Loading page…
Master the fundamental concepts of simd & vectorization through this focused micro-challenge.
You have read the whole brief, and the concepts above stay free on every task. Writing and running the code needs a plan.
Three hints are available for this task, revealed one at a time inside the code workspace so you can struggle productively before seeing them.
Every task includes starter code, theory, and hidden tests so you can implement and verify locally in the browser.
How it worksRGBA images store four 8-bit channels per pixel (16 million colors times alpha). Inverting colors is 255 - channel, a perfect SIMD byte operation across 16 pixels per __m128i chunk.
cLoading…
Image kernels often hit DRAM bandwidth before compute limits. Align rows, process full vectors in the hot loop, and handle width remainder scalarly.
For example, a 1920x1080 frame has 2,073,600 pixels; vectorized inversion touches the same bytes as scalar code but issues fewer instructions.
For this exercise, you will invert a buffer with SSE2 byte intrinsics and compare ms/frame to scalar loops. This task asks you to treat alpha separately if premultiplication matters for your test image format.
Keep the relevant datasheet, ISA manual, or architecture textbook chapter open while you implement. When your output disagrees with the reference trace on the same program, the bug is usually a mis-decoded opcode, a stale register read, or a flag bit left unchanged after arithmetic.
For this exercise, you will use those habits while implementing the requirement in the starter code. Microarchitectural product names change across CPU generations, but the control ideas (fetch, bypass, cache lines, vector lanes) stay stable enough to debug from first principles.
Invert the colours of an RGBA image with SIMD while preserving alpha. Each pixel is 4 bytes, R G B A in memory. Loaded as a little-endian uint32, it is A<<24 | B<<16 | G<<8 | R, so inverting R, G and B is a single XOR with 0x00FFFFFF. A 128-bit register holds 4 pixels and a 256-bit one holds 8, so one vector XOR with a broadcast mask inverts 4 or 8 pixels at once.
cLoading…
W·H ≤ 1000000. A generated image's pixel i (row-major) is the value x after i + 1 steps of x = (x · 1103515245 + 12345) mod 2^31 from x = SEED, taken as R = x & 255, G = (x >> 8) & 255, B = (x >> 16) & 255, A = (x >> 24) & 127.
Per image:
cLoading…
sse2/avx2 count full vectors (4 or 8 pixels) and the leftover pixels handled one at a time, for the whole image as one flat array.checksum is 32-bit FNV-1a over the inverted image's bytes in memory order (R G B A for each pixel): start from h = 2166136261, and for each byte h = (h XOR byte) · 16777619 mod 2^32.alpha preserved if every output alpha equals its input alpha, else alpha CHANGED.Print a blank line between images.
Input:
cLoading…
Output:
cLoading…
00 FF FF FF in every pixel (_mm_xor_si128 with _mm_set1_epi32(0x00FFFFFF), or a GCC vector type). Invert the leftovers with the same XOR on a uint32_t.Hidden tests cover a single-pixel image, an image whose pixel count is an exact multiple of 8, a large generated image, and an image with alpha 0 and 255 pixels.