add latency and TP information through ibench

This commit is contained in:
Qingcai Jiang
2021-12-01 11:42:36 +08:00
parent fb2dbfa83f
commit be48dd6c1b

View File

@@ -1,8 +1,8 @@
osaca_version: 0.4.6
micro_architecture: TaiShan v110
micro_architecture: TaiShan v110 # https://en.wikichip.org/wiki/hisilicon/microarchitectures/taishan_v110
arch_code: tsv110
isa: AArch64
ROB_size: 128
ROB_size: 128 # https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AArch64/AArch64SchedTSV110.td#L21
retired_uOps_per_cycle: 4
scheduler_size: ~ # unknown
hidden_loads: false
@@ -29,6 +29,7 @@ instruction_forms:
throughput: 0.5
latency: 0.0
port_pressure: [[1, '12']]
# arithmetic instructions: add (from AArch64SchedTSV110.td and ibench)
- name: add
operands:
- class: register
@@ -40,6 +41,7 @@ instruction_forms:
throughput: 0.3333
latency: 1.0
port_pressure: [[1, '012']]
# memory instructions: ldur (data from AArch64SchedTSV110.td)
- name: ldur
operands:
- class: register
@@ -54,6 +56,7 @@ instruction_forms:
throughput: 0.5
latency: 4.0
port_pressure: [[1, '67']]
# arithmetic instructions: fmla (latency and throughput from ibench, port data missed)
- name: fmla
operands:
- class: register
@@ -65,10 +68,11 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
latency: 4.0
port_pressure: ~
throughput: 0.5
uops: ~
# arithmetic instructions: fdiv (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: fdiv
operands:
- class: register
@@ -80,10 +84,11 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 26.0
port_pressure: [[1, '45']]
throughput: 22.0
uops: 1
# arithmetic instructions: fsqrt (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: fsqrt
operands:
- class: register
@@ -95,10 +100,11 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: ~
latency: 22.0
port_pressure: [[1, '45']]
throughput: 34.0
uops: ~
# arithmetic instructions: fadd (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: fadd
operands:
- class: register
@@ -110,10 +116,10 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
latency: 4.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: ~
uops: 1
- name: add
operands:
- class: register
@@ -137,10 +143,10 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
latency: 4.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: ~
uops: 1
- name: fmul
operands:
- class: register
@@ -149,10 +155,10 @@ instruction_forms:
prefix: d
- class: register
prefix: d
latency: ~
port_pressure: ~
latency: 5.0
port_pressure: [[1, '45']]
throughput: 0.5
uops: ~
uops: 1
- name: fdiv
operands:
- class: register
@@ -164,10 +170,10 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 40.0
port_pressure: [[1, '45']]
throughput: 36.0
uops: 1
- name: frecpe
operands:
- class: register
@@ -176,10 +182,11 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
latency: 3.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: ~
uops: 1
# arithmetic instructions: subs (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: subs
operands:
- class: register
@@ -188,25 +195,27 @@ instruction_forms:
prefix: x
- class: immediate
imd: int
latency: ~
port_pressure: ~
latency: 1.0
port_pressure: [[1, '12']]
throughput: 0.5
uops: ~
uops: 1
# memory instructions: stur (data from AArch64SchedTSV110.td)
- name: stur
operands:
- class: register
prefix: q
- class: memory
base: x
offset: ~
index: ~
scale: 1
offset: '*'
index: '*'
scale: '*'
pre-indexed: false
post-indexed: false
latency: 0.0
port_pressure: ~
throughput: ~
uops: ~
throughput: 0.5
latency: 1.0
port_pressure: [[1, '67']]
uops: 1
# arithmetic instructions: fmla (latency and throughput from ibench, port data missed)
- name: fmla
operands:
- class: register
@@ -218,9 +227,9 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
latency: 5.0
port_pressure: ~
throughput: 1.0
throughput: 1.322
uops: ~
- name: mov
operands:
@@ -270,10 +279,11 @@ instruction_forms:
prefix: w
- class: immediate
imd: int
latency: ~
port_pressure: ~
latency: 1.0
port_pressure: [[1, '012']]
throughput: 0.33333
uops: ~
uops: 1
# miscellaneous instructions: dup (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: dup
operands:
- class: register
@@ -281,10 +291,11 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 2.0
port_pressure: [2, '45']
throughput: 0.667
uops: 2
# arithmetic instructions: frecpe (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: frecpe
operands:
- class: register
@@ -293,10 +304,10 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 3.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: 1
- name: fmul
operands:
- class: register
@@ -308,10 +319,10 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: 0.5
uops: ~
latency: 5.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: 1
- name: fadd
operands:
- class: register
@@ -320,10 +331,10 @@ instruction_forms:
prefix: d
- class: register
prefix: d
latency: ~
port_pressure: ~
latency: 4.0
port_pressure: [[1, '45']]
throughput: 0.5
uops: ~
uops: 1
- name: fsqrt
operands:
- class: register
@@ -335,10 +346,11 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 22.0
port_pressure: [[1, '45']]
throughput: 64.0
uops: 1
# arithmetic instructions: adds (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: adds
operands:
- class: register
@@ -347,25 +359,25 @@ instruction_forms:
prefix: x
- class: immediate
imd: int
latency: ~
port_pressure: ~
latency: 1.0
port_pressure: [[1, '12']]
throughput: 0.5
uops: ~
uops: 1
- name: stur
operands:
- class: register
prefix: d
- class: memory
base: x
offset: ~
index: ~
scale: 1
offset: '*'
index: '*'
scale: '*'
pre-indexed: false
post-indexed: false
latency: 0.0
port_pressure: ~
throughput: 1.0
uops: ~
throughput: 0.5
latency: 1.0
port_pressure: [[1, '67']]
uops: 1
- name: fsub
operands:
- class: register
@@ -377,10 +389,10 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: 0.5
uops: ~
latency: 5.0
port_pressure: [[1, '45']]
throughput: 1.321
uops: 1
- name: fmul
operands:
- class: register
@@ -392,10 +404,11 @@ instruction_forms:
- class: register
prefix: v
shape: d
latency: ~
port_pressure: ~
latency: 5.0
port_pressure: [[1, '45']]
throughput: 1.0
uops: ~
uops: 1
# arithmetic instructions: mul (latency and throughput from ibench, port data from AArch64SchedTSV110.td)
- name: mul
operands:
- class: register
@@ -404,10 +417,10 @@ instruction_forms:
prefix: x
- class: register
prefix: x
latency: ~
port_pressure: ~
latency: 4.0
port_pressure: [[1, '3']]
throughput: 1.0
uops: ~
uops: 1
- name: fadd
operands:
- class: register
@@ -419,10 +432,10 @@ instruction_forms:
- class: register
prefix: v
shape: s
latency: ~
port_pressure: ~
throughput: ~
uops: ~
latency: 5.0
port_pressure: [[1, '45']]
throughput: 1.321
uops: 1
- name: add
operands:
- class: register
@@ -447,36 +460,38 @@ instruction_forms:
port_pressure: []
throughput: 0.0
uops: 0
# miscellaneous instructions: cmp (throughput from ibench, latency and port data from AArch64SchedTSV110.td)
- name: cmp
operands:
- class: register
prefix: x
- class: register
prefix: x
latency: ~
port_pressure: ~
latency: 1.0
port_pressure: [1, '12']
throughput: 0.5
uops: ~
uops: 1
# miscellaneous instructions: mov (assumed free register renaming, register to register moves without conversion)
- name: fmov
operands:
- class: register
prefix: s
- class: immediate
imd: int
latency: ~
port_pressure: ~
throughput: 0.5
uops: ~
latency: 0.0
port_pressure: []
throughput: 0.0
uops: 0
- name: cmp
operands:
- class: register
prefix: w
- class: register
prefix: w
latency: ~
port_pressure: ~
latency: 1.0
port_pressure: [1, '12']
throughput: 0.5
uops: ~
uops: 1
- name: ldp
operands:
- class: register
@@ -494,6 +509,24 @@ instruction_forms:
latency: ~
port_pressure: [[2, '67']
uops: 2
# memory instructions: ldp (data from AArch64SchedTSV110.td)
- name: ldp
operands:
- class: register
prefix: x
- class: register
prefix: x
- class: memory
base: x
offset: '*'
index: '*'
scale: '*'
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: 8.0
port_pressure: [[1, '67'], [1, '012']]
uops: 2
- name: ldp
operands:
- class: register
@@ -508,7 +541,7 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
@@ -525,7 +558,7 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
@@ -542,8 +575,8 @@ instruction_forms:
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: ~
port_pressure: [[2, '67']]
latency: 8.0
port_pressure: [[1, '67'], [1, '012']]
uops: 2
- name: ldp
operands:
@@ -559,7 +592,7 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
@@ -576,7 +609,7 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
@@ -593,9 +626,9 @@ instruction_forms:
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: ~
port_pressure: [[2, '67']]
uops: 2
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
operands:
- class: register
@@ -610,7 +643,7 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
- name: ldp
@@ -627,9 +660,10 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: ~
latency: 9.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
# memory instructions: stp (data from AArch64SchedTSV110.td)
- name: stp
operands:
- class: register
@@ -644,7 +678,7 @@ instruction_forms:
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: 0.0
latency: 2.0
port_pressure: [[2, '67']
uops: 2
- name: stp
@@ -661,9 +695,9 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 2.0
port_pressure: [[1, '67'], [1, '012']]
uops: 2
- name: stp
operands:
- class: register
@@ -678,9 +712,9 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 2.0
port_pressure: [[1, '67'], [1, '012']]
uops: 2
- name: stp
operands:
- class: register
@@ -695,8 +729,8 @@ instruction_forms:
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67']]
latency: 2.0
port_pressure: [[2, '67']
uops: 2
- name: stp
operands:
@@ -712,9 +746,9 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 2.0
port_pressure: [[2, '67'], [1, '012']]
uops: 3
- name: stp
operands:
- class: register
@@ -729,9 +763,9 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 2.0
port_pressure: [[2, '67'], [1, '012']]
uops: 3
- name: stp
operands:
- class: register
@@ -746,7 +780,7 @@ instruction_forms:
pre-indexed: false
post-indexed: false
throughput: 1.0
latency: 0.0
latency: 2.0
port_pressure: [[2, '67']]
uops: 2
- name: stp
@@ -763,9 +797,9 @@ instruction_forms:
pre-indexed: true
post-indexed: false
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 3.0
port_pressure: [[2, '67'], [1, '012']]
uops: 3
- name: stp
operands:
- class: register
@@ -780,9 +814,9 @@ instruction_forms:
pre-indexed: false
post-indexed: true
throughput: 1.0
latency: 0.0
port_pressure: [[2, '67'], [2, '012']]
uops: 4
latency: 2.0
port_pressure: [[2, '67'], [1, '012']]
uops: 3
- name: ldr
operands:
- class: register