Add the vec4 %subi instruction

tgt-vvp/eval_vec4.c

@@ -138,6 +138,9 @@ static void draw_binary_vec4_arith(ivl_expr_t expr)
 		case '+':
 		  draw_immediate_vec4(re, "%addi");
 		  return;
+		case '-':
+		  draw_immediate_vec4(re, "%subi");
+		  return;
 		case '*':
 		  draw_immediate_vec4(re, "%muli");
 		  return;

vvp/codes.h

@@ -226,6 +226,7 @@ extern bool of_STORE_STRA(vthread_t thr, vvp_code_t code);
 extern bool of_STORE_VEC4(vthread_t thr, vvp_code_t code);
 extern bool of_STORE_VEC4A(vthread_t thr, vvp_code_t code);
 extern bool of_SUB(vthread_t thr, vvp_code_t code);
+extern bool of_SUBI(vthread_t thr, vvp_code_t code);
 extern bool of_SUB_WR(vthread_t thr, vvp_code_t code);
 extern bool of_SUBSTR(vthread_t thr, vvp_code_t code);
 extern bool of_SUBSTR_VEC4(vthread_t thr, vvp_code_t code);

vvp/compile.cc

@@ -274,6 +274,7 @@ static const struct opcode_table_s opcode_table[] = {
       { "%store/vec4a",   of_STORE_VEC4A,   3, {OA_ARR_PTR, OA_BIT1, OA_BIT2} },
       { "%sub",    of_SUB,    0,  {OA_NONE,     OA_NONE,     OA_NONE} },
       { "%sub/wr", of_SUB_WR, 0,  {OA_NONE,     OA_NONE,     OA_NONE} },
+      { "%subi",   of_SUBI,   3,  {OA_BIT1,     OA_BIT2,     OA_NUMBER} },
       { "%substr",     of_SUBSTR,     2,{OA_BIT1,    OA_BIT2, OA_NONE} },
       { "%substr/vec4",of_SUBSTR_VEC4,2,{OA_BIT1,    OA_BIT2, OA_NONE} },
       { "%test_nul",     of_TEST_NUL,     1,{OA_FUNC_PTR,OA_NONE,    OA_NONE} },

vvp/vthread.cc

@@ -5464,6 +5464,30 @@ bool of_SUB(vthread_t thr, vvp_code_t)
       return true;
 }
 
+/*
+ * %subi <vala>, <valb>, <wid>
+ *
+ * Pop1 operand, get the other operand from the arguments, and push
+ * the result.
+ */
+bool of_SUBI(vthread_t thr, vvp_code_t cp)
+{
+      unsigned wid = cp->number;
+
+      vvp_vector4_t&l = thr->peek_vec4();
+
+	// I expect that most of the bits of an immediate value are
+	// going to be zero, so start the result vector with all zero
+	// bits. Then we only need to replace the bits that are different.
+      vvp_vector4_t r (wid, BIT4_0);
+      get_immediate_rval (cp, r);
+
+      l.sub(r);
+
+      return true;
+
+}
+
 bool of_SUB_WR(vthread_t thr, vvp_code_t)
 {
       double r = thr->pop_real();