target/i386: reimplement 0x0f 0xd8-0xdf, 0xe8-0xef, 0xf8-0xff, add AVX

These are more simple integer instructions present in both MMX and SSE/AVX,
with no holes that were later occupied by newer instructions.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

3 files changed, 63 insertions, 1 deletions

diff --git a/target/i386/tcg/decode-new.c.inc b/target/i386/tcg/decode-new.c.inc
index 8a749f3..a160a0b 100644
--- a/target/i386/tcg/decode-new.c.inc
+++ b/target/i386/tcg/decode-new.c.inc
@@ -273,6 +273,34 @@ static const X86OpEntry opcodes_0F[256] = {
     [0x6d] = X86_OP_ENTRY3(PUNPCKHQDQ, V,x, H,x, W,x,  vex4 p_66 avx2_256),
     [0x6e] = X86_OP_ENTRY3(MOVD_to,    V,x, None,None, E,y, vex5 mmx p_00_66),  /* wrong dest Vy on SDM! */
     [0x6f] = X86_OP_GROUP0(0F6F),
+
+    /* Incorrectly missing from 2-17 */
+    [0xd8] = X86_OP_ENTRY3(PSUBUSB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xd9] = X86_OP_ENTRY3(PSUBUSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xda] = X86_OP_ENTRY3(PMINUB,   V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xdb] = X86_OP_ENTRY3(PAND,     V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xdc] = X86_OP_ENTRY3(PADDUSB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xdd] = X86_OP_ENTRY3(PADDUSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xde] = X86_OP_ENTRY3(PMAXUB,   V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xdf] = X86_OP_ENTRY3(PANDN,    V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+
+    [0xe8] = X86_OP_ENTRY3(PSUBSB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xe9] = X86_OP_ENTRY3(PSUBSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xea] = X86_OP_ENTRY3(PMINSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xeb] = X86_OP_ENTRY3(POR,     V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xec] = X86_OP_ENTRY3(PADDSB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xed] = X86_OP_ENTRY3(PADDSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xee] = X86_OP_ENTRY3(PMAXSW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xef] = X86_OP_ENTRY3(PXOR,    V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+
+    [0xf8] = X86_OP_ENTRY3(PSUBB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xf9] = X86_OP_ENTRY3(PSUBW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xfa] = X86_OP_ENTRY3(PSUBD,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xfb] = X86_OP_ENTRY3(PSUBQ,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xfc] = X86_OP_ENTRY3(PADDB,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xfd] = X86_OP_ENTRY3(PADDW,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    [0xfe] = X86_OP_ENTRY3(PADDD,  V,x, H,x, W,x,  vex4 mmx avx2_256 p_00_66),
+    /* 0xff = UD0 */
 };
 
 static void do_decode_0F(DisasContext *s, CPUX86State *env, X86OpEntry *entry, uint8_t *b)
diff --git a/target/i386/tcg/emit.c.inc b/target/i386/tcg/emit.c.inc
index 8dbacc2..0d35a86 100644
--- a/target/i386/tcg/emit.c.inc
+++ b/target/i386/tcg/emit.c.inc
@@ -328,9 +328,31 @@ static void gen_##uname(DisasContext *s, CPUX86State *env, X86DecodedInsn *decod
          decode->op[2].offset, vec_len, vec_len);                                  \
 }
 
+BINARY_INT_GVEC(PADDB,   tcg_gen_gvec_add, MO_8)
+BINARY_INT_GVEC(PADDW,   tcg_gen_gvec_add, MO_16)
+BINARY_INT_GVEC(PADDD,   tcg_gen_gvec_add, MO_32)
+BINARY_INT_GVEC(PADDSB,  tcg_gen_gvec_ssadd, MO_8)
+BINARY_INT_GVEC(PADDSW,  tcg_gen_gvec_ssadd, MO_16)
+BINARY_INT_GVEC(PADDUSB, tcg_gen_gvec_usadd, MO_8)
+BINARY_INT_GVEC(PADDUSW, tcg_gen_gvec_usadd, MO_16)
+BINARY_INT_GVEC(PAND,    tcg_gen_gvec_and, MO_64)
 BINARY_INT_GVEC(PCMPGTB, tcg_gen_gvec_cmp, TCG_COND_GT, MO_8)
 BINARY_INT_GVEC(PCMPGTW, tcg_gen_gvec_cmp, TCG_COND_GT, MO_16)
 BINARY_INT_GVEC(PCMPGTD, tcg_gen_gvec_cmp, TCG_COND_GT, MO_32)
+BINARY_INT_GVEC(PMAXSW,  tcg_gen_gvec_smax, MO_16)
+BINARY_INT_GVEC(PMAXUB,  tcg_gen_gvec_umax, MO_8)
+BINARY_INT_GVEC(PMINSW,  tcg_gen_gvec_smin, MO_16)
+BINARY_INT_GVEC(PMINUB,  tcg_gen_gvec_umin, MO_8)
+BINARY_INT_GVEC(POR,     tcg_gen_gvec_or, MO_64)
+BINARY_INT_GVEC(PSUBB,   tcg_gen_gvec_sub, MO_8)
+BINARY_INT_GVEC(PSUBW,   tcg_gen_gvec_sub, MO_16)
+BINARY_INT_GVEC(PSUBD,   tcg_gen_gvec_sub, MO_32)
+BINARY_INT_GVEC(PSUBQ,   tcg_gen_gvec_sub, MO_64)
+BINARY_INT_GVEC(PSUBSB,  tcg_gen_gvec_sssub, MO_8)
+BINARY_INT_GVEC(PSUBSW,  tcg_gen_gvec_sssub, MO_16)
+BINARY_INT_GVEC(PSUBUSB, tcg_gen_gvec_ussub, MO_8)
+BINARY_INT_GVEC(PSUBUSW, tcg_gen_gvec_ussub, MO_16)
+BINARY_INT_GVEC(PXOR,    tcg_gen_gvec_xor, MO_64)
 
 
 /*
@@ -608,6 +630,16 @@ static void gen_MULX(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 
 }
 
+static void gen_PANDN(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
+{
+    int vec_len = vector_len(s, decode);
+
+    /* Careful, operand order is reversed!  */
+    tcg_gen_gvec_andc(MO_64,
+                      decode->op[0].offset, decode->op[2].offset,
+                      decode->op[1].offset, vec_len, vec_len);
+}
+
 static void gen_PDEP(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 {
     MemOp ot = decode->op[1].ot;
diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c
index 5133b8c..ff5c4bb 100644
--- a/target/i386/tcg/translate.c
+++ b/target/i386/tcg/translate.c
@@ -4781,7 +4781,9 @@ static bool disas_insn(DisasContext *s, CPUState *cpu)
 #ifndef CONFIG_USER_ONLY
         use_new &= b <= limit;
 #endif
-        if (use_new && (b >= 0x160 && b <= 0x16f)) {
+        if (use_new &&
+            ((b >= 0x160 && b <= 0x16f) ||
+             (b >= 0x1d8 && b <= 0x1ff && (b & 8)))) {
             disas_insn_new(s, cpu, b + 0x100);
             return s->pc;
         }