Fix vftint_w_d code, reported by Yang Liu

jiegec · Dec 26, 2024 · 1df495b · 1df495b
1 parent d7b542f
commit 1df495b
Show file tree

Hide file tree

Showing 7 changed files with 45 additions and 17 deletions.
diff --git a/code/common.h b/code/common.h
@@ -185,7 +185,8 @@ union v256 {
 void test();
 
 void print(const char *s, v128 num) {
-  printf("v128 as __m128i %s: 0x%016lx 0x%016lx\n", s, num.dword[0], num.dword[1]);
+  printf("v128 as __m128i %s: 0x%016lx 0x%016lx\n", s, num.dword[0],
+         num.dword[1]);
   printf("v128 as __m128 %s: %f %f %f %f\n", s, num.fp32[0], num.fp32[1],
          num.fp32[2], num.fp32[3]);
   printf("v128 as __m128d %s: %lf %lf\n", s, num.fp64[0], num.fp64[1]);
@@ -204,8 +205,8 @@ void print(const char *s, __m128d num) {
 }
 
 void print(const char *s, v256 num) {
-  printf("v256 as __m256i %s: 0x%016lx 0x%016lx 0x%016lx 0x%016lx\n", s, num.dword[0],
-         num.dword[1], num.dword[2], num.dword[3]);
+  printf("v256 as __m256i %s: 0x%016lx 0x%016lx 0x%016lx 0x%016lx\n", s,
+         num.dword[0], num.dword[1], num.dword[2], num.dword[3]);
   printf("v256 as __m256 %s: %f %f %f %f %f %f %f %f\n", s, num.fp32[0],
          num.fp32[1], num.fp32[2], num.fp32[3], num.fp32[4], num.fp32[5],
          num.fp32[6], num.fp32[7]);
@@ -214,8 +215,8 @@ void print(const char *s, v256 num) {
 }
 
 void print(const char *s, __m256i num) {
-  printf("__m256i %s: 0x%016llx 0x%016llx 0x%016llx 0x%016llx\n", s, num[0], num[1],
-         num[2], num[3]);
+  printf("__m256i %s: 0x%016llx 0x%016llx 0x%016llx 0x%016llx\n", s, num[0],
+         num[1], num[2], num[3]);
 }
 
 void print(const char *s, __m256 num) {

diff --git a/code/vftint_w_d.cpp b/code/vftint_w_d.cpp
@@ -0,0 +1,19 @@
+#include "common.h"
+
+v128 vftint_w_d(v128 a, v128 b) {
+  v128 dst;
+#include "vftint_w_d.h"
+  return dst;
+}
+
+void test() {
+  {
+    __m128d a = {1.0, 2.0};
+    __m128d b = {3.0, 4.0};
+    PRINT(a);
+    PRINT(b);
+    PRINT(__lsx_vftint_w_d(a, b));
+    PRINT(vftint_w_d(a, b));
+    assert(vftint_w_d(a, b) == __lsx_vftint_w_d(a, b));
+  }
+}
diff --git a/code/vftint_w_d.h b/code/vftint_w_d.h
@@ -1,5 +1,5 @@
-for (int i = 0; i < 2; i++) {
-  dst.dword[i] = (i < 1)
-                     ? (s64)a.fp64[i]
-                     : (s64)b.fp64[i]; // rounding mode is not expressed in C
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = (i < 2)
+                    ? (s64)b.fp64[i]
+                    : (s64)a.fp64[i - 2]; // rounding mode is not expressed in C
 }
diff --git a/code/xvsetanyeqz_b.cpp b/code/xvsetanyeqz_b.cpp
@@ -19,11 +19,13 @@ void test() {
   assert(__lasx_xbz_b(a));
   assert(__lasx_xbz_b(a) == xbz_b(a));
 
-  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(!__lasx_xbz_b(a));
   assert(__lasx_xbz_b(a) == xbz_b(a));
 
-  a = __m256i{0x1111111111111100, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111111100, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(__lasx_xbz_b(a));
   assert(__lasx_xbz_b(a) == xbz_b(a));
 }
diff --git a/code/xvsetanyeqz_d.cpp b/code/xvsetanyeqz_d.cpp
@@ -19,11 +19,13 @@ void test() {
   assert(__lasx_xbz_d(a));
   assert(__lasx_xbz_d(a) == xbz_d(a));
 
-  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(!__lasx_xbz_d(a));
   assert(__lasx_xbz_d(a) == xbz_d(a));
 
-  a = __m256i{0x0000000000000000, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x0000000000000000, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(__lasx_xbz_d(a));
   assert(__lasx_xbz_d(a) == xbz_d(a));
 }
diff --git a/code/xvsetanyeqz_h.cpp b/code/xvsetanyeqz_h.cpp
@@ -19,11 +19,13 @@ void test() {
   assert(__lasx_xbz_h(a));
   assert(__lasx_xbz_h(a) == xbz_h(a));
 
-  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(!__lasx_xbz_h(a));
   assert(__lasx_xbz_h(a) == xbz_h(a));
 
-  a = __m256i{0x1111111111110000, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111110000, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(__lasx_xbz_h(a));
   assert(__lasx_xbz_h(a) == xbz_h(a));
 }
diff --git a/code/xvsetanyeqz_w.cpp b/code/xvsetanyeqz_w.cpp
@@ -19,11 +19,13 @@ void test() {
   assert(__lasx_xbz_w(a));
   assert(__lasx_xbz_w(a) == xbz_w(a));
 
-  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111111111111, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(!__lasx_xbz_w(a));
   assert(__lasx_xbz_w(a) == xbz_w(a));
 
-  a = __m256i{0x1111111100000000, 0x1111111111111111, 0x1111111111111111, 0x1111111111111111};
+  a = __m256i{0x1111111100000000, 0x1111111111111111, 0x1111111111111111,
+              0x1111111111111111};
   assert(__lasx_xbz_w(a));
   assert(__lasx_xbz_w(a) == xbz_w(a));
 }