projects / qemu.git / commitdiff
? search:
summary | shortlog | log | commit | commitdiff | tree
raw | patch | inline | side by side (parent: a5b3680)
target/ppc: Use gvec to decode XVCPSGN[SD]P
authorLucas Mateus Castro (alqotel) <lucas.araujo@eldorado.org.br>
Wed, 19 Oct 2022 12:50:37 +0000 (09:50 -0300)
committerDaniel Henrique Barboza <danielhb413@gmail.com>
Fri, 28 Oct 2022 16:15:22 +0000 (13:15 -0300)
Moved XVCPSGNSP and XVCPSGNDP to decodetree and used gvec to translate
them.

xvcpsgnsp:
rept    loop    master             patch
8       12500   0,00561400         0,00537900 (-4.2%)
25      4000    0,00562100         0,00400000 (-28.8%)
100     1000    0,00696900         0,00416300 (-40.3%)
500     200     0,02211900         0,00840700 (-62.0%)
2500    40      0,09328600         0,02728300 (-70.8%)
8000    12      0,27295300         0,06867800 (-74.8%)

xvcpsgndp:
rept    loop    master             patch
8       12500   0,00556300         0,00584200 (+5.0%)
25      4000    0,00482700         0,00431700 (-10.6%)
100     1000    0,00585800         0,00464400 (-20.7%)
500     200     0,01565300         0,00839700 (-46.4%)
2500    40      0,05766500         0,02430600 (-57.8%)
8000    12      0,19875300         0,07947100 (-60.0%)

Like the previous instructions there seemed to be a improvement on
translation time.

Signed-off-by: Lucas Mateus Castro (alqotel) <lucas.araujo@eldorado.org.br>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-Id: <20221019125040.48028-10-lucas.araujo@eldorado.org.br>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
target/ppc/insn32.decode patch | blob | history
target/ppc/translate/vsx-impl.c.inc patch | blob | history
target/ppc/translate/vsx-ops.c.inc patch | blob | history

diff --git a/target/ppc/insn32.decode b/target/ppc/insn32.decode
index 395339ce408e505fbf1a7401cc23d078c8216d57..3594c0c9607fed34b40c033fb2950d85bda8e48e 100644 (file)
--- a/target/ppc/insn32.decode
+++ b/target/ppc/insn32.decode
@@ -762,6 +762,8 @@ XVNABSDP        111100 ..... 00000 ..... 111101001 ..   @XX2
 XVNABSSP        111100 ..... 00000 ..... 110101001 ..   @XX2
 XVNEGDP         111100 ..... 00000 ..... 111111001 ..   @XX2
 XVNEGSP         111100 ..... 00000 ..... 110111001 ..   @XX2
+XVCPSGNDP       111100 ..... ..... ..... 11110000 ...   @XX3
+XVCPSGNSP       111100 ..... ..... ..... 11010000 ...   @XX3
 
 ## VSX Scalar Multiply-Add Instructions
 
diff --git a/target/ppc/translate/vsx-impl.c.inc b/target/ppc/translate/vsx-impl.c.inc
index 8717e20d089b1715d8b9674b18b70cb42cea5dc1..1c289238ecbd16d0345f41b1b08e744df5495d66 100644 (file)
--- a/target/ppc/translate/vsx-impl.c.inc
+++ b/target/ppc/translate/vsx-impl.c.inc
@@ -729,62 +729,6 @@ VSX_SCALAR_MOVE_QP(xsnabsqp, OP_NABS, SGN_MASK_DP)
 VSX_SCALAR_MOVE_QP(xsnegqp, OP_NEG, SGN_MASK_DP)
 VSX_SCALAR_MOVE_QP(xscpsgnqp, OP_CPSGN, SGN_MASK_DP)
 
-#define VSX_VECTOR_MOVE(name, op, sgn_mask)                      \
-static void glue(gen_, name)(DisasContext *ctx)                  \
-    {                                                            \
-        TCGv_i64 xbh, xbl, sgm;                                  \
-        if (unlikely(!ctx->vsx_enabled)) {                       \
-            gen_exception(ctx, POWERPC_EXCP_VSXU);               \
-            return;                                              \
-        }                                                        \
-        xbh = tcg_temp_new_i64();                                \
-        xbl = tcg_temp_new_i64();                                \
-        sgm = tcg_temp_new_i64();                                \
-        get_cpu_vsr(xbh, xB(ctx->opcode), true);                 \
-        get_cpu_vsr(xbl, xB(ctx->opcode), false);                \
-        tcg_gen_movi_i64(sgm, sgn_mask);                         \
-        switch (op) {                                            \
-            case OP_ABS: {                                       \
-                tcg_gen_andc_i64(xbh, xbh, sgm);                 \
-                tcg_gen_andc_i64(xbl, xbl, sgm);                 \
-                break;                                           \
-            }                                                    \
-            case OP_NABS: {                                      \
-                tcg_gen_or_i64(xbh, xbh, sgm);                   \
-                tcg_gen_or_i64(xbl, xbl, sgm);                   \
-                break;                                           \
-            }                                                    \
-            case OP_NEG: {                                       \
-                tcg_gen_xor_i64(xbh, xbh, sgm);                  \
-                tcg_gen_xor_i64(xbl, xbl, sgm);                  \
-                break;                                           \
-            }                                                    \
-            case OP_CPSGN: {                                     \
-                TCGv_i64 xah = tcg_temp_new_i64();               \
-                TCGv_i64 xal = tcg_temp_new_i64();               \
-                get_cpu_vsr(xah, xA(ctx->opcode), true);         \
-                get_cpu_vsr(xal, xA(ctx->opcode), false);        \
-                tcg_gen_and_i64(xah, xah, sgm);                  \
-                tcg_gen_and_i64(xal, xal, sgm);                  \
-                tcg_gen_andc_i64(xbh, xbh, sgm);                 \
-                tcg_gen_andc_i64(xbl, xbl, sgm);                 \
-                tcg_gen_or_i64(xbh, xbh, xah);                   \
-                tcg_gen_or_i64(xbl, xbl, xal);                   \
-                tcg_temp_free_i64(xah);                          \
-                tcg_temp_free_i64(xal);                          \
-                break;                                           \
-            }                                                    \
-        }                                                        \
-        set_cpu_vsr(xT(ctx->opcode), xbh, true);                 \
-        set_cpu_vsr(xT(ctx->opcode), xbl, false);                \
-        tcg_temp_free_i64(xbh);                                  \
-        tcg_temp_free_i64(xbl);                                  \
-        tcg_temp_free_i64(sgm);                                  \
-    }
-
-VSX_VECTOR_MOVE(xvcpsgndp, OP_CPSGN, SGN_MASK_DP)
-VSX_VECTOR_MOVE(xvcpsgnsp, OP_CPSGN, SGN_MASK_SP)
-
 #define TCG_OP_IMM_i64(FUNC, OP, IMM)                           \
     static void FUNC(TCGv_i64 t, TCGv_i64 b)                    \
     {                                                           \
@@ -852,6 +796,59 @@ TRANS(XVABSSP, do_vsx_msb_op, MO_32, do_xvabs_vec, do_xvabssp_i64)
 TRANS(XVNABSSP, do_vsx_msb_op, MO_32, do_xvnabs_vec, do_xvnabssp_i64)
 TRANS(XVNEGSP, do_vsx_msb_op, MO_32, do_xvneg_vec, do_xvnegsp_i64)
 
+static void do_xvcpsgndp_i64(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+    tcg_gen_andi_i64(a, a, SGN_MASK_DP);
+    tcg_gen_andi_i64(b, b, ~SGN_MASK_DP);
+    tcg_gen_or_i64(t, a, b);
+}
+
+static void do_xvcpsgnsp_i64(TCGv_i64 t, TCGv_i64 a, TCGv_i64 b)
+{
+    tcg_gen_andi_i64(a, a, SGN_MASK_SP);
+    tcg_gen_andi_i64(b, b, ~SGN_MASK_SP);
+    tcg_gen_or_i64(t, a, b);
+}
+
+static void do_xvcpsgn_vec(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
+{
+    uint64_t msb = (vece == MO_32) ? SGN_MASK_SP : SGN_MASK_DP;
+    tcg_gen_bitsel_vec(vece, t, tcg_constant_vec_matching(t, vece, msb), a, b);
+}
+
+static bool do_xvcpsgn(DisasContext *ctx, arg_XX3 *a, unsigned vece)
+{
+    static const TCGOpcode vecop_list[] = {
+        0
+    };
+
+    static const GVecGen3 op[] = {
+        {
+            .fni8 = do_xvcpsgnsp_i64,
+            .fniv = do_xvcpsgn_vec,
+            .opt_opc = vecop_list,
+            .vece = MO_32
+        },
+        {
+            .fni8 = do_xvcpsgndp_i64,
+            .fniv = do_xvcpsgn_vec,
+            .opt_opc = vecop_list,
+            .vece = MO_64
+        },
+    };
+
+    REQUIRE_INSNS_FLAGS2(ctx, VSX);
+    REQUIRE_VSX(ctx);
+
+    tcg_gen_gvec_3(vsr_full_offset(a->xt), vsr_full_offset(a->xa),
+                   vsr_full_offset(a->xb), 16, 16, &op[vece - MO_32]);
+
+    return true;
+}
+
+TRANS(XVCPSGNSP, do_xvcpsgn, MO_32)
+TRANS(XVCPSGNDP, do_xvcpsgn, MO_64)
+
 #define VSX_CMP(name, op1, op2, inval, type)                                  \
 static void gen_##name(DisasContext *ctx)                                     \
 {                                                                             \
diff --git a/target/ppc/translate/vsx-ops.c.inc b/target/ppc/translate/vsx-ops.c.inc
index b77324e0a8fdaec5566b92ef5685983fc9834eca..f7d7377379f3183af4b5e66818c8f26f92e7bca5 100644 (file)
--- a/target/ppc/translate/vsx-ops.c.inc
+++ b/target/ppc/translate/vsx-ops.c.inc
@@ -165,9 +165,6 @@ GEN_XX3FORM(name, opc2, opc3 | 1, fl2)
 GEN_XX2FORM_DCMX(xvtstdcdp, 0x14, 0x1E, PPC2_ISA300),
 GEN_XX2FORM_DCMX(xvtstdcsp, 0x14, 0x1A, PPC2_ISA300),
 
-GEN_XX3FORM(xvcpsgndp, 0x00, 0x1E, PPC2_VSX),
-GEN_XX3FORM(xvcpsgnsp, 0x00, 0x1A, PPC2_VSX),
-
 GEN_XX3FORM(xsadddp, 0x00, 0x04, PPC2_VSX),
 GEN_VSX_XFORM_300(xsaddqp, 0x04, 0x00, 0x0),
 GEN_XX3FORM(xssubdp, 0x00, 0x05, PPC2_VSX),
Unnamed repository; edit this file 'description' to name the repository.