ggml-cpu: fix wrong displacement

taronaeo · taronaeo · commit 6b4469b882d3 · 2025-06-20T22:46:49.000+08:00
Signed-off-by: Aaron Teo &lt;aaron.teo1@ibm.com&gt;
diff --git a/ggml/src/ggml-cpu/ggml-cpu.c b/ggml/src/ggml-cpu/ggml-cpu.c
@@ -3143,7 +3143,7 @@ void ggml_cpu_fp32_to_fp16(const float * x, ggml_fp16_t * y, int64_t n) {
         float32x4_t v_x1 = vec_xl(i + 0, x);
         float32x4_t v_x2 = vec_xl(i + 4, x);
         uint16x8_t v_dlf16 = vec_round_from_fp32(v_x1, v_x2, 0);
-        vec_xst(v_dlf16, i, (uint16_t *)y);
+        vec_xst(v_dlf16, 0, (uint16_t *)(y + i));
     }
     // TODO: Enable bottom code once checks are done
     // for (; i + 3 < n; i += 4) {

Original file line number	Diff line number	Diff line change
`@@ -3143,7 +3143,7 @@ void ggml_cpu_fp32_to_fp16(const float * x, ggml_fp16_t * y, int64_t n) {`
`3143`	`3143`	`float32x4_t v_x1 = vec_xl(i + 0, x);`
`3144`	`3144`	`float32x4_t v_x2 = vec_xl(i + 4, x);`
`3145`	`3145`	`uint16x8_t v_dlf16 = vec_round_from_fp32(v_x1, v_x2, 0);`
`3146`		`- vec_xst(v_dlf16, i, (uint16_t *)y);`
	`3146`	`+ vec_xst(v_dlf16, 0, (uint16_t *)(y + i));`
`3147`	`3147`	`}`
`3148`	`3148`	`// TODO: Enable bottom code once checks are done`
`3149`	`3149`	`// for (; i + 3 < n; i += 4) {`