diff mbox series

[v5,24/49] target/ppc: move vrl[bhwd]nm/vrl[bhwd]mi to decodetree

Message ID 20220225210936.1749575-25-matheus.ferst@eldorado.org.br (mailing list archive)
State New, archived
Headers show
Series target/ppc: PowerISA Vector/VSX instruction batch | expand

Commit Message

Matheus K. Ferst Feb. 25, 2022, 9:09 p.m. UTC
From: Matheus Ferst <matheus.ferst@eldorado.org.br>

Signed-off-by: Matheus Ferst <matheus.ferst@eldorado.org.br>
---
I couldn't figure out how to use tcg_gen_gvec_rotlv here. Since the code
is in the fniv implementation, we have TCGv_vec instead of offsets. I'm
keeping the masking for now, so the generated code has the desired
effect.
---
 target/ppc/helper.h                 |   8 +-
 target/ppc/insn32.decode            |   6 ++
 target/ppc/int_helper.c             |  50 ++++-----
 target/ppc/translate/vmx-impl.c.inc | 152 ++++++++++++++++++++++++++--
 target/ppc/translate/vmx-ops.c.inc  |   5 +-
 5 files changed, 182 insertions(+), 39 deletions(-)

Comments

Richard Henderson Feb. 25, 2022, 9:31 p.m. UTC | #1
On 2/25/22 11:09, matheus.ferst@eldorado.org.br wrote:
> From: Matheus Ferst<matheus.ferst@eldorado.org.br>
> 
> Signed-off-by: Matheus Ferst<matheus.ferst@eldorado.org.br>
> ---
> I couldn't figure out how to use tcg_gen_gvec_rotlv here. Since the code
> is in the fniv implementation, we have TCGv_vec instead of offsets. I'm
> keeping the masking for now, so the generated code has the desired
> effect.

Fair.

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>

r~
diff mbox series

Patch

diff --git a/target/ppc/helper.h b/target/ppc/helper.h
index e763093dbb..6bd7fad70c 100644
--- a/target/ppc/helper.h
+++ b/target/ppc/helper.h
@@ -271,10 +271,10 @@  DEF_HELPER_4(vmaxfp, void, env, avr, avr, avr)
 DEF_HELPER_4(vminfp, void, env, avr, avr, avr)
 DEF_HELPER_3(vrefp, void, env, avr, avr)
 DEF_HELPER_3(vrsqrtefp, void, env, avr, avr)
-DEF_HELPER_3(vrlwmi, void, avr, avr, avr)
-DEF_HELPER_3(vrldmi, void, avr, avr, avr)
-DEF_HELPER_3(vrldnm, void, avr, avr, avr)
-DEF_HELPER_3(vrlwnm, void, avr, avr, avr)
+DEF_HELPER_FLAGS_4(VRLWMI, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VRLDMI, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VRLDNM, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
+DEF_HELPER_FLAGS_4(VRLWNM, TCG_CALL_NO_RWG, void, avr, avr, avr, i32)
 DEF_HELPER_5(vmaddfp, void, env, avr, avr, avr, avr)
 DEF_HELPER_5(vnmsubfp, void, env, avr, avr, avr, avr)
 DEF_HELPER_3(vexptefp, void, env, avr, avr)
diff --git a/target/ppc/insn32.decode b/target/ppc/insn32.decode
index d918e2d0f2..e788dc5152 100644
--- a/target/ppc/insn32.decode
+++ b/target/ppc/insn32.decode
@@ -492,6 +492,12 @@  VRLH            000100 ..... ..... ..... 00001000100    @VX
 VRLW            000100 ..... ..... ..... 00010000100    @VX
 VRLD            000100 ..... ..... ..... 00011000100    @VX
 
+VRLWMI          000100 ..... ..... ..... 00010000101    @VX
+VRLDMI          000100 ..... ..... ..... 00011000101    @VX
+
+VRLWNM          000100 ..... ..... ..... 00110000101    @VX
+VRLDNM          000100 ..... ..... ..... 00111000101    @VX
+
 ## Vector Integer Arithmetic Instructions
 
 VEXTSB2W        000100 ..... 10000 ..... 11000000010    @VX_tb
diff --git a/target/ppc/int_helper.c b/target/ppc/int_helper.c
index 71b31fbd89..f52242ca81 100644
--- a/target/ppc/int_helper.c
+++ b/target/ppc/int_helper.c
@@ -1275,33 +1275,33 @@  void helper_vrsqrtefp(CPUPPCState *env, ppc_avr_t *r, ppc_avr_t *b)
     }
 }
 
-#define VRLMI(name, size, element, insert)                            \
-void helper_##name(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b)          \
-{                                                                     \
-    int i;                                                            \
-    for (i = 0; i < ARRAY_SIZE(r->element); i++) {                    \
-        uint##size##_t src1 = a->element[i];                          \
-        uint##size##_t src2 = b->element[i];                          \
-        uint##size##_t src3 = r->element[i];                          \
-        uint##size##_t begin, end, shift, mask, rot_val;              \
-                                                                      \
-        shift = extract##size(src2, 0, 6);                            \
-        end   = extract##size(src2, 8, 6);                            \
-        begin = extract##size(src2, 16, 6);                           \
-        rot_val = rol##size(src1, shift);                             \
-        mask = mask_u##size(begin, end);                              \
-        if (insert) {                                                 \
-            r->element[i] = (rot_val & mask) | (src3 & ~mask);        \
-        } else {                                                      \
-            r->element[i] = (rot_val & mask);                         \
-        }                                                             \
-    }                                                                 \
+#define VRLMI(name, size, element, insert)                                  \
+void helper_##name(ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b, uint32_t desc) \
+{                                                                           \
+    int i;                                                                  \
+    for (i = 0; i < ARRAY_SIZE(r->element); i++) {                          \
+        uint##size##_t src1 = a->element[i];                                \
+        uint##size##_t src2 = b->element[i];                                \
+        uint##size##_t src3 = r->element[i];                                \
+        uint##size##_t begin, end, shift, mask, rot_val;                    \
+                                                                            \
+        shift = extract##size(src2, 0, 6);                                  \
+        end   = extract##size(src2, 8, 6);                                  \
+        begin = extract##size(src2, 16, 6);                                 \
+        rot_val = rol##size(src1, shift);                                   \
+        mask = mask_u##size(begin, end);                                    \
+        if (insert) {                                                       \
+            r->element[i] = (rot_val & mask) | (src3 & ~mask);              \
+        } else {                                                            \
+            r->element[i] = (rot_val & mask);                               \
+        }                                                                   \
+    }                                                                       \
 }
 
-VRLMI(vrldmi, 64, u64, 1);
-VRLMI(vrlwmi, 32, u32, 1);
-VRLMI(vrldnm, 64, u64, 0);
-VRLMI(vrlwnm, 32, u32, 0);
+VRLMI(VRLDMI, 64, u64, 1);
+VRLMI(VRLWMI, 32, u32, 1);
+VRLMI(VRLDNM, 64, u64, 0);
+VRLMI(VRLWNM, 32, u32, 0);
 
 void helper_vsel(CPUPPCState *env, ppc_avr_t *r, ppc_avr_t *a, ppc_avr_t *b,
                  ppc_avr_t *c)
diff --git a/target/ppc/translate/vmx-impl.c.inc b/target/ppc/translate/vmx-impl.c.inc
index f24b78d42e..09d6c88e62 100644
--- a/target/ppc/translate/vmx-impl.c.inc
+++ b/target/ppc/translate/vmx-impl.c.inc
@@ -799,7 +799,6 @@  static void trans_vclzd(DisasContext *ctx)
 }
 
 GEN_VXFORM_V(vmuluwm, MO_32, tcg_gen_gvec_mul, 4, 2);
-GEN_VXFORM(vrlwnm, 2, 6);
 GEN_VXFORM(vsrv, 2, 28);
 GEN_VXFORM(vslv, 2, 29);
 GEN_VXFORM(vslo, 6, 16);
@@ -839,6 +838,152 @@  TRANS_FLAGS(ALTIVEC, VRLH, do_vector_gvec3_VX, MO_16, tcg_gen_gvec_rotlv)
 TRANS_FLAGS(ALTIVEC, VRLW, do_vector_gvec3_VX, MO_32, tcg_gen_gvec_rotlv)
 TRANS_FLAGS2(ALTIVEC_207, VRLD, do_vector_gvec3_VX, MO_64, tcg_gen_gvec_rotlv)
 
+static TCGv_vec do_vrl_mask_vec(unsigned vece, TCGv_vec vrb)
+{
+    TCGv_vec t0 = tcg_temp_new_vec_matching(vrb),
+             t1 = tcg_temp_new_vec_matching(vrb),
+             t2 = tcg_temp_new_vec_matching(vrb),
+             ones = tcg_constant_vec_matching(vrb, vece, -1);
+
+    /* Extract b and e */
+    tcg_gen_dupi_vec(vece, t2, (8 << vece) - 1);
+
+    tcg_gen_shri_vec(vece, t0, vrb, 16);
+    tcg_gen_and_vec(vece, t0, t0, t2);
+
+    tcg_gen_shri_vec(vece, t1, vrb, 8);
+    tcg_gen_and_vec(vece, t1, t1, t2);
+
+    /* Compare b and e to negate the mask where begin > end */
+    tcg_gen_cmp_vec(TCG_COND_GT, vece, t2, t0, t1);
+
+    /* Create the mask with (~0 >> b) ^ ((~0 >> e) >> 1) */
+    tcg_gen_shrv_vec(vece, t0, ones, t0);
+    tcg_gen_shrv_vec(vece, t1, ones, t1);
+    tcg_gen_shri_vec(vece, t1, t1, 1);
+    tcg_gen_xor_vec(vece, t0, t0, t1);
+
+    /* negate the mask */
+    tcg_gen_xor_vec(vece, t0, t0, t2);
+
+    tcg_temp_free_vec(t1);
+    tcg_temp_free_vec(t2);
+
+    return t0;
+}
+
+static void gen_vrlnm_vec(unsigned vece, TCGv_vec vrt, TCGv_vec vra,
+                          TCGv_vec vrb)
+{
+    TCGv_vec mask, n = tcg_temp_new_vec_matching(vrt);
+
+    /* Create the mask */
+    mask = do_vrl_mask_vec(vece, vrb);
+
+    /* Extract n */
+    tcg_gen_dupi_vec(vece, n, (8 << vece) - 1);
+    tcg_gen_and_vec(vece, n, vrb, n);
+
+    /* Rotate and mask */
+    tcg_gen_rotlv_vec(vece, vrt, vra, n);
+    tcg_gen_and_vec(vece, vrt, vrt, mask);
+
+    tcg_temp_free_vec(n);
+    tcg_temp_free_vec(mask);
+}
+
+static bool do_vrlnm(DisasContext *ctx, arg_VX *a, int vece)
+{
+    static const TCGOpcode vecop_list[] = {
+        INDEX_op_cmp_vec, INDEX_op_rotlv_vec, INDEX_op_sari_vec,
+        INDEX_op_shli_vec, INDEX_op_shri_vec, INDEX_op_shrv_vec, 0
+    };
+    static const GVecGen3 ops[2] = {
+        {
+            .fniv = gen_vrlnm_vec,
+            .fno = gen_helper_VRLWNM,
+            .opt_opc = vecop_list,
+            .load_dest = true,
+            .vece = MO_32
+        },
+        {
+            .fniv = gen_vrlnm_vec,
+            .fno = gen_helper_VRLDNM,
+            .opt_opc = vecop_list,
+            .load_dest = true,
+            .vece = MO_64
+        }
+    };
+
+    REQUIRE_INSNS_FLAGS2(ctx, ISA300);
+    REQUIRE_VSX(ctx);
+
+    tcg_gen_gvec_3(avr_full_offset(a->vrt), avr_full_offset(a->vra),
+                   avr_full_offset(a->vrb), 16, 16, &ops[vece - 2]);
+
+    return true;
+}
+
+TRANS(VRLWNM, do_vrlnm, MO_32)
+TRANS(VRLDNM, do_vrlnm, MO_64)
+
+static void gen_vrlmi_vec(unsigned vece, TCGv_vec vrt, TCGv_vec vra,
+                          TCGv_vec vrb)
+{
+    TCGv_vec mask, n = tcg_temp_new_vec_matching(vrt),
+             tmp = tcg_temp_new_vec_matching(vrt);
+
+    /* Create the mask */
+    mask = do_vrl_mask_vec(vece, vrb);
+
+    /* Extract n */
+    tcg_gen_dupi_vec(vece, n, (8 << vece) - 1);
+    tcg_gen_and_vec(vece, n, vrb, n);
+
+    /* Rotate and insert */
+    tcg_gen_rotlv_vec(vece, tmp, vra, n);
+    tcg_gen_bitsel_vec(vece, vrt, mask, tmp, vrt);
+
+    tcg_temp_free_vec(n);
+    tcg_temp_free_vec(tmp);
+    tcg_temp_free_vec(mask);
+}
+
+static bool do_vrlmi(DisasContext *ctx, arg_VX *a, int vece)
+{
+    static const TCGOpcode vecop_list[] = {
+        INDEX_op_cmp_vec, INDEX_op_rotlv_vec, INDEX_op_sari_vec,
+        INDEX_op_shli_vec, INDEX_op_shri_vec, INDEX_op_shrv_vec, 0
+    };
+    static const GVecGen3 ops[2] = {
+        {
+            .fniv = gen_vrlmi_vec,
+            .fno = gen_helper_VRLWMI,
+            .opt_opc = vecop_list,
+            .load_dest = true,
+            .vece = MO_32
+        },
+        {
+            .fniv = gen_vrlnm_vec,
+            .fno = gen_helper_VRLDMI,
+            .opt_opc = vecop_list,
+            .load_dest = true,
+            .vece = MO_64
+        }
+    };
+
+    REQUIRE_INSNS_FLAGS2(ctx, ISA300);
+    REQUIRE_VSX(ctx);
+
+    tcg_gen_gvec_3(avr_full_offset(a->vrt), avr_full_offset(a->vra),
+                   avr_full_offset(a->vrb), 16, 16, &ops[vece - 2]);
+
+    return true;
+}
+
+TRANS(VRLWMI, do_vrlmi, MO_32)
+TRANS(VRLDMI, do_vrlmi, MO_64)
+
 static bool do_vector_shift_quad(DisasContext *ctx, arg_VX *a, bool right,
                                  bool alg)
 {
@@ -975,12 +1120,7 @@  GEN_VXFORM3(vsubeuqm, 31, 0);
 GEN_VXFORM3(vsubecuq, 31, 0);
 GEN_VXFORM_DUAL(vsubeuqm, PPC_NONE, PPC2_ALTIVEC_207, \
             vsubecuq, PPC_NONE, PPC2_ALTIVEC_207)
-GEN_VXFORM(vrlwmi, 2, 2);
-GEN_VXFORM(vrldmi, 2, 3);
 GEN_VXFORM_TRANS(vsl, 2, 7);
-GEN_VXFORM(vrldnm, 2, 7);
-GEN_VXFORM_DUAL(vsl, PPC_ALTIVEC, PPC_NONE, \
-                vrldnm, PPC_NONE, PPC2_ISA300)
 GEN_VXFORM_TRANS(vsr, 2, 11);
 GEN_VXFORM_ENV(vpkuhum, 7, 0);
 GEN_VXFORM_ENV(vpkuwum, 7, 1);
diff --git a/target/ppc/translate/vmx-ops.c.inc b/target/ppc/translate/vmx-ops.c.inc
index a7acea3ca7..3a8a9cc564 100644
--- a/target/ppc/translate/vmx-ops.c.inc
+++ b/target/ppc/translate/vmx-ops.c.inc
@@ -102,7 +102,6 @@  GEN_VXFORM_300(vextubrx, 6, 28),
 GEN_VXFORM_300(vextuhrx, 6, 29),
 GEN_VXFORM_DUAL(vmrgew, vextuwrx, 6, 30, PPC_NONE, PPC2_ALTIVEC_207),
 GEN_VXFORM_207(vmuluwm, 4, 2),
-GEN_VXFORM_300(vrlwnm, 2, 6),
 GEN_VXFORM_300(vsrv, 2, 28),
 GEN_VXFORM_300(vslv, 2, 29),
 GEN_VXFORM(vslo, 6, 16),
@@ -133,9 +132,7 @@  GEN_VXFORM_DUAL(vaddeuqm, vaddecuq, 30, 0xFF, PPC_NONE, PPC2_ALTIVEC_207),
 GEN_VXFORM_DUAL(vsubuqm, bcdtrunc, 0, 20, PPC2_ALTIVEC_207, PPC2_ISA300),
 GEN_VXFORM_DUAL(vsubcuq, bcdutrunc, 0, 21, PPC2_ALTIVEC_207, PPC2_ISA300),
 GEN_VXFORM_DUAL(vsubeuqm, vsubecuq, 31, 0xFF, PPC_NONE, PPC2_ALTIVEC_207),
-GEN_VXFORM_300(vrlwmi, 2, 2),
-GEN_VXFORM_300(vrldmi, 2, 3),
-GEN_VXFORM_DUAL(vsl, vrldnm, 2, 7, PPC_ALTIVEC, PPC_NONE),
+GEN_VXFORM(vsl, 2, 7),
 GEN_VXFORM(vsr, 2, 11),
 GEN_VXFORM(vpkuhum, 7, 0),
 GEN_VXFORM(vpkuwum, 7, 1),