diff --git a/src/gallium/drivers/radeonsi/si_descriptors.c b/src/gallium/drivers/radeonsi/si_descriptors.c
index 0e95f48..a4a1875 100644
--- a/src/gallium/drivers/radeonsi/si_descriptors.c
+++ b/src/gallium/drivers/radeonsi/si_descriptors.c
@@ -244,7 +244,8 @@ static void si_emit_descriptors(struct si_context *sctx,
 			radeon_emit(cs, (va >> 32UL) & 0xFFFFFFFFUL);
 		}
 
-		radeon_emit_array(cs, descriptors[i], desc->element_dw_size);
+		radeon_emit_array_cpu_to_le32(cs, descriptors[i],
+					      desc->element_dw_size);
 
 		last_index = i;
 	}
@@ -611,18 +612,18 @@ void si_update_vertex_buffers(struct si_context *sctx)
 		va = rbuffer->gpu_address + offset;
 
 		/* Fill in T# buffer resource description */
-		desc[0] = va & 0xFFFFFFFF;
-		desc[1] = S_008F04_BASE_ADDRESS_HI(va >> 32) |
-			  S_008F04_STRIDE(vb->stride);
+		desc[0] = util_cpu_to_le32(va);
+		desc[1] = util_cpu_to_le32(S_008F04_BASE_ADDRESS_HI(va >> 32) |
+				      S_008F04_STRIDE(vb->stride));
 		if (vb->stride)
 			/* Round up by rounding down and adding 1 */
-			desc[2] = (vb->buffer->width0 - offset -
-				   sctx->vertex_elements->format_size[i]) /
-				  vb->stride + 1;
+			desc[2] = util_cpu_to_le32((vb->buffer->width0 - offset -
+						    sctx->vertex_elements->format_size[i]) /
+						   vb->stride + 1);
 		else
-			desc[2] = vb->buffer->width0 - offset;
+			desc[2] = util_cpu_to_le32(vb->buffer->width0 - offset);
 
-		desc[3] = sctx->vertex_elements->rsrc_word3[i];
+		desc[3] = util_cpu_to_le32(sctx->vertex_elements->rsrc_word3[i]);
 
 		if (!bound[ve->vertex_buffer_index]) {
 			r600_context_bo_reloc(&sctx->b, &sctx->b.rings.gfx,
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index dbd58f1..797630d 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -586,4 +586,12 @@ static INLINE void radeon_emit_array(struct radeon_winsys_cs *cs,
     cs->cdw += count;
 }
 
+static INLINE void radeon_emit_array_cpu_to_le32(struct radeon_winsys_cs *cs,
+                                                 const uint32_t *values,
+                                                 unsigned count)
+{
+    util_memcpy_cpu_to_le32(cs->buf+cs->cdw, values, count * 4);
+    cs->cdw += count;
+}
+
 #endif