tcg: Add gvec expanders for variable shift

The gvec expanders perform a modulo on the shift count.  If the target
requires alternate behaviour, then it cannot use the generic gvec
expanders anyway, and will have to have its own custom code.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

4 files changed, 225 insertions, 0 deletions

diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index 35ebc5a201..061ef329f1 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -2555,6 +2555,201 @@ void tcg_gen_gvec_sari(unsigned vece, uint32_t dofs, uint32_t aofs,
     }
 }
 
+/*
+ * Expand D = A << (B % element bits)
+ *
+ * Unlike scalar shifts, where it is easy for the target front end
+ * to include the modulo as part of the expansion.  If the target
+ * naturally includes the modulo as part of the operation, great!
+ * If the target has some other behaviour from out-of-range shifts,
+ * then it could not use this function anyway, and would need to
+ * do it's own expansion with custom functions.
+ */
+static void tcg_gen_shlv_mod_vec(unsigned vece, TCGv_vec d,
+                                 TCGv_vec a, TCGv_vec b)
+{
+    TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+    tcg_gen_dupi_vec(vece, t, (8 << vece) - 1);
+    tcg_gen_and_vec(vece, t, t, b);
+    tcg_gen_shlv_vec(vece, d, a, t);
+    tcg_temp_free_vec(t);
+}
+
+static void tcg_gen_shl_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 t = tcg_temp_new_i32();
+
+    tcg_gen_andi_i32(t, b, 31);
+    tcg_gen_shl_i32(d, a, t);
+    tcg_temp_free_i32(t);
+}
+
+static void tcg_gen_shl_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 t = tcg_temp_new_i64();
+
+    tcg_gen_andi_i64(t, b, 63);
+    tcg_gen_shl_i64(d, a, t);
+    tcg_temp_free_i64(t);
+}
+
+void tcg_gen_gvec_shlv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const TCGOpcode vecop_list[] = { INDEX_op_shlv_vec, 0 };
+    static const GVecGen3 g[4] = {
+        { .fniv = tcg_gen_shlv_mod_vec,
+          .fno = gen_helper_gvec_shl8v,
+          .opt_opc = vecop_list,
+          .vece = MO_8 },
+        { .fniv = tcg_gen_shlv_mod_vec,
+          .fno = gen_helper_gvec_shl16v,
+          .opt_opc = vecop_list,
+          .vece = MO_16 },
+        { .fni4 = tcg_gen_shl_mod_i32,
+          .fniv = tcg_gen_shlv_mod_vec,
+          .fno = gen_helper_gvec_shl32v,
+          .opt_opc = vecop_list,
+          .vece = MO_32 },
+        { .fni8 = tcg_gen_shl_mod_i64,
+          .fniv = tcg_gen_shlv_mod_vec,
+          .fno = gen_helper_gvec_shl64v,
+          .opt_opc = vecop_list,
+          .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+          .vece = MO_64 },
+    };
+
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
+/*
+ * Similarly for logical right shifts.
+ */
+
+static void tcg_gen_shrv_mod_vec(unsigned vece, TCGv_vec d,
+                                 TCGv_vec a, TCGv_vec b)
+{
+    TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+    tcg_gen_dupi_vec(vece, t, (8 << vece) - 1);
+    tcg_gen_and_vec(vece, t, t, b);
+    tcg_gen_shrv_vec(vece, d, a, t);
+    tcg_temp_free_vec(t);
+}
+
+static void tcg_gen_shr_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 t = tcg_temp_new_i32();
+
+    tcg_gen_andi_i32(t, b, 31);
+    tcg_gen_shr_i32(d, a, t);
+    tcg_temp_free_i32(t);
+}
+
+static void tcg_gen_shr_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 t = tcg_temp_new_i64();
+
+    tcg_gen_andi_i64(t, b, 63);
+    tcg_gen_shr_i64(d, a, t);
+    tcg_temp_free_i64(t);
+}
+
+void tcg_gen_gvec_shrv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const TCGOpcode vecop_list[] = { INDEX_op_shrv_vec, 0 };
+    static const GVecGen3 g[4] = {
+        { .fniv = tcg_gen_shrv_mod_vec,
+          .fno = gen_helper_gvec_shr8v,
+          .opt_opc = vecop_list,
+          .vece = MO_8 },
+        { .fniv = tcg_gen_shrv_mod_vec,
+          .fno = gen_helper_gvec_shr16v,
+          .opt_opc = vecop_list,
+          .vece = MO_16 },
+        { .fni4 = tcg_gen_shr_mod_i32,
+          .fniv = tcg_gen_shrv_mod_vec,
+          .fno = gen_helper_gvec_shr32v,
+          .opt_opc = vecop_list,
+          .vece = MO_32 },
+        { .fni8 = tcg_gen_shr_mod_i64,
+          .fniv = tcg_gen_shrv_mod_vec,
+          .fno = gen_helper_gvec_shr64v,
+          .opt_opc = vecop_list,
+          .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+          .vece = MO_64 },
+    };
+
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
+/*
+ * Similarly for arithmetic right shifts.
+ */
+
+static void tcg_gen_sarv_mod_vec(unsigned vece, TCGv_vec d,
+                                 TCGv_vec a, TCGv_vec b)
+{
+    TCGv_vec t = tcg_temp_new_vec_matching(d);
+
+    tcg_gen_dupi_vec(vece, t, (8 << vece) - 1);
+    tcg_gen_and_vec(vece, t, t, b);
+    tcg_gen_sarv_vec(vece, d, a, t);
+    tcg_temp_free_vec(t);
+}
+
+static void tcg_gen_sar_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 t = tcg_temp_new_i32();
+
+    tcg_gen_andi_i32(t, b, 31);
+    tcg_gen_sar_i32(d, a, t);
+    tcg_temp_free_i32(t);
+}
+
+static void tcg_gen_sar_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 t = tcg_temp_new_i64();
+
+    tcg_gen_andi_i64(t, b, 63);
+    tcg_gen_sar_i64(d, a, t);
+    tcg_temp_free_i64(t);
+}
+
+void tcg_gen_gvec_sarv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const TCGOpcode vecop_list[] = { INDEX_op_sarv_vec, 0 };
+    static const GVecGen3 g[4] = {
+        { .fniv = tcg_gen_sarv_mod_vec,
+          .fno = gen_helper_gvec_sar8v,
+          .opt_opc = vecop_list,
+          .vece = MO_8 },
+        { .fniv = tcg_gen_sarv_mod_vec,
+          .fno = gen_helper_gvec_sar16v,
+          .opt_opc = vecop_list,
+          .vece = MO_16 },
+        { .fni4 = tcg_gen_sar_mod_i32,
+          .fniv = tcg_gen_sarv_mod_vec,
+          .fno = gen_helper_gvec_sar32v,
+          .opt_opc = vecop_list,
+          .vece = MO_32 },
+        { .fni8 = tcg_gen_sar_mod_i64,
+          .fniv = tcg_gen_sarv_mod_vec,
+          .fno = gen_helper_gvec_sar64v,
+          .opt_opc = vecop_list,
+          .prefer_i64 = TCG_TARGET_REG_BITS == 64,
+          .vece = MO_64 },
+    };
+
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
 /* Expand OPSZ bytes worth of three-operand operations using i32 elements.  */
 static void expand_cmp_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
                            uint32_t oprsz, TCGCond cond)
diff --git a/tcg/tcg-op-gvec.h b/tcg/tcg-op-gvec.h
index ac744ff7c9..84a6247b16 100644
--- a/tcg/tcg-op-gvec.h
+++ b/tcg/tcg-op-gvec.h
@@ -318,6 +318,17 @@ void tcg_gen_gvec_shri(unsigned vece, uint32_t dofs, uint32_t aofs,
 void tcg_gen_gvec_sari(unsigned vece, uint32_t dofs, uint32_t aofs,
                        int64_t shift, uint32_t oprsz, uint32_t maxsz);
 
+/*
+ * Perform vector shift by vector element, modulo the element size.
+ * E.g.  D[i] = A[i] << (B[i] % (8 << vece)).
+ */
+void tcg_gen_gvec_shlv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_shrv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_sarv(unsigned vece, uint32_t dofs, uint32_t aofs,
+                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+
 void tcg_gen_gvec_cmp(TCGCond cond, 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 213d2e22aa..96317dbd10 100644
--- a/tcg/tcg-op-vec.c
+++ b/tcg/tcg-op-vec.c
@@ -583,3 +583,18 @@ void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
 {
     do_op3(vece, r, a, b, INDEX_op_umax_vec);
 }
+
+void tcg_gen_shlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
+{
+    do_op3(vece, r, a, b, INDEX_op_shlv_vec);
+}
+
+void tcg_gen_shrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
+{
+    do_op3(vece, r, a, b, INDEX_op_shrv_vec);
+}
+
+void tcg_gen_sarv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
+{
+    do_op3(vece, r, a, b, INDEX_op_sarv_vec);
+}
diff --git a/tcg/tcg-op.h b/tcg/tcg-op.h
index 9fff9864f6..833c6330b5 100644
--- a/tcg/tcg-op.h
+++ b/tcg/tcg-op.h
@@ -986,6 +986,10 @@ 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);
 void tcg_gen_sari_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i);
 
+void tcg_gen_shlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec s);
+void tcg_gen_shrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec s);
+void tcg_gen_sarv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec s);
+
 void tcg_gen_cmp_vec(TCGCond cond, unsigned vece, TCGv_vec r,
                      TCGv_vec a, TCGv_vec b);