From e42abef9bf3adb45c99de6632ae72da3e8b727e0 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sun, 11 Dec 2022 15:52:38 +0000 Subject: [PATCH] [MCA][X86] Add test coverage for ERI instructions --- .../llvm-mca/X86/Generic/resources-avx512er.s | 373 +++++++++++++++++++++ 1 file changed, 373 insertions(+) create mode 100644 llvm/test/tools/llvm-mca/X86/Generic/resources-avx512er.s diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512er.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512er.s new file mode 100644 index 0000000..034fc6d --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512er.s @@ -0,0 +1,373 @@ +# 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 + +vexp2pd %zmm16, %zmm19 +vexp2pd (%rax), %zmm19 +vexp2pd (%rax){1to8}, %zmm19 +vexp2pd %zmm16, %zmm19 {k1} +vexp2pd (%rax), %zmm19 {k1} +vexp2pd (%rax){1to8}, %zmm19 {k1} +vexp2pd %zmm16, %zmm19 {z}{k1} +vexp2pd (%rax), %zmm19 {z}{k1} +vexp2pd (%rax){1to8}, %zmm19 {z}{k1} + +vexp2pd {sae}, %zmm16, %zmm19 +vexp2pd {sae}, %zmm16, %zmm19 {k1} +vexp2pd {sae}, %zmm16, %zmm19 {z}{k1} + +vexp2ps %zmm16, %zmm19 +vexp2ps (%rax), %zmm19 +vexp2ps (%rax){1to16}, %zmm19 +vexp2ps %zmm16, %zmm19 {k1} +vexp2ps (%rax), %zmm19 {k1} +vexp2ps (%rax){1to16}, %zmm19 {k1} +vexp2ps %zmm16, %zmm19 {z}{k1} +vexp2ps (%rax), %zmm19 {z}{k1} +vexp2ps (%rax){1to16}, %zmm19 {z}{k1} + +vexp2ps {sae}, %zmm16, %zmm19 +vexp2ps {sae}, %zmm16, %zmm19 {k1} +vexp2ps {sae}, %zmm16, %zmm19 {z}{k1} + +vrcp28pd %zmm16, %zmm19 +vrcp28pd (%rax), %zmm19 +vrcp28pd (%rax){1to8}, %zmm19 +vrcp28pd %zmm16, %zmm19 {k1} +vrcp28pd (%rax), %zmm19 {k1} +vrcp28pd (%rax){1to8}, %zmm19 {k1} +vrcp28pd %zmm16, %zmm19 {z}{k1} +vrcp28pd (%rax), %zmm19 {z}{k1} +vrcp28pd (%rax){1to8}, %zmm19 {z}{k1} + +vrcp28pd {sae}, %zmm16, %zmm19 +vrcp28pd {sae}, %zmm16, %zmm19 {k1} +vrcp28pd {sae}, %zmm16, %zmm19 {z}{k1} + +vrcp28ps %zmm16, %zmm19 +vrcp28ps (%rax), %zmm19 +vrcp28ps (%rax){1to16}, %zmm19 +vrcp28ps %zmm16, %zmm19 {k1} +vrcp28ps (%rax), %zmm19 {k1} +vrcp28ps (%rax){1to16}, %zmm19 {k1} +vrcp28ps %zmm16, %zmm19 {z}{k1} +vrcp28ps (%rax), %zmm19 {z}{k1} +vrcp28ps (%rax){1to16}, %zmm19 {z}{k1} + +vrcp28ps {sae}, %zmm16, %zmm19 +vrcp28ps {sae}, %zmm16, %zmm19 {k1} +vrcp28ps {sae}, %zmm16, %zmm19 {z}{k1} + +vrcp28sd %xmm16, %xmm17, %xmm19 +vrcp28sd (%rax), %xmm17, %xmm19 +vrcp28sd %xmm16, %xmm17, %xmm19 {k1} +vrcp28sd (%rax), %xmm17, %xmm19 {k1} +vrcp28sd %xmm16, %xmm17, %xmm19 {z}{k1} +vrcp28sd (%rax), %xmm17, %xmm19 {z}{k1} + +vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 +vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {k1} +vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {z}{k1} + +vrcp28ss %xmm16, %xmm17, %xmm19 +vrcp28ss (%rax), %xmm17, %xmm19 +vrcp28ss %xmm16, %xmm17, %xmm19 {k1} +vrcp28ss (%rax), %xmm17, %xmm19 {k1} +vrcp28ss %xmm16, %xmm17, %xmm19 {z}{k1} +vrcp28ss (%rax), %xmm17, %xmm19 {z}{k1} + +vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 +vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {k1} +vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {z}{k1} + +vrsqrt28pd %zmm16, %zmm19 +vrsqrt28pd (%rax), %zmm19 +vrsqrt28pd (%rax){1to8}, %zmm19 +vrsqrt28pd %zmm16, %zmm19 {k1} +vrsqrt28pd (%rax), %zmm19 {k1} +vrsqrt28pd (%rax){1to8}, %zmm19 {k1} +vrsqrt28pd %zmm16, %zmm19 {z}{k1} +vrsqrt28pd (%rax), %zmm19 {z}{k1} +vrsqrt28pd (%rax){1to8}, %zmm19 {z}{k1} + +vrsqrt28pd {sae}, %zmm16, %zmm19 +vrsqrt28pd {sae}, %zmm16, %zmm19 {k1} +vrsqrt28pd {sae}, %zmm16, %zmm19 {z}{k1} + +vrsqrt28ps %zmm16, %zmm19 +vrsqrt28ps (%rax), %zmm19 +vrsqrt28ps (%rax){1to16}, %zmm19 +vrsqrt28ps %zmm16, %zmm19 {k1} +vrsqrt28ps (%rax), %zmm19 {k1} +vrsqrt28ps (%rax){1to16}, %zmm19 {k1} +vrsqrt28ps %zmm16, %zmm19 {z}{k1} +vrsqrt28ps (%rax), %zmm19 {z}{k1} +vrsqrt28ps (%rax){1to16}, %zmm19 {z}{k1} + +vrsqrt28ps {sae}, %zmm16, %zmm19 +vrsqrt28ps {sae}, %zmm16, %zmm19 {k1} +vrsqrt28ps {sae}, %zmm16, %zmm19 {z}{k1} + +vrsqrt28sd %xmm16, %xmm17, %xmm19 +vrsqrt28sd (%rax), %xmm17, %xmm19 +vrsqrt28sd %xmm16, %xmm17, %xmm19 {k1} +vrsqrt28sd (%rax), %xmm17, %xmm19 {k1} +vrsqrt28sd %xmm16, %xmm17, %xmm19 {z}{k1} +vrsqrt28sd (%rax), %xmm17, %xmm19 {z}{k1} + +vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 +vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {k1} +vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {z}{k1} + +vrsqrt28ss %xmm16, %xmm17, %xmm19 +vrsqrt28ss (%rax), %xmm17, %xmm19 +vrsqrt28ss %xmm16, %xmm17, %xmm19 {k1} +vrsqrt28ss (%rax), %xmm17, %xmm19 {k1} +vrsqrt28ss %xmm16, %xmm17, %xmm19 {z}{k1} +vrsqrt28ss (%rax), %xmm17, %xmm19 {z}{k1} + +vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 +vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {k1} +vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {z}{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 vexp2pd %zmm16, %zmm19 +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax), %zmm19 +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: 1 3 1.00 vexp2pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: 1 3 1.00 vexp2pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 1.00 * vexp2pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: 1 3 1.00 vexp2pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 1 3 1.00 vexp2pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 1 3 1.00 vexp2pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 1 3 1.00 vexp2ps %zmm16, %zmm19 +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax), %zmm19 +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: 1 3 1.00 vexp2ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: 1 3 1.00 vexp2ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 1.00 * vexp2ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: 1 3 1.00 vexp2ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 1 3 1.00 vexp2ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 1 3 1.00 vexp2ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrcp28pd %zmm16, %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax), %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrcp28pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrcp28pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrcp28pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrcp28pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrcp28pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrcp28pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrcp28ps %zmm16, %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax), %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrcp28ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrcp28ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrcp28ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrcp28ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrcp28ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrcp28ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrcp28sd %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 2 11 1.00 * vrcp28sd (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrcp28sd %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 11 1.00 * vrcp28sd (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrcp28sd %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 11 1.00 * vrcp28sd (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrcp28ss %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 2 11 1.00 * vrcp28ss (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrcp28ss %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 11 1.00 * vrcp28ss (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrcp28ss %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 11 1.00 * vrcp28ss (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd %zmm16, %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax), %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrsqrt28pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps %zmm16, %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax), %zmm19 +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: 4 14 2.00 * vrsqrt28ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: 3 7 2.00 vrsqrt28ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 2 11 1.00 * vrsqrt28sd (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 11 1.00 * vrsqrt28sd (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 11 1.00 * vrsqrt28sd (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 2 11 1.00 * vrsqrt28ss (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 11 1.00 * vrsqrt28ss (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 11 1.00 * vrsqrt28ss (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 1.00 vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} + +# 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: - - 156.00 24.00 - 24.00 24.00 24.00 + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd %zmm16, %zmm19 +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax), %zmm19 +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - - - vexp2pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps %zmm16, %zmm19 +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax), %zmm19 +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - 0.50 0.50 vexp2ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - - - - vexp2ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax), %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax), %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrcp28ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrcp28ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28sd (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28sd (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28sd (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28ss (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28ss (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrcp28ss (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrcp28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax), %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax){1to8}, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax){1to8}, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28pd (%rax){1to8}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28pd {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax), %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax){1to16}, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax), %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax){1to16}, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax), %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 0.50 0.50 vrsqrt28ps (%rax){1to16}, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps {sae}, %zmm16, %zmm19 +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps {sae}, %zmm16, %zmm19 {%k1} +# CHECK-NEXT: - - 2.50 - - 0.50 - - vrsqrt28ps {sae}, %zmm16, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28sd (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28sd (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28sd (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28sd {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28ss (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28ss (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - 0.50 0.50 vrsqrt28ss (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 1.00 - - - - - vrsqrt28ss {sae}, %xmm16, %xmm17, %xmm19 {%k1} {z} -- 2.7.4