A flame graph visualizes hierarchical call stack data from performance profiling, where each horizontal bar represents a function in the call stack and its width is proportional to the time (or samples) spent in that function. Stacks are layered bottom-to-top showing caller-to-callee relationships. Invented by Brendan Gregg, flame graphs are the standard visualization for identifying CPU bottlenecks and hot code paths across all major programming languages and profiling tools.
#' anyplot.ai
#' flamegraph-basic: Flame Graph for Performance Profiling
#' Library: ggplot2 3.5.1 | R 4.4.1
#' Quality: 86/100 | Created: 2026-06-08
library(ggplot2)
library(dplyr)
library(tidyr)
library(ragg)
set.seed(42)
# Theme tokens — Imprint, theme-adaptive chrome
THEME <- Sys.getenv("ANYPLOT_THEME", "light")
PAGE_BG <- if (THEME == "light") "#FAF8F1" else "#1A1A17"
ELEVATED_BG <- if (THEME == "light") "#FFFDF6" else "#242420"
INK <- if (THEME == "light") "#1A1A17" else "#F0EFE8"
INK_SOFT <- if (THEME == "light") "#4A4A44" else "#B8B7B0"
INK_MUTED <- if (THEME == "light") "#6B6A63" else "#A8A79F"
# Imprint warm anchors — semantic exception for the conventional
# flame-graph aesthetic (matte red → ochre → amber).
FLAME_BASE <- "#AE3030" # matte red (Imprint pos 5) — base of flame
FLAME_MID <- "#BD8233" # ochre (Imprint pos 4)
FLAME_TIP <- "#DDCC77" # amber (Imprint anchor) — tip of flame
# Dual label inks — picked per bar via fill luminance so deep-flame rows
# (matte-red base) read in light ink while warm tip rows keep dark ink.
LABEL_DARK <- "#1A1A17"
LABEL_LIGHT <- "#F0EFE8"
# Data — simulated CPU profile of a model-training script.
# Each row is a leaf call stack and its sample count; parent widths
# are derived by summing all descendant leaves through the prefix.
stacks <- tibble::tribble(
~stack, ~value,
"main;parse_config;tokenize;regex_match", 24,
"main;parse_config;tokenize;trim_ws", 8,
"main;parse_config;validate_schema;type_check", 28,
"main;parse_config;validate_schema;range_check", 14,
"main;load_dataset;read_csv;buffered_read;syscall_read", 56,
"main;load_dataset;read_csv;buffered_read;prefetch", 12,
"main;load_dataset;read_csv;parse_row;cast_types", 38,
"main;load_dataset;read_csv;parse_row;split_columns", 21,
"main;load_dataset;decode_utf8", 22,
"main;train_model;preprocess;normalize", 31,
"main;train_model;preprocess;impute_missing", 14,
"main;train_model;forward_pass;matmul;simd_kernel", 152,
"main;train_model;forward_pass;matmul;dispatch_blas", 38,
"main;train_model;forward_pass;activation_relu", 28,
"main;train_model;backward_pass;grad_matmul;simd_kernel", 124,
"main;train_model;backward_pass;grad_matmul;dispatch_blas", 32,
"main;train_model;backward_pass;update_weights;step_sgd", 36,
"main;train_model;backward_pass;update_weights;apply_momentum", 18,
"main;evaluate;forward_pass;matmul;simd_kernel", 22,
"main;evaluate;forward_pass;matmul;dispatch_blas", 9,
"main;evaluate;forward_pass;activation_relu", 6,
"main;evaluate;compute_metrics;accuracy_top_k", 8,
"main;evaluate;compute_metrics;confusion_matrix", 6,
"main;render_report;format_summary", 9,
"main;render_report;write_output", 6,
"main;render_report;compress_artifacts", 4
)
# Expand each leaf stack into per-depth prefix frames, then aggregate.
expand_stack <- function(stack_str, val) {
parts <- strsplit(stack_str, ";", fixed = TRUE)[[1]]
tibble::tibble(
depth = seq_along(parts),
func = parts,
prefix = vapply(seq_along(parts),
function(i) paste(parts[seq_len(i)], collapse = ";"),
character(1)),
value = val
)
}
frames <- do.call(rbind, Map(expand_stack, stacks$stack, stacks$value)) %>%
group_by(depth, prefix) %>%
summarise(value = sum(value), .groups = "drop") %>%
mutate(
func = vapply(strsplit(prefix, ";", fixed = TRUE),
function(p) p[length(p)], character(1)),
parent = vapply(strsplit(prefix, ";", fixed = TRUE),
function(p) if (length(p) == 1) NA_character_
else paste(p[-length(p)], collapse = ";"),
character(1))
) %>%
arrange(depth, prefix) %>%
as.data.frame()
# Lay out x positions: each child sits within its parent's span,
# siblings ordered alphabetically (standard flame-graph convention).
frames$xmin <- NA_real_
frames$xmax <- NA_real_
for (d in sort(unique(frames$depth))) {
if (d == 1) {
cum <- 0
for (i in which(frames$depth == 1)) {
frames$xmin[i] <- cum
frames$xmax[i] <- cum + frames$value[i]
cum <- cum + frames$value[i]
}
} else {
for (par in unique(frames$parent[frames$depth == d])) {
par_row <- which(frames$prefix == par)
cum <- frames$xmin[par_row]
kids <- which(frames$parent == par & frames$depth == d)
kids <- kids[order(frames$prefix[kids])]
for (i in kids) {
frames$xmin[i] <- cum
frames$xmax[i] <- cum + frames$value[i]
cum <- cum + frames$value[i]
}
}
}
}
# Y geometry — one row per depth, narrow gap between rows.
frames$ymin <- frames$depth - 1 + 0.05
frames$ymax <- frames$depth - 0.05
# Conditional labels — truncate function names that don't fit the box,
# drop entirely when the box is too narrow for even three characters.
total_width <- max(frames$xmax)
frames$frac <- (frames$xmax - frames$xmin) / total_width
char_budget <- 95
frames$max_chars <- floor(frames$frac * char_budget)
frames$label <- vapply(seq_len(nrow(frames)), function(i) {
name <- frames$func[i]
max_n <- frames$max_chars[i]
if (max_n >= nchar(name)) return(name)
if (max_n >= 4) return(paste0(substr(name, 1, max_n - 1L), "…"))
""
}, character(1))
max_depth <- max(frames$depth)
# Pick a contrasting label ink per bar — WCAG relative luminance on the
# interpolated gradient fill, threshold tuned so depth-1/2 dark-red rows
# get light text (AA) and ochre/amber rows keep dark text.
fill_ramp <- grDevices::colorRamp(c(FLAME_BASE, FLAME_MID, FLAME_TIP))
depth_norm <- (frames$depth - 1) / (max_depth - 1)
fill_rgb <- fill_ramp(depth_norm) / 255
fill_lin <- ifelse(fill_rgb <= 0.03928,
fill_rgb / 12.92,
((fill_rgb + 0.055) / 1.055) ^ 2.4)
frames$lum <- 0.2126 * fill_lin[, 1] +
0.7152 * fill_lin[, 2] +
0.0722 * fill_lin[, 3]
frames$label_color <- ifelse(frames$lum < 0.25, LABEL_LIGHT, LABEL_DARK)
# Plot
p <- ggplot(frames,
aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax, fill = depth)) +
geom_rect(color = ELEVATED_BG, linewidth = 0.5) +
geom_text(
aes(x = (xmin + xmax) / 2, y = (ymin + ymax) / 2,
label = label, color = label_color),
size = 3.0, family = "sans"
) +
scale_color_identity() +
scale_fill_gradientn(
colors = c(FLAME_BASE, FLAME_MID, FLAME_TIP),
name = "depth"
) +
scale_x_continuous(
expand = expansion(add = c(0, 0)),
labels = scales::label_comma()
) +
scale_y_continuous(
expand = expansion(add = c(0.05, 0.25)),
breaks = seq(0.5, max_depth - 0.5, by = 1),
labels = as.character(seq_len(max_depth))
) +
labs(
title = "flamegraph-basic · r · ggplot2 · anyplot.ai",
x = "Samples (CPU time)",
y = "Call stack depth"
) +
theme_minimal(base_size = 8) +
theme(
plot.background = element_rect(fill = PAGE_BG, color = PAGE_BG),
panel.background = element_rect(fill = ELEVATED_BG, color = NA),
panel.grid.major = element_blank(),
panel.grid.minor = element_blank(),
axis.title = element_text(color = INK, size = 10),
axis.title.x = element_text(margin = margin(t = 8)),
axis.title.y = element_text(margin = margin(r = 8)),
axis.text = element_text(color = INK_SOFT, size = 8),
axis.ticks.x = element_line(color = INK_SOFT),
axis.ticks.length = unit(3, "pt"),
axis.line.x = element_line(color = INK_SOFT),
axis.line.y = element_line(color = INK_SOFT, linewidth = 0.3),
plot.title = element_text(color = INK, size = 12,
margin = margin(b = 10)),
plot.margin = margin(t = 12, r = 18, b = 10, l = 12),
legend.position = "none"
)
ggsave(
filename = sprintf("plot-%s.png", THEME),
plot = p,
device = ragg::agg_png,
width = 8,
height = 4.5,
units = "in",
dpi = 400
)
Runnable source as JSON, for any HTTP client: https://api.anyplot.ai/specs/flamegraph-basic/ggplot2/code. Any spec id and library id listed in llms-full.txt fit the same URL shape; every URL below is complete and callable.
{
"spec_id": "flamegraph-basic",
"language": "r",
"library": "ggplot2",
"page": "https://anyplot.ai/flamegraph-basic/r/ggplot2",
"hub": "https://anyplot.ai/flamegraph-basic",
"code_json": "https://api.anyplot.ai/specs/flamegraph-basic/ggplot2/code",
"spec_json": "https://api.anyplot.ai/specs/flamegraph-basic",
"render_light_png": "https://storage.googleapis.com/anyplot-images/plots/flamegraph-basic/r/ggplot2/plot-light.png",
"render_dark_png": "https://storage.googleapis.com/anyplot-images/plots/flamegraph-basic/r/ggplot2/plot-dark.png",
"quality_score": 86.0,
"license": "MIT",
"guide": "https://anyplot.ai/llms.txt"
}Part of Flame Graph for Performance Profiling on anyplot.ai.