The shuf behavior is 3c5000 only

code/Makefile

@@ -1,6 +1,7 @@
 SRCS=$(wildcard *.cpp)
 HDRS=$(wildcard *.h)
 EXES=$(patsubst %.cpp,%,$(SRCS))
+CXX?=g++
 
 all: $(EXES)
 
@@ -11,4 +12,4 @@ clean:
 	rm -rf $(EXES)
 
 %: %.cpp $(HDRS)
-	g++ $< -mlsx -mlasx -o $@
+	$(CXX) $< -mlsx -mlasx -o $@

code/common.h

@@ -14,6 +14,10 @@ typedef uint32_t u32;
 typedef int64_t s64;
 typedef uint64_t u64;
 
+#ifndef MACHINE_3C5000
+#define MACHINE_3C5000 0
+#endif
+
 union v128 {
   __m128i m128i;
   __m128 m128;

code/vshuf_b.h

@@ -1,6 +1,6 @@
 for (int i = 0; i < 16; i++) {
-  if (c.byte[i] >= 64) {
-    // Caveat: observed in 3C5000, but not in QEMU
+  if (c.byte[i] >= 64 && MACHINE_3C5000) {
+    // Caveat: observed in 3C5000
     dst.byte[i] = 0;
   } else if ((c.byte[i] % 32) < 16) {
     dst.byte[i] = b.byte[c.byte[i] % 16];

code/vshuf_d.h

@@ -1,6 +1,6 @@
 for (int i = 0; i < 2; i++) {
-  if ((a.dword[i] % 256) >= 64) {
-    // Caveat: observed in 3C5000, but not in QEMU
+  if ((a.dword[i] % 256) >= 64 && MACHINE_3C5000) {
+    // Caveat: observed in 3C5000
     dst.dword[i] = 0;
   } else if ((a.dword[i] % 4) < 2) {
     dst.dword[i] = c.dword[a.dword[i] % 2];

code/vshuf_h.h

@@ -1,6 +1,6 @@
 for (int i = 0; i < 8; i++) {
-  if ((a.half[i] % 256) >= 64) {
-    // Caveat: observed in 3C5000, but not in QEMU
+  if ((a.half[i] % 256) >= 64 && MACHINE_3C5000) {
+    // Caveat: observed in 3C5000
     dst.half[i] = 0;
   } else if ((a.half[i] % 16) < 8) {
     dst.half[i] = c.half[a.half[i] % 8];

code/vshuf_w.h

@@ -1,6 +1,6 @@
 for (int i = 0; i < 4; i++) {
-  if ((a.word[i] % 256) >= 64) {
-    // Caveat: observed in 3C5000, but not in QEMU
+  if ((a.word[i] % 256) >= 64 && MACHINE_3C5000) {
+    // Caveat: observed in 3C5000
     dst.word[i] = 0;
   } else if ((a.word[i] % 8) < 4) {
     dst.word[i] = c.word[a.word[i] % 4];