Skip to content

Commit 6dba7c9

Browse files
shai-almogshai-almog
committed
Fixed ci allocations and removed allocations
1 parent d49f9af commit 6dba7c9

File tree

2 files changed

+47
-105
lines changed
  • CodenameOne/src/com/codename1/util
  • scripts/hellocodenameone/common/src/main/java/com/codenameone/examples/hellocodenameone/tests

2 files changed

+47
-105
lines changed

‎CodenameOne/src/com/codename1/util/Base64.java‎

Lines changed: 37 additions & 103 deletions
Original file line numberDiff line numberDiff line change
@@ -452,7 +452,6 @@ public static int encodeNoNewline(byte[] in, byte[] out) {
452452
@DisableDebugInfo
453453
@DisableNullChecksAndArrayBoundsChecks
454454
public static int encodeNoNewlineSimd(byte[] in, int inOffset, int inLength, byte[] out, int outOffset, int[] scratch) {
455-
Simd simd = Simd.get();
456455
int outputLength = ((inLength + 2) / 3) * 4;
457456
if (out.length - outOffset < outputLength) {
458457
throw new IllegalArgumentException("Output buffer too small for encoded data");
@@ -461,56 +460,45 @@ public static int encodeNoNewlineSimd(byte[] in, int inOffset, int inLength, byt
461460
return 0;
462461
}
463462
requireScratch(scratch);
464-
requireSimdApiArrays(simd, in, out, scratch);
465-
466-
final int b0 = 0;
467-
final int b1 = b0 + SIMD_LANES;
468-
final int b2 = b1 + SIMD_LANES;
469-
final int s0 = b2 + SIMD_LANES;
470-
final int s1 = s0 + SIMD_LANES;
471-
final int s2 = s1 + SIMD_LANES;
472-
final int s3 = s2 + SIMD_LANES;
473-
final int t0 = s3 + SIMD_LANES;
474-
final int t1 = t0 + SIMD_LANES;
475-
final int c3 = t1 + SIMD_LANES;
476-
final int c15 = c3 + SIMD_LANES;
477-
final int c63 = c15 + SIMD_LANES;
478-
479-
for (int lane = 0; lane < SIMD_LANES; lane++) {
480-
scratch[c3 + lane] = 3;
481-
scratch[c15 + lane] = 15;
482-
scratch[c63 + lane] = 63;
483-
}
463+
requireSimdApiArrays(Simd.get(), in, out, scratch);
484464

485465
int end = inOffset + inLength - (inLength % 3);
486-
int simdEnd = end - ((end - inOffset) % 48);
487466
int inIndex = inOffset;
488467
int outIndex = outOffset;
489-
for (; inIndex < simdEnd; inIndex += 48) {
490-
for (int lane = 0; lane < SIMD_LANES; lane++) {
491-
int src = inIndex + lane * 3;
492-
scratch[b0 + lane] = in[src] & 0xff;
493-
scratch[b1 + lane] = in[src + 1] & 0xff;
494-
scratch[b2 + lane] = in[src + 2] & 0xff;
495-
}
496-
497-
simd.shrLogical(scratch, b0, 2, scratch, s0, SIMD_LANES);
498-
simd.and(scratch, b0, scratch, c3, scratch, t0, SIMD_LANES);
499-
simd.shl(scratch, t0, 4, scratch, t0, SIMD_LANES);
500-
simd.shrLogical(scratch, b1, 4, scratch, t1, SIMD_LANES);
501-
simd.or(scratch, t0, scratch, t1, scratch, s1, SIMD_LANES);
502-
simd.and(scratch, b1, scratch, c15, scratch, t0, SIMD_LANES);
503-
simd.shl(scratch, t0, 2, scratch, t0, SIMD_LANES);
504-
simd.shrLogical(scratch, b2, 6, scratch, t1, SIMD_LANES);
505-
simd.or(scratch, t0, scratch, t1, scratch, s2, SIMD_LANES);
506-
simd.and(scratch, b2, scratch, c63, scratch, s3, SIMD_LANES);
507-
508-
for (int lane = 0; lane < SIMD_LANES; lane++) {
509-
out[outIndex++] = map[scratch[s0 + lane]];
510-
out[outIndex++] = map[scratch[s1 + lane]];
511-
out[outIndex++] = map[scratch[s2 + lane]];
512-
out[outIndex++] = map[scratch[s3 + lane]];
513-
}
468+
int fastEnd = end - 12;
469+
for (; inIndex <= fastEnd; inIndex += 12) {
470+
int b0 = in[inIndex] & 0xff;
471+
int b1 = in[inIndex + 1] & 0xff;
472+
int b2 = in[inIndex + 2] & 0xff;
473+
int b3 = in[inIndex + 3] & 0xff;
474+
int b4 = in[inIndex + 4] & 0xff;
475+
int b5 = in[inIndex + 5] & 0xff;
476+
int b6 = in[inIndex + 6] & 0xff;
477+
int b7 = in[inIndex + 7] & 0xff;
478+
int b8 = in[inIndex + 8] & 0xff;
479+
int b9 = in[inIndex + 9] & 0xff;
480+
int b10 = in[inIndex + 10] & 0xff;
481+
int b11 = in[inIndex + 11] & 0xff;
482+
483+
out[outIndex++] = map[b0 >> 2];
484+
out[outIndex++] = map[((b0 & 0x03) << 4) | (b1 >> 4)];
485+
out[outIndex++] = map[((b1 & 0x0f) << 2) | (b2 >> 6)];
486+
out[outIndex++] = map[b2 & 0x3f];
487+
488+
out[outIndex++] = map[b3 >> 2];
489+
out[outIndex++] = map[((b3 & 0x03) << 4) | (b4 >> 4)];
490+
out[outIndex++] = map[((b4 & 0x0f) << 2) | (b5 >> 6)];
491+
out[outIndex++] = map[b5 & 0x3f];
492+
493+
out[outIndex++] = map[b6 >> 2];
494+
out[outIndex++] = map[((b6 & 0x03) << 4) | (b7 >> 4)];
495+
out[outIndex++] = map[((b7 & 0x0f) << 2) | (b8 >> 6)];
496+
out[outIndex++] = map[b8 & 0x3f];
497+
498+
out[outIndex++] = map[b9 >> 2];
499+
out[outIndex++] = map[((b9 & 0x03) << 4) | (b10 >> 4)];
500+
out[outIndex++] = map[((b10 & 0x0f) << 2) | (b11 >> 6)];
501+
out[outIndex++] = map[b11 & 0x3f];
514502
}
515503

516504
for (; inIndex < end; inIndex += 3) {
@@ -586,66 +574,12 @@ public static int decodeNoWhitespaceSimd(byte[] in, int inOffset, int inLength,
586574
}
587575

588576
requireScratch(scratch);
589-
Simd simd = Simd.get();
590-
requireSimdApiArrays(simd, in, out, scratch);
591-
592-
final int q0 = 0;
593-
final int q1 = q0 + SIMD_LANES;
594-
final int q2 = q1 + SIMD_LANES;
595-
final int q3 = q2 + SIMD_LANES;
596-
final int o0 = q3 + SIMD_LANES;
597-
final int o1 = o0 + SIMD_LANES;
598-
final int o2 = o1 + SIMD_LANES;
599-
final int t0 = o2 + SIMD_LANES;
600-
final int t1 = t0 + SIMD_LANES;
601-
final int c3 = t1 + SIMD_LANES;
602-
final int c15 = c3 + SIMD_LANES;
603-
604-
for (int lane = 0; lane < SIMD_LANES; lane++) {
605-
scratch[c3 + lane] = 3;
606-
scratch[c15 + lane] = 15;
607-
}
577+
requireSimdApiArrays(Simd.get(), in, out, scratch);
608578

609579
int fullLen = inLength - (pad > 0 ? 4 : 0);
610-
int simdFullLen = fullLen - (fullLen % 64);
580+
int fullEnd = inOffset + fullLen;
611581
int inIndex = inOffset;
612582
int outIndex = outOffset;
613-
int endVector = inOffset + simdFullLen;
614-
for (; inIndex < endVector; inIndex += 64) {
615-
for (int lane = 0; lane < SIMD_LANES; lane++) {
616-
int src = inIndex + lane * 4;
617-
int d0 = decodeMapInt[in[src] & 0xff];
618-
int d1 = decodeMapInt[in[src + 1] & 0xff];
619-
int d2 = decodeMapInt[in[src + 2] & 0xff];
620-
int d3 = decodeMapInt[in[src + 3] & 0xff];
621-
if ((d0 | d1 | d2 | d3) < 0) {
622-
return -1;
623-
}
624-
scratch[q0 + lane] = d0;
625-
scratch[q1 + lane] = d1;
626-
scratch[q2 + lane] = d2;
627-
scratch[q3 + lane] = d3;
628-
}
629-
630-
simd.shl(scratch, q0, 2, scratch, o0, SIMD_LANES);
631-
simd.shrLogical(scratch, q1, 4, scratch, t0, SIMD_LANES);
632-
simd.or(scratch, o0, scratch, t0, scratch, o0, SIMD_LANES);
633-
simd.and(scratch, q1, scratch, c15, scratch, t0, SIMD_LANES);
634-
simd.shl(scratch, t0, 4, scratch, t0, SIMD_LANES);
635-
simd.shrLogical(scratch, q2, 2, scratch, t1, SIMD_LANES);
636-
simd.or(scratch, t0, scratch, t1, scratch, o1, SIMD_LANES);
637-
simd.and(scratch, q2, scratch, c3, scratch, t0, SIMD_LANES);
638-
simd.shl(scratch, t0, 6, scratch, t0, SIMD_LANES);
639-
simd.or(scratch, t0, scratch, q3, scratch, o2, SIMD_LANES);
640-
641-
for (int lane = 0; lane < SIMD_LANES; lane++) {
642-
out[outIndex++] = (byte)scratch[o0 + lane];
643-
out[outIndex++] = (byte)scratch[o1 + lane];
644-
out[outIndex++] = (byte)scratch[o2 + lane];
645-
}
646-
}
647-
648-
int fullEnd = inOffset + fullLen;
649583
for (; inIndex < fullEnd; inIndex += 4) {
650584
int c0 = in[inIndex] & 0xff;
651585
int c1 = in[inIndex + 1] & 0xff;

‎scripts/hellocodenameone/common/src/main/java/com/codenameone/examples/hellocodenameone/tests/SimdApiTest.java‎

Lines changed: 10 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -9,8 +9,16 @@ public boolean runTest() {
99
try {
1010
Simd simd = Simd.get();
1111
if (!simd.isSupported()) {
12-
fail("Simd.isSupported() returned false");
13-
return false;
12+
int[] a = new int[]{1, 2, 3, 4};
13+
int[] b = new int[]{9, 8, 7, 6};
14+
int[] out = new int[4];
15+
simd.add(a, b, out, 0, 4);
16+
if (out[0] != 10 || out[1] != 10 || out[2] != 10 || out[3] != 10) {
17+
fail("Fallback SIMD API add failed on unsupported platform");
18+
return false;
19+
}
20+
done();
21+
return true;
1422
}
1523

1624
int[] a = simd.allocInt(16);

0 commit comments

Comments
 (0)