From 474e41f1b9ff6683df0afc937a17ae13b68a97b6 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sat, 19 Nov 2022 21:46:14 +0000 Subject: [PATCH] [MCA][X86] Add test coverage for BF16 instructions --- .../llvm-mca/X86/Generic/resources-avx512bf16.s | 113 ++++++++++++ .../llvm-mca/X86/Generic/resources-avx512bf16vl.s | 197 +++++++++++++++++++++ 2 files changed, 310 insertions(+) create mode 100644 llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16.s create mode 100644 llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16vl.s diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16.s new file mode 100644 index 0000000..892c008 --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16.s @@ -0,0 +1,113 @@ +# 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 + +vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 +vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 +vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 +vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {k1} +vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {k1} +vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {k1} +vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {k1}{z} +vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {k1}{z} +vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {k1}{z} + +vcvtneps2bf16 %zmm16, %ymm19 +vcvtneps2bf16 (%rax), %ymm19 +vcvtneps2bf16 (%rax){1to16}, %ymm19 +vcvtneps2bf16 %zmm16, %ymm19 {k1} +vcvtneps2bf16 (%rax), %ymm19 {k1} +vcvtneps2bf16 (%rax){1to16}, %ymm19 {k1} +vcvtneps2bf16 %zmm16, %ymm19 {k1}{z} +vcvtneps2bf16 (%rax), %ymm19 {k1}{z} +vcvtneps2bf16 (%rax){1to16}, %ymm19 {k1}{z} + +vdpbf16ps %zmm16, %zmm17, %zmm19 +vdpbf16ps (%rax), %zmm17, %zmm19 +vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 +vdpbf16ps %zmm16, %zmm17, %zmm19 {k1} +vdpbf16ps (%rax), %zmm17, %zmm19 {k1} +vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {k1} +vdpbf16ps %zmm16, %zmm17, %zmm19 {k1}{z} +vdpbf16ps (%rax), %zmm17, %zmm19 {k1}{z} +vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {k1}{z} + +# 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: 2 4 1.00 vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %zmm16, %ymm19 +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax), %ymm19 +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to16}, %ymm19 +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %zmm16, %ymm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax), %ymm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to16}, %ymm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %zmm16, %ymm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax), %ymm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to16}, %ymm19 {%k1} {z} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %zmm16, %zmm17, %zmm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %zmm17, %zmm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %zmm16, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %zmm16, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {%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: - - 4.50 22.50 - 18.00 9.00 9.00 + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %zmm16, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to16}, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %zmm16, %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax), %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to16}, %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %zmm16, %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax), %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to16}, %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %zmm16, %ymm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax), %ymm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to16}, %ymm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %zmm16, %zmm17, %zmm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %zmm17, %zmm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %zmm16, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %zmm16, %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %zmm17, %zmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to16}, %zmm17, %zmm19 {%k1} {z} diff --git a/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16vl.s b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16vl.s new file mode 100644 index 0000000..e743245 --- /dev/null +++ b/llvm/test/tools/llvm-mca/X86/Generic/resources-avx512bf16vl.s @@ -0,0 +1,197 @@ +# 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 + +vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 +vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 +vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 +vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {k1} +vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {k1} +vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {k1} +vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {k1}{z} +vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {k1}{z} +vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {k1}{z} + +vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 +vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 +vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 +vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {k1} +vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {k1} +vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {k1} +vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {k1}{z} +vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {k1}{z} +vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {k1}{z} + +vcvtneps2bf16x %xmm16, %xmm19 +vcvtneps2bf16x (%rax), %xmm19 +vcvtneps2bf16 (%rax){1to4}, %xmm19 +vcvtneps2bf16 %xmm16, %xmm19 {k1} +vcvtneps2bf16x (%rax), %xmm19 {k1} +vcvtneps2bf16 (%rax){1to4}, %xmm19 {k1} +vcvtneps2bf16 %xmm16, %xmm19 {k1}{z} +vcvtneps2bf16x (%rax), %xmm19 {k1}{z} +vcvtneps2bf16 (%rax){1to4}, %xmm19 {k1}{z} + +vcvtneps2bf16 %ymm16, %xmm19 +vcvtneps2bf16y (%rax), %xmm19 +vcvtneps2bf16 (%rax){1to8}, %xmm19 +vcvtneps2bf16 %ymm16, %xmm19 {k1} +vcvtneps2bf16y (%rax), %xmm19 {k1} +vcvtneps2bf16 (%rax){1to8}, %xmm19 {k1} +vcvtneps2bf16 %ymm16, %xmm19 {k1}{z} +vcvtneps2bf16y (%rax), %xmm19 {k1}{z} +vcvtneps2bf16 (%rax){1to8}, %xmm19 {k1}{z} + +vdpbf16ps %xmm16, %xmm17, %xmm19 +vdpbf16ps (%rax), %xmm17, %xmm19 +vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 +vdpbf16ps %xmm16, %xmm17, %xmm19 {k1} +vdpbf16ps (%rax), %xmm17, %xmm19 {k1} +vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {k1} +vdpbf16ps %xmm16, %xmm17, %xmm19 {k1}{z} +vdpbf16ps (%rax), %xmm17, %xmm19 {k1}{z} +vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {k1}{z} + +vdpbf16ps %ymm16, %ymm17, %ymm19 +vdpbf16ps (%rax), %ymm17, %ymm19 +vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 +vdpbf16ps %ymm16, %ymm17, %ymm19 {k1} +vdpbf16ps (%rax), %ymm17, %ymm19 {k1} +vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {k1} +vdpbf16ps %ymm16, %ymm17, %ymm19 {k1}{z} +vdpbf16ps (%rax), %ymm17, %ymm19 {k1}{z} +vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {k1}{z} + +# 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: 2 4 1.00 vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 3 10 1.00 * vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %xmm16, %xmm19 +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16x (%rax), %xmm19 +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16 (%rax){1to4}, %xmm19 +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %xmm16, %xmm19 {%k1} +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16x (%rax), %xmm19 {%k1} +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16 (%rax){1to4}, %xmm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %xmm16, %xmm19 {%k1} {z} +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16x (%rax), %xmm19 {%k1} {z} +# CHECK-NEXT: 3 10 1.00 * vcvtneps2bf16 (%rax){1to4}, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %ymm16, %xmm19 +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16y (%rax), %xmm19 +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to8}, %xmm19 +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %ymm16, %xmm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16y (%rax), %xmm19 {%k1} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to8}, %xmm19 {%k1} +# CHECK-NEXT: 2 4 1.00 vcvtneps2bf16 %ymm16, %xmm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16y (%rax), %xmm19 {%k1} {z} +# CHECK-NEXT: 3 11 1.00 * vcvtneps2bf16 (%rax){1to8}, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %xmm17, %xmm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %ymm16, %ymm17, %ymm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %ymm17, %ymm19 +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %ymm16, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: 1 5 0.50 vdpbf16ps %ymm16, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax), %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: 2 10 0.50 * vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {%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: - - 9.00 45.00 - 36.00 18.00 18.00 + +# CHECK: Resource pressure by instruction: +# CHECK-NEXT: [0] [1] [2] [3] [4] [5] [6.0] [6.1] Instructions: +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to4}, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtne2ps2bf16 %ymm16, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax), %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtne2ps2bf16 (%rax){1to8}, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %xmm16, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16x (%rax), %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to4}, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %xmm16, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16x (%rax), %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to4}, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %xmm16, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16x (%rax), %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to4}, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %ymm16, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16y (%rax), %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to8}, %xmm19 +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %ymm16, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16y (%rax), %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to8}, %xmm19 {%k1} +# CHECK-NEXT: - - - 1.00 - 1.00 - - vcvtneps2bf16 %ymm16, %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16y (%rax), %xmm19 {%k1} {z} +# CHECK-NEXT: - - - 1.00 - 1.00 0.50 0.50 vcvtneps2bf16 (%rax){1to8}, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %xmm16, %xmm17, %xmm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %xmm17, %xmm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %xmm16, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %xmm16, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to4}, %xmm17, %xmm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %ymm16, %ymm17, %ymm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %ymm17, %ymm19 +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %ymm16, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {%k1} +# CHECK-NEXT: - - 0.50 0.50 - - - - vdpbf16ps %ymm16, %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax), %ymm17, %ymm19 {%k1} {z} +# CHECK-NEXT: - - 0.50 0.50 - - 0.50 0.50 vdpbf16ps (%rax){1to8}, %ymm17, %ymm19 {%k1} {z} -- 2.7.4