SBPort0 +# CHECK-NEXT: [3] - SBPort1 +# CHECK-NEXT: [4] - SBPort4 +# CHECK-NEXT: [5] - SBPort5 +# CHECK-NEXT: [6.0] - SBPort23 +# CHECK-NEXT: [6.1] - SBPort23 + +# CHECK: Resource pressure per iteration: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] +# CHECK-NEXT: - - 8.00 12.00 - 12.00 8.00 8.00 + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %xmm1, %xmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %xmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %xmm1, %xmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %xmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %ymm1, %ymm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %ymm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %ymm1, %ymm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %ymm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntb %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntb (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %xmm1, %xmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %xmm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %xmm1, %xmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %xmm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %ymm1, %ymm0 +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %ymm0 +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %ymm1, %ymm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %ymm0 {%k1} +# CHECK-NEXT: - - - 0.50 - 0.50 - - vpopcntw %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: - - - 0.50 - 0.50 0.50 0.50 vpopcntw (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %xmm16, %xmm17, %k2 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %xmm17, %k2 +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %ymm16, %ymm17, %k2 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %ymm17, %k2 +# CHECK-NEXT: - - 1.00 - - - - - vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vpshufbitqmb (%rdi), %ymm17, %k2 {%k1} diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalg.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalg.s new file mode 100644 index 0000000..b331835 --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalg.s @@ -0,0 +1,84 @@ +# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py +# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -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 1 1.00 vpopcntb %zmm1, %zmm0 +# CHECK-NEXT: 2 8 1.00 * vpopcntb (%rdi), %zmm0 +# CHECK-NEXT: 1 1 1.00 vpopcntb %zmm1, %zmm0 {%k1} +# CHECK-NEXT: 2 8 1.00 * vpopcntb (%rdi), %zmm0 {%k1} +# CHECK-NEXT: 1 1 1.00 vpopcntb %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: 2 8 1.00 * vpopcntb (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: 1 1 1.00 vpopcntw %zmm1, %zmm0 +# CHECK-NEXT: 2 8 1.00 * vpopcntw (%rdi), %zmm0 +# CHECK-NEXT: 1 1 1.00 vpopcntw %zmm1, %zmm0 {%k1} +# CHECK-NEXT: 2 8 1.00 * vpopcntw (%rdi), %zmm0 {%k1} +# CHECK-NEXT: 1 1 1.00 vpopcntw %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: 2 8 1.00 * vpopcntw (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %zmm16, %zmm17, %k2 +# CHECK-NEXT: 2 12 0.50 * vpshufbitqmb (%rdi), %zmm17, %k2 +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %zmm16, %zmm17, %k2 {%k1} +# CHECK-NEXT: 2 12 0.50 * vpshufbitqmb (%rdi), %zmm17, %k2 {%k1} + +# CHECK: Resources: +# CHECK-NEXT: [0] - ICXDivider +# CHECK-NEXT: [1] - ICXFPDivider +# CHECK-NEXT: [2] - ICXPort0 +# CHECK-NEXT: [3] - ICXPort1 +# CHECK-NEXT: [4] - ICXPort2 +# CHECK-NEXT: [5] - ICXPort3 +# CHECK-NEXT: [6] - ICXPort4 +# CHECK-NEXT: [7] - ICXPort5 +# CHECK-NEXT: [8] - ICXPort6 +# CHECK-NEXT: [9] - ICXPort7 +# CHECK-NEXT: [10] - ICXPort8 +# CHECK-NEXT: [11] - ICXPort9 + +# CHECK: Resource pressure per iteration: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] +# CHECK-NEXT: - - 14.00 - 4.00 4.00 - 2.00 - - - - + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntb %zmm1, %zmm0 +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntb (%rdi), %zmm0 +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntb %zmm1, %zmm0 {%k1} +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntb (%rdi), %zmm0 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntb %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntb (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntw %zmm1, %zmm0 +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntw (%rdi), %zmm0 +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntw %zmm1, %zmm0 {%k1} +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntw (%rdi), %zmm0 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - - - - - vpopcntw %zmm1, %zmm0 {%k1} {z} +# CHECK-NEXT: - - 1.00 - 0.50 0.50 - - - - - - vpopcntw (%rdi), %zmm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 - - - - 0.50 - - - - vpshufbitqmb %zmm16, %zmm17, %k2 +# CHECK-NEXT: - - 0.50 - 0.50 0.50 - 0.50 - - - - vpshufbitqmb (%rdi), %zmm17, %k2 +# CHECK-NEXT: - - 0.50 - - - - 0.50 - - - - vpshufbitqmb %zmm16, %zmm17, %k2 {%k1} +# CHECK-NEXT: - - 0.50 - 0.50 0.50 - 0.50 - - - - vpshufbitqmb (%rdi), %zmm17, %k2 {%k1} diff --git a/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalgvl.s b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalgvl.s new file mode 100644 index 0000000..4b56fea --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/IceLakeServer/resources-avx512bitalgvl.s @@ -0,0 +1,135 @@ +# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py +# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -instruction-tables < %s | FileCheck %s + +vpopcntb %xmm1, %xmm0 +vpopcntb (%rdi), %xmm0 +vpopcntb %xmm1, %xmm0 {%k1} +vpopcntb (%rdi), %xmm0 {%k1} +vpopcntb %xmm1, %xmm0 {%k1} {z} +vpopcntb (%rdi), %xmm0 {%k1} {z} + +vpopcntb %ymm1, %ymm0 +vpopcntb (%rdi), %ymm0 +vpopcntb %ymm1, %ymm0 {%k1} +vpopcntb (%rdi), %ymm0 {%k1} +vpopcntb %ymm1, %ymm0 {%k1} {z} +vpopcntb (%rdi), %ymm0 {%k1} {z} + +vpopcntw %xmm1, %xmm0 +vpopcntw (%rdi), %xmm0 +vpopcntw %xmm1, %xmm0 {%k1} +vpopcntw (%rdi), %xmm0 {%k1} +vpopcntw %xmm1, %xmm0 {%k1} {z} +vpopcntw (%rdi), %xmm0 {%k1} {z} + +vpopcntw %ymm1, %ymm0 +vpopcntw (%rdi), %ymm0 +vpopcntw %ymm1, %ymm0 {%k1} +vpopcntw (%rdi), %ymm0 {%k1} +vpopcntw %ymm1, %ymm0 {%k1} {z} +vpopcntw (%rdi), %ymm0 {%k1} {z} + +vpshufbitqmb %xmm16, %xmm17, %k2 +vpshufbitqmb (%rdi), %xmm17, %k2 +vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} + +vpshufbitqmb %ymm16, %ymm17, %k2 +vpshufbitqmb (%rdi), %ymm17, %k2 +vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +vpshufbitqmb (%rdi), %ymm17, %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 1 0.50 vpopcntb %xmm1, %xmm0 +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntb %xmm1, %xmm0 {%k1} +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntb %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: 2 7 0.50 * vpopcntb (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntb %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntb (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 {%k1} +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntw %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: 2 7 0.50 * vpopcntw (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 {%k1} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 {%k1} +# CHECK-NEXT: 1 1 0.50 vpopcntw %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: 2 8 0.50 * vpopcntw (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %xmm16, %xmm17, %k2 +# CHECK-NEXT: 2 11 0.50 * vpshufbitqmb (%rdi), %xmm17, %k2 +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +# CHECK-NEXT: 2 11 0.50 * vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %ymm16, %ymm17, %k2 +# CHECK-NEXT: 2 12 0.50 * vpshufbitqmb (%rdi), %ymm17, %k2 +# CHECK-NEXT: 1 5 0.50 vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +# CHECK-NEXT: 2 12 0.50 * vpshufbitqmb (%rdi), %ymm17, %k2 {%k1} + +# CHECK: Resources: +# CHECK-NEXT: [0] - ICXDivider +# CHECK-NEXT: [1] - ICXFPDivider +# CHECK-NEXT: [2] - ICXPort0 +# CHECK-NEXT: [3] - ICXPort1 +# CHECK-NEXT: [4] - ICXPort2 +# CHECK-NEXT: [5] - ICXPort3 +# CHECK-NEXT: [6] - ICXPort4 +# CHECK-NEXT: [7] - ICXPort5 +# CHECK-NEXT: [8] - ICXPort6 +# CHECK-NEXT: [9] - ICXPort7 +# CHECK-NEXT: [10] - ICXPort8 +# CHECK-NEXT: [11] - ICXPort9 + +# CHECK: Resource pressure per iteration: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] +# CHECK-NEXT: - - 16.00 16.00 8.00 8.00 - - - - - - + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] Instructions: +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %xmm1, %xmm0 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %xmm0 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %xmm1, %xmm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %xmm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %ymm1, %ymm0 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %ymm0 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %ymm1, %ymm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %ymm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntb %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntb (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %xmm1, %xmm0 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %xmm0 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %xmm1, %xmm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %xmm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %xmm1, %xmm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %xmm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %ymm1, %ymm0 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %ymm0 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %ymm1, %ymm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %ymm0 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpopcntw %ymm1, %ymm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpopcntw (%rdi), %ymm0 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpshufbitqmb %xmm16, %xmm17, %k2 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpshufbitqmb (%rdi), %xmm17, %k2 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpshufbitqmb %xmm16, %xmm17, %k2 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpshufbitqmb (%rdi), %xmm17, %k2 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpshufbitqmb %ymm16, %ymm17, %k2 +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpshufbitqmb (%rdi), %ymm17, %k2 +# CHECK-NEXT: - - 0.50 0.50 - - - - - - - - vpshufbitqmb %ymm16, %ymm17, %k2 {%k1} +# CHECK-NEXT: - - 0.50 0.50 0.50 0.50 - - - - - - vpshufbitqmb (%rdi), %ymm17, %k2 {%k1}