blackbeard/miner
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases 2 Wiki Activity

Use big-endian bytes consistently in qpow kernelSHA3 digest and

Browse Source 
comparisons
This commit is contained in:
rob thijssen
2025-09-13 10:36:23 +03:00
parent 860503792c
commit ff477754d2
1 changed files with 13 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/engine-gpu-cuda/src/kernels/qpow_kernel.cu
+13 -29
View File
@@ -593,36 +593,19 @@ const uint64_t n0i = C_CONSTS_READY ? C_N0_INV : n0_inv;
uint64_t yhat[8];
to_mont_512(y0_loc, r2_loc, n_loc, n0i, yhat);
// Prepare target/threshold lane-LE bytes (match SHA3 output layout) for XOR with h_le
// Prepare target/threshold big-endian bytes (numeric)
uint8_t target_be_bytes[64], thresh_be_bytes[64];
if (C_CONSTS_READY) {
#pragma unroll
for (int i = 0; i < 8; ++i) {
// C_TARGET/C_THRESH are big-endian 64-bit limbs [0]=MSL; map to lane-LE with reversed limb order
store64_le(&target_be_bytes[(7 - i) * 8], C_TARGET[i]);
store64_le(&thresh_be_bytes[(7 - i) * 8], C_THRESH[i]);
store64_be(&target_be_bytes[i * 8], C_TARGET[i]);
store64_be(&thresh_be_bytes[i * 8], C_THRESH[i]);
}
} else {
#pragma unroll
for (int i = 0; i < 8; ++i) {
// Convert input big-endian bytes to lane-LE with reversed limb order
target_be_bytes[(7 - i) * 8 + 0] = target_be[i * 8 + 7];
target_be_bytes[(7 - i) * 8 + 1] = target_be[i * 8 + 6];
target_be_bytes[(7 - i) * 8 + 2] = target_be[i * 8 + 5];
target_be_bytes[(7 - i) * 8 + 3] = target_be[i * 8 + 4];
target_be_bytes[(7 - i) * 8 + 4] = target_be[i * 8 + 3];
target_be_bytes[(7 - i) * 8 + 5] = target_be[i * 8 + 2];
target_be_bytes[(7 - i) * 8 + 6] = target_be[i * 8 + 1];
target_be_bytes[(7 - i) * 8 + 7] = target_be[i * 8 + 0];
thresh_be_bytes[(7 - i) * 8 + 0] = threshold_be[i * 8 + 7];
thresh_be_bytes[(7 - i) * 8 + 1] = threshold_be[i * 8 + 6];
thresh_be_bytes[(7 - i) * 8 + 2] = threshold_be[i * 8 + 5];
thresh_be_bytes[(7 - i) * 8 + 3] = threshold_be[i * 8 + 4];
thresh_be_bytes[(7 - i) * 8 + 4] = threshold_be[i * 8 + 3];
thresh_be_bytes[(7 - i) * 8 + 5] = threshold_be[i * 8 + 2];
thresh_be_bytes[(7 - i) * 8 + 6] = threshold_be[i * 8 + 1];
thresh_be_bytes[(7 - i) * 8 + 7] = threshold_be[i * 8 + 0];
for (int i = 0; i < 64; ++i) {
target_be_bytes[i] = target_be[i];
thresh_be_bytes[i] = threshold_be[i];
}
}
@@ -654,11 +637,12 @@ for (uint32_t j = 0; j < iters; ++j) {
uint8_t h_le[64];
sha3_512_64bytes_le(y_be, h_le);
// Use raw SHA3 output bytes (lane-LE) directly for distance and debug
// Convert digest to big-endian numeric bytes
uint8_t digest_be[64];
#pragma unroll
for (int i = 0; i < 64; ++i) {
digest_be[i] = h_le[i];
for (int i = 0; i < 8; ++i) {
uint64_t w = load64_le(&h_le[i * 8]);
store64_be(&digest_be[(7 - i) * 8], w);
}
// distance = target_be XOR digest_be (bytewise, big-endian order)
@@ -668,7 +652,7 @@ for (uint32_t j = 0; j < iters; ++j) {
dist_be[i] = target_be_bytes[i] ^ digest_be[i];
}
// Compare distance <= threshold (lexicographic on lane-LE bytes)
// Compare distance <= threshold (lexicographic on big-endian bytes)
bool decision = be64_leq(dist_be, thresh_be_bytes);
// Optional sampler (first thread/iter): capture y/H/target/thresh for parity
@@ -676,7 +660,7 @@ for (uint32_t j = 0; j < iters; ++j) {
#pragma unroll
for (int i = 0; i < 64; ++i) {
C_SAMPLER_Y_BE[i] = y_be[i];
C_SAMPLER_H_BE[i] = h_le[i];
C_SAMPLER_H_BE[i] = digest_be[i];
C_SAMPLER_TARGET_BE[i] = target_be_bytes[i];
C_SAMPLER_THRESH_BE[i] = thresh_be_bytes[i];
}
@@ -707,7 +691,7 @@ for (uint32_t j = 0; j < iters; ++j) {
if (out_dbg_h_be) {
#pragma unroll
for (int i = 0; i < 64; ++i) {
out_dbg_h_be[i] = h_le[i];
out_dbg_h_be[i] = digest_be[i];
}
}
}