mirror of
https://github.com/alex-s168/website.git
synced 2025-09-10 01:05:07 +02:00
c
This commit is contained in:
@@ -32,7 +32,7 @@
|
|||||||
#context wimage(res-path()+"favicon.png", width:w)
|
#context wimage(res-path()+"favicon.png", width:w)
|
||||||
]
|
]
|
||||||
|
|
||||||
#let ic_url = "https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.sciencedirect.com/science/article/pii/S0890540197926432/pdf%3Fmd5%3D30965cec6dd7605a865bbec4076f65e4%26pid%3D1-s2.0-S0890540197926432-main.pdf&ved=2ahUKEwjc2NHahqaOAxXFGxAIHRgsIp0QFnoECBMQAQ&usg=AOvVaw0yzy07VkWVoJu4XSqaOqj3"
|
#let ic_url = "https://www.sciencedirect.com/science/article/pii/S0890540197926432"
|
||||||
|
|
||||||
#section[
|
#section[
|
||||||
This represents an #flink(ic_url)[interaction combinator] tree, that can be interpreted as a
|
This represents an #flink(ic_url)[interaction combinator] tree, that can be interpreted as a
|
||||||
|
@@ -34,8 +34,8 @@
|
|||||||
]
|
]
|
||||||
|
|
||||||
#section[
|
#section[
|
||||||
== Comparision with CPUs
|
== Comparison with CPUs
|
||||||
GPUs focus on operating on a lot of data at once (triangles, vertecies, pixels, ...),
|
GPUs focus on operating on a lot of data at once (triangles, vertices, pixels, ...),
|
||||||
while CPUs focus on high performance on a single core, and low compute delay.
|
while CPUs focus on high performance on a single core, and low compute delay.
|
||||||
]
|
]
|
||||||
|
|
||||||
@@ -54,7 +54,7 @@
|
|||||||
== Waves
|
== Waves
|
||||||
A wave is a SIMD processing unit consisting of typically 32 "lanes" (sometimes called threads).
|
A wave is a SIMD processing unit consisting of typically 32 "lanes" (sometimes called threads).
|
||||||
|
|
||||||
Each wave in a CU has seperate control flow, and doesn't have to be related.
|
Each wave in a CU has separate control flow, and doesn't have to be related.
|
||||||
|
|
||||||
Instructions that waves support:
|
Instructions that waves support:
|
||||||
- arithmetic operations
|
- arithmetic operations
|
||||||
@@ -86,7 +86,7 @@
|
|||||||
|
|
||||||
#section[
|
#section[
|
||||||
=== Why are the banks interlaved?
|
=== Why are the banks interlaved?
|
||||||
When the whole wave wants to read a contiguos array of `f32`, so when each wave performs `some_f32_array[lane_id()]`,
|
When the whole wave wants to read a contiguous array of `f32`, so when each wave performs `some_f32_array[lane_id()]`,
|
||||||
all 32 banks can be used at the same time.
|
all 32 banks can be used at the same time.
|
||||||
]
|
]
|
||||||
|
|
||||||
@@ -147,7 +147,7 @@
|
|||||||
- no fancy out of order or superscalar execution
|
- no fancy out of order or superscalar execution
|
||||||
- support standard 32 bit floating point, without exceptions.
|
- support standard 32 bit floating point, without exceptions.
|
||||||
|
|
||||||
Note that we won't specifiy the exact instruction encoding.
|
Note that we won't specify the exact instruction encoding.
|
||||||
]
|
]
|
||||||
|
|
||||||
#section[
|
#section[
|
||||||
|
Reference in New Issue
Block a user