vertexcodec: Fix compilation on gcc/ARM for NEON code

zeux · zeux · commit 4d53b1999332 · 2025-01-16T06:46:52.000-08:00
gcc uses strongly typed NEON intrinsics and vadd_u16 call was missing vreinteprret_u16_u8. Fixes #833.
diff --git a/src/vertexcodec.cpp b/src/vertexcodec.cpp
@@ -1365,7 +1365,7 @@ SIMD_TARGET inline uint8x8_t rebase(uint8x8_t npi, uint8x16_t r0, uint8x16_t r1,
 	{
 		uint16x8_t rsum = vaddq_u16(vaddq_u16(vreinterpretq_u16_u8(r0), vreinterpretq_u16_u8(r1)), vaddq_u16(vreinterpretq_u16_u8(r2), vreinterpretq_u16_u8(r3)));
 		uint16x4_t rsumx = vadd_u16(vget_low_u16(rsum), vget_high_u16(rsum));
-		return vreinterpret_u8_u16(vadd_u16(vadd_u16(npi, rsumx), vext_u16(rsumx, rsumx, 2)));
+		return vreinterpret_u8_u16(vadd_u16(vadd_u16(vreinterpret_u16_u8(npi), rsumx), vext_u16(rsumx, rsumx, 2)));
 	}
 	case 2:
 	{

Original file line number	Diff line number	Diff line change
`@@ -1365,7 +1365,7 @@ SIMD_TARGET inline uint8x8_t rebase(uint8x8_t npi, uint8x16_t r0, uint8x16_t r1,`
`1365`	`1365`	`{`
`1366`	`1366`	`uint16x8_t rsum = vaddq_u16(vaddq_u16(vreinterpretq_u16_u8(r0), vreinterpretq_u16_u8(r1)), vaddq_u16(vreinterpretq_u16_u8(r2), vreinterpretq_u16_u8(r3)));`
`1367`	`1367`	`uint16x4_t rsumx = vadd_u16(vget_low_u16(rsum), vget_high_u16(rsum));`
`1368`		`- return vreinterpret_u8_u16(vadd_u16(vadd_u16(npi, rsumx), vext_u16(rsumx, rsumx, 2)));`
	`1368`	`+ return vreinterpret_u8_u16(vadd_u16(vadd_u16(vreinterpret_u16_u8(npi), rsumx), vext_u16(rsumx, rsumx, 2)));`
`1369`	`1369`	`}`
`1370`	`1370`	`case 2:`
`1371`	`1371`	`{`