target/ppc, tcg, tcg/i386: Optimize emulation of some Altivec instructions

[Qemu-devel] [PATCH v4 11/13] tcg: Add opcodes for verctor vmrgl instructions
Posted by Stefan Brankovic 6 years, 7 months ago
Signed-off-by: Stefan Brankovic <stefan.brankovic@rt-rk.com>
---
 accel/tcg/tcg-runtime-gvec.c | 42 ++++++++++++++++++++++++++++++++++++++++++
 accel/tcg/tcg-runtime.h      |  4 ++++
 tcg/i386/tcg-target.h        |  1 +
 tcg/tcg-op-gvec.c            | 24 ++++++++++++++++++++++++
 tcg/tcg-op-gvec.h            |  2 ++
 tcg/tcg-op-vec.c             |  5 +++++
 tcg/tcg-op.h                 |  1 +
 tcg/tcg-opc.h                |  1 +
 tcg/tcg.c                    |  2 ++
 tcg/tcg.h                    |  1 +
 10 files changed, 83 insertions(+)

diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c
index 28173ae..152f277 100644
--- a/accel/tcg/tcg-runtime-gvec.c
+++ b/accel/tcg/tcg-runtime-gvec.c
@@ -1500,3 +1500,45 @@ void HELPER(gvec_vmrgh32)(void *d, void *a, void *b, uint32_t desc)
     }
     clear_high(d, oprsz, desc);
 }
+
+void HELPER(gvec_vmrgl8)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < (oprsz / 2); i += sizeof(uint8_t)) {
+        uint8_t aa = *(uint8_t *)(a + i);
+        uint8_t bb = *(uint8_t *)(b + i);
+        *(uint8_t *)(d + 2 * i) = bb;
+        *(uint8_t *)(d + 2 * i + sizeof(uint8_t)) = aa;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_vmrgl16)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < (oprsz / 2); i += sizeof(uint16_t)) {
+        uint16_t aa = *(uint16_t *)(a + i);
+        uint16_t bb = *(uint16_t *)(b + i);
+        *(uint16_t *)(d + 2 * i) = bb;
+        *(uint16_t *)(d + 2 * i + sizeof(uint16_t)) = aa;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_vmrgl32)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
+        uint32_t aa = *(uint32_t *)(a + i);
+        uint32_t bb = *(uint32_t *)(b + i);
+        *(uint32_t *)(d + 2 * i) = bb;
+        *(uint32_t *)(d + 2 * i + sizeof(uint32_t)) = aa;
+    }
+    clear_high(d, oprsz, desc);
+}
diff --git a/accel/tcg/tcg-runtime.h b/accel/tcg/tcg-runtime.h
index 089956f..fd0ba1e 100644
--- a/accel/tcg/tcg-runtime.h
+++ b/accel/tcg/tcg-runtime.h
@@ -309,3 +309,7 @@ DEF_HELPER_FLAGS_5(gvec_bitsel, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(gvec_vmrgh8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(gvec_vmrgh16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(gvec_vmrgh32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_vmrgl8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_vmrgl16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_vmrgl32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index daae35f..e825324 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -193,6 +193,7 @@ extern bool have_avx2;
 #define TCG_TARGET_HAS_bitsel_vec       0
 #define TCG_TARGET_HAS_cmpsel_vec       -1
 #define TCG_TARGET_HAS_vmrgh_vec        1
+#define TCG_TARGET_HAS_vmrgl_vec        0
 
 #define TCG_TARGET_deposit_i32_valid(ofs, len) \
     (((ofs) == 0 && (len) == 8) || ((ofs) == 8 && (len) == 8) || \
diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index 2560fb6..da1d272 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -2125,6 +2125,30 @@ void tcg_gen_gvec_vmrgh(unsigned vece, uint32_t dofs, uint32_t aofs,
     tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
 }
 
+static const TCGOpcode vecop_list_vmrgl[] = { INDEX_op_vmrgl_vec, 0 };
+
+void tcg_gen_gvec_vmrgl(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const GVecGen3 g[3] = {
+        { .fniv = tcg_gen_vmrgl_vec,
+          .fno = gen_helper_gvec_vmrgl8,
+          .opt_opc = vecop_list_vmrgl,
+          .vece = MO_8 },
+        { .fniv = tcg_gen_vmrgl_vec,
+          .fno = gen_helper_gvec_vmrgl16,
+          .opt_opc = vecop_list_vmrgl,
+          .vece = MO_16 },
+        {
+          .fniv = tcg_gen_vmrgl_vec,
+          .fno = gen_helper_gvec_vmrgl32,
+          .opt_opc = vecop_list_vmrgl,
+          .vece = MO_32 }
+    };
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
 /* Perform a vector negation using normal negation and a mask.
    Compare gen_subv_mask above.  */
 static void gen_negv_mask(TCGv_i64 d, TCGv_i64 b, TCGv_i64 m)
diff --git a/tcg/tcg-op-gvec.h b/tcg/tcg-op-gvec.h
index 8c04d71..a2eb45c 100644
--- a/tcg/tcg-op-gvec.h
+++ b/tcg/tcg-op-gvec.h
@@ -275,6 +275,8 @@ void tcg_gen_gvec_umax(unsigned vece, uint32_t dofs, uint32_t aofs,
 /* Vector merge. */
 void tcg_gen_gvec_vmrgh(unsigned vece, uint32_t dofs, uint32_t aofs,
                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_vmrgl(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
 
 void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs,
                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
diff --git a/tcg/tcg-op-vec.c b/tcg/tcg-op-vec.c
index fb0b83e..ab22335 100644
--- a/tcg/tcg-op-vec.c
+++ b/tcg/tcg-op-vec.c
@@ -668,6 +668,11 @@ void tcg_gen_vmrgh_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
     do_op3(vece, r, a, b, INDEX_op_vmrgh_vec);
 }
 
+void tcg_gen_vmrgl_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
+{
+    do_op3(vece, r, a, b, INDEX_op_vmrgl_vec);
+}
+
 void tcg_gen_shlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
 {
     do_op3_nofail(vece, r, a, b, INDEX_op_shlv_vec);
diff --git a/tcg/tcg-op.h b/tcg/tcg-op.h
index d8de022..c101170 100644
--- a/tcg/tcg-op.h
+++ b/tcg/tcg-op.h
@@ -986,6 +986,7 @@ void tcg_gen_smax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
 void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
 
 void tcg_gen_vmrgh_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
+void tcg_gen_vmrgl_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
 
 void tcg_gen_shli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i);
 void tcg_gen_shri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i);
diff --git a/tcg/tcg-opc.h b/tcg/tcg-opc.h
index 2bc3bdf..d99131a 100644
--- a/tcg/tcg-opc.h
+++ b/tcg/tcg-opc.h
@@ -236,6 +236,7 @@ DEF(smax_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
 DEF(umax_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
 
 DEF(vmrgh_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_vmrgh_vec))
+DEF(vmrgl_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_vmrgl_vec))
 
 DEF(and_vec, 1, 2, 0, IMPLVEC)
 DEF(or_vec, 1, 2, 0, IMPLVEC)
diff --git a/tcg/tcg.c b/tcg/tcg.c
index fed9a6f..01245d5 100644
--- a/tcg/tcg.c
+++ b/tcg/tcg.c
@@ -1648,6 +1648,8 @@ bool tcg_op_supported(TCGOpcode op)
         return have_vec && TCG_TARGET_HAS_minmax_vec;
     case INDEX_op_vmrgh_vec:
         return have_vec && TCG_TARGET_HAS_vmrgh_vec;
+    case INDEX_op_vmrgl_vec:
+        return have_vec && TCG_TARGET_HAS_vmrgl_vec;
     case INDEX_op_bitsel_vec:
         return have_vec && TCG_TARGET_HAS_bitsel_vec;
     case INDEX_op_cmpsel_vec:
diff --git a/tcg/tcg.h b/tcg/tcg.h
index 05b9b51..6f9f333 100644
--- a/tcg/tcg.h
+++ b/tcg/tcg.h
@@ -187,6 +187,7 @@ typedef uint64_t TCGRegSet;
 #define TCG_TARGET_HAS_sat_vec          0
 #define TCG_TARGET_HAS_minmax_vec       0
 #define TCG_TARGET_HAS_vmrgh_vec        0
+#define TCG_TARGET_HAS_vmrgl_vec        0
 #define TCG_TARGET_HAS_bitsel_vec       0
 #define TCG_TARGET_HAS_cmpsel_vec       0
 #else
-- 
2.7.4