[Qemu-devel] [RFC PATCH v1 15/22] target/i386: reimplement (V)PADDUS(B, W)

[Jan Bobek]

Use the gvec infrastructure to achieve the desired functionality.

Signed-off-by: Jan Bobek <address@hidden>
---
 target/i386/ops_sse.h        |  5 -----
 target/i386/ops_sse_header.h |  2 --
 target/i386/translate.c      | 19 +++++++++++++++++--
 3 files changed, 17 insertions(+), 9 deletions(-)

diff --git a/target/i386/ops_sse.h b/target/i386/ops_sse.h
index 8829dcb781..8c9b47fca4 100644
--- a/target/i386/ops_sse.h
+++ b/target/i386/ops_sse.h
@@ -337,9 +337,6 @@ static inline int satsw(int x)
     }
 }
 
-#define FADDUB(a, b) satub((a) + (b))
-#define FADDUW(a, b) satuw((a) + (b))
-
 #define FSUBUB(a, b) satub((a) - (b))
 #define FSUBUW(a, b) satuw((a) - (b))
 #define FSUBSB(a, b) satsb((int8_t)(a) - (int8_t)(b))
@@ -362,11 +359,9 @@ static inline int satsw(int x)
 #define FAVG(a, b) (((a) + (b) + 1) >> 1)
 #endif
 
-SSE_HELPER_B(helper_paddusb, FADDUB)
 SSE_HELPER_B(helper_psubusb, FSUBUB)
 SSE_HELPER_B(helper_psubsb, FSUBSB)
 
-SSE_HELPER_W(helper_paddusw, FADDUW)
 SSE_HELPER_W(helper_psubusw, FSUBUW)
 SSE_HELPER_W(helper_psubsw, FSUBSW)
 
diff --git a/target/i386/ops_sse_header.h b/target/i386/ops_sse_header.h
index 78203e80a5..8a31ade70c 100644
--- a/target/i386/ops_sse_header.h
+++ b/target/i386/ops_sse_header.h
@@ -60,11 +60,9 @@ DEF_HELPER_3(glue(pslldq, SUFFIX), void, env, Reg, Reg)
 #define SSE_HELPER_Q(name, F)\
     DEF_HELPER_3(glue(name, SUFFIX), void, env, Reg, Reg)
 
-SSE_HELPER_B(paddusb, FADDUB)
 SSE_HELPER_B(psubusb, FSUBUB)
 SSE_HELPER_B(psubsb, FSUBSB)
 
-SSE_HELPER_W(paddusw, FADDUW)
 SSE_HELPER_W(psubusw, FSUBUW)
 SSE_HELPER_W(psubsw, FSUBSW)
 
diff --git a/target/i386/translate.c b/target/i386/translate.c
index 5ea5014d99..e2ed8c20b3 100644
--- a/target/i386/translate.c
+++ b/target/i386/translate.c
@@ -2826,8 +2826,8 @@ static const SSEFunc_0_epp sse_op_table1[256][4] = {
     [0xd9] = MMX_OP2(psubusw),
     [0xda] = MMX_OP2(pminub),
     [0xdb] = { SSE_TOMBSTONE, SSE_TOMBSTONE },
-    [0xdc] = MMX_OP2(paddusb),
-    [0xdd] = MMX_OP2(paddusw),
+    [0xdc] = { SSE_TOMBSTONE, SSE_TOMBSTONE },
+    [0xdd] = { SSE_TOMBSTONE, SSE_TOMBSTONE },
     [0xde] = MMX_OP2(pmaxub),
     [0xdf] = { SSE_TOMBSTONE, SSE_TOMBSTONE },
     [0xe0] = MMX_OP2(pavgb),
@@ -3176,6 +3176,11 @@ static inline void gen_gvec_ld_modrm_3(CPUX86State *env, 
DisasContext *s,
 #define gen_vpadds_xmm(env, s, modrm, vece) gen_gvec_ld_modrm_vxmm((env), (s), 
(modrm), (vece), tcg_gen_gvec_ssadd, 0123)
 #define gen_vpadds_ymm(env, s, modrm, vece) gen_gvec_ld_modrm_vymm((env), (s), 
(modrm), (vece), tcg_gen_gvec_ssadd, 0123)
 
+#define gen_paddus_mm(env, s, modrm, vece)   gen_gvec_ld_modrm_mm  ((env), 
(s), (modrm), (vece), tcg_gen_gvec_usadd, 0112)
+#define gen_paddus_xmm(env, s, modrm, vece)  gen_gvec_ld_modrm_xmm ((env), 
(s), (modrm), (vece), tcg_gen_gvec_usadd, 0112)
+#define gen_vpaddus_xmm(env, s, modrm, vece) gen_gvec_ld_modrm_vxmm((env), 
(s), (modrm), (vece), tcg_gen_gvec_usadd, 0123)
+#define gen_vpaddus_ymm(env, s, modrm, vece) gen_gvec_ld_modrm_vymm((env), 
(s), (modrm), (vece), tcg_gen_gvec_usadd, 0123)
+
 #define gen_psub_mm(env, s, modrm, vece)   gen_gvec_ld_modrm_mm  ((env), (s), 
(modrm), (vece), tcg_gen_gvec_sub, 0112)
 #define gen_psub_xmm(env, s, modrm, vece)  gen_gvec_ld_modrm_xmm ((env), (s), 
(modrm), (vece), tcg_gen_gvec_sub, 0112)
 #define gen_vpsub_xmm(env, s, modrm, vece) gen_gvec_ld_modrm_vxmm((env), (s), 
(modrm), (vece), tcg_gen_gvec_sub, 0123)
@@ -3336,6 +3341,16 @@ static void gen_sse(CPUX86State *env, DisasContext *s, 
int b)
     case 0xd4 | M_0F | P_66 | VEX_128: gen_vpadd_xmm(env, s, modrm, MO_64); 
return;
     case 0xd4 | M_0F | P_66 | VEX_256: gen_vpadd_ymm(env, s, modrm, MO_64); 
return;
 
+    case 0xdc | M_0F:                  gen_paddus_mm(env, s, modrm, MO_8); 
return;
+    case 0xdc | M_0F | P_66:           gen_paddus_xmm(env, s, modrm, MO_8); 
return;
+    case 0xdc | M_0F | P_66 | VEX_128: gen_vpaddus_xmm(env, s, modrm, MO_8); 
return;
+    case 0xdc | M_0F | P_66 | VEX_256: gen_vpaddus_ymm(env, s, modrm, MO_8); 
return;
+
+    case 0xdd | M_0F:                  gen_paddus_mm(env, s, modrm, MO_16); 
return;
+    case 0xdd | M_0F | P_66:           gen_paddus_xmm(env, s, modrm, MO_16); 
return;
+    case 0xdd | M_0F | P_66 | VEX_128: gen_vpaddus_xmm(env, s, modrm, MO_16); 
return;
+    case 0xdd | M_0F | P_66 | VEX_256: gen_vpaddus_ymm(env, s, modrm, MO_16); 
return;
+
     case 0xec | M_0F:                  gen_padds_mm(env, s, modrm, MO_8); 
return;
     case 0xec | M_0F | P_66:           gen_padds_xmm(env, s, modrm, MO_8); 
return;
     case 0xec | M_0F | P_66 | VEX_128: gen_vpadds_xmm(env, s, modrm, MO_8); 
return;
-- 
2.20.1