heat: add more figures from perf counters

This commit is contained in:
Rodrigo Arias 2021-03-09 18:21:59 +01:00
parent 3bcbc62a98
commit 7b4da07dbf

View File

@ -2,6 +2,7 @@ library(ggplot2)
library(dplyr)
library(scales)
library(jsonlite)
library(viridis)
args=commandArgs(trailingOnly=TRUE)
@ -14,7 +15,7 @@ dataset = jsonlite::stream_in(file(input_file)) %>%
jsonlite::flatten()
# We only need the nblocks and time
df = select(dataset, config.cbs, config.rbs, perf.cache_misses) %>%
df = select(dataset, config.cbs, config.rbs, perf.cache_misses, perf.instructions, perf.cycles, time) %>%
rename(cbs=config.cbs, rbs=config.rbs)
df$cbs = as.factor(df$cbs)
@ -22,30 +23,55 @@ df$rbs = as.factor(df$rbs)
# Normalize the time by the median
df=group_by(df, cbs, rbs) %>%
mutate(median.time = median(time)) %>%
mutate(log.median.time = log(median.time)) %>%
mutate(median.misses = median(perf.cache_misses)) %>%
mutate(log.median.misses = log(median.misses)) %>%
ungroup()
mutate(median.instr= median(perf.instructions)) %>%
mutate(log.median.instr= log(median.instr)) %>%
mutate(median.cycles = median(perf.cycles)) %>%
mutate(median.cpi = median.cycles / median.instr) %>%
mutate(median.ipc = median.instr / median.cycles) %>%
mutate(median.ips = median.instr / median.time) %>%
mutate(median.cps = median.cycles / median.time) %>%
ungroup()# %>%
ppi=300
h=5
w=5
heatmap_plot = function(df, colname, title) {
p = ggplot(df, aes(x=cbs, y=rbs, fill=!!ensym(colname))) +
geom_raster() +
#scale_fill_gradient(high="black", low="white") +
scale_fill_viridis(option="plasma") +
coord_fixed() +
theme_bw() +
theme(axis.text.x=element_text(angle = -45, hjust = 0)) +
theme(plot.subtitle=element_text(size=8)) +
#guides(fill = guide_colorbar(barwidth=15, title.position="top")) +
guides(fill = guide_colorbar(barwidth=12, title.vjust=0.8)) +
labs(x="cbs", y="rbs",
title=sprintf("Heat granularity: %s", title),
subtitle=input_file) +
theme(legend.position="bottom")
k=1
ggsave(sprintf("%s.png", colname), plot=p, width=4.8*k, height=5*k, dpi=300)
ggsave(sprintf("%s.pdf", colname), plot=p, width=4.8*k, height=5*k, dpi=300)
}
png("heatmap.png", width=1.5*w*ppi, height=h*ppi, res=ppi)
#
## Create the plot with the normalized time vs nblocks
p = ggplot(df, aes(x=cbs, y=rbs, fill=log.median.misses)) +
geom_raster() +
scale_fill_gradient(high="black", low="white") +
coord_fixed() +
theme_bw() +
theme(plot.subtitle=element_text(size=8)) +
labs(x="cbs", y="rbs",
title=sprintf("Heat granularity: cache misses"),
subtitle=input_file)
heatmap_plot(df, "median.misses", "cache misses")
heatmap_plot(df, "log.median.misses", "cache misses")
heatmap_plot(df, "median.instr", "instructions")
heatmap_plot(df, "log.median.instr", "instructions")
heatmap_plot(df, "median.cycles", "cycles")
heatmap_plot(df, "median.ipc", "IPC")
heatmap_plot(df, "median.cpi", "cycles/instruction")
heatmap_plot(df, "median.ips", "instructions/second")
heatmap_plot(df, "median.cps", "cycles/second")
# Render the plot
print(p)
cutlevel = 0.5
# To plot the median.time we crop the larger values:
df_filtered = filter(df, between(median.time,
median(time) - (cutlevel * sd(time)),
median(time) + (cutlevel * sd(time))))
# Save the png image
dev.off()
heatmap_plot(df_filtered, "median.time", "execution time (seconds)")
heatmap_plot(df_filtered, "log.median.time", "execution time")