diff mbox series

[v4,17/48] target/loongarch: Implement xvmul/xvmuh/xvmulw{ev/od}

Message ID 20230830084902.2113960-18-gaosong@loongson.cn
State New
Headers show
Series Add LoongArch LASX instructions | expand

Commit Message

gaosong Aug. 30, 2023, 8:48 a.m. UTC 
This patch includes:
- XVMUL.{B/H/W/D};
- XVMUH.{B/H/W/D}[U];
- XVMULW{EV/OD}.{H.B/W.H/D.W/Q.D}[U];
- XVMULW{EV/OD}.{H.BU.B/W.HU.H/D.WU.W/Q.DU.D}.

Signed-off-by: Song Gao <gaosong@loongson.cn>
---
 target/loongarch/vec.h                       |  2 +
 target/loongarch/insns.decode                | 38 +++++++++++++
 target/loongarch/disas.c                     | 38 +++++++++++++
 target/loongarch/vec_helper.c                | 57 ++++++++++---------
 target/loongarch/insn_trans/trans_lasx.c.inc | 42 ++++++++++++++
 target/loongarch/insn_trans/trans_lsx.c.inc  | 60 ++++++++++----------
 6 files changed, 180 insertions(+), 57 deletions(-)

Comments

Richard Henderson Aug. 30, 2023, 6:23 p.m. UTC | #1 
On 8/30/23 01:48, Song Gao wrote:
> This patch includes:
> - XVMUL.{B/H/W/D};
> - XVMUH.{B/H/W/D}[U];
> - XVMULW{EV/OD}.{H.B/W.H/D.W/Q.D}[U];
> - XVMULW{EV/OD}.{H.BU.B/W.HU.H/D.WU.W/Q.DU.D}.
> 
> Signed-off-by: Song Gao<gaosong@loongson.cn>
> ---
>   target/loongarch/vec.h                       |  2 +
>   target/loongarch/insns.decode                | 38 +++++++++++++
>   target/loongarch/disas.c                     | 38 +++++++++++++
>   target/loongarch/vec_helper.c                | 57 ++++++++++---------
>   target/loongarch/insn_trans/trans_lasx.c.inc | 42 ++++++++++++++
>   target/loongarch/insn_trans/trans_lsx.c.inc  | 60 ++++++++++----------
>   6 files changed, 180 insertions(+), 57 deletions(-)
> 
> diff --git a/target/loongarch/vec.h b/target/loongarch/vec.h
> index cd6f6a72fd..6fc84c8c5a 100644
> --- a/target/loongarch/vec.h
> +++ b/target/loongarch/vec.h
> @@ -60,4 +60,6 @@
>   #define DO_MIN(a, b)    (a < b ? a : b)
>   #define DO_MAX(a, b)    (a > b ? a : b)
>   
> +#define DO_MUL(a, b)    (a * b)
> +

No need to move this.

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


r~
diff mbox series

Patch

diff --git a/target/loongarch/vec.h b/target/loongarch/vec.h
index cd6f6a72fd..6fc84c8c5a 100644
--- a/target/loongarch/vec.h
+++ b/target/loongarch/vec.h
@@ -60,4 +60,6 @@ 
 #define DO_MIN(a, b)    (a < b ? a : b)
 #define DO_MAX(a, b)    (a > b ? a : b)
 
+#define DO_MUL(a, b)    (a * b)
+
 #endif /* LOONGARCH_VEC_H */
diff --git a/target/loongarch/insns.decode b/target/loongarch/insns.decode
index 99aefcb651..0f9ebe641f 100644
--- a/target/loongarch/insns.decode
+++ b/target/loongarch/insns.decode
@@ -1473,6 +1473,44 @@  xvmini_hu        0111 01101001 01101 ..... ..... .....    @vv_ui5
 xvmini_wu        0111 01101001 01110 ..... ..... .....    @vv_ui5
 xvmini_du        0111 01101001 01111 ..... ..... .....    @vv_ui5
 
+xvmul_b          0111 01001000 01000 ..... ..... .....    @vvv
+xvmul_h          0111 01001000 01001 ..... ..... .....    @vvv
+xvmul_w          0111 01001000 01010 ..... ..... .....    @vvv
+xvmul_d          0111 01001000 01011 ..... ..... .....    @vvv
+xvmuh_b          0111 01001000 01100 ..... ..... .....    @vvv
+xvmuh_h          0111 01001000 01101 ..... ..... .....    @vvv
+xvmuh_w          0111 01001000 01110 ..... ..... .....    @vvv
+xvmuh_d          0111 01001000 01111 ..... ..... .....    @vvv
+xvmuh_bu         0111 01001000 10000 ..... ..... .....    @vvv
+xvmuh_hu         0111 01001000 10001 ..... ..... .....    @vvv
+xvmuh_wu         0111 01001000 10010 ..... ..... .....    @vvv
+xvmuh_du         0111 01001000 10011 ..... ..... .....    @vvv
+
+xvmulwev_h_b     0111 01001001 00000 ..... ..... .....    @vvv
+xvmulwev_w_h     0111 01001001 00001 ..... ..... .....    @vvv
+xvmulwev_d_w     0111 01001001 00010 ..... ..... .....    @vvv
+xvmulwev_q_d     0111 01001001 00011 ..... ..... .....    @vvv
+xvmulwod_h_b     0111 01001001 00100 ..... ..... .....    @vvv
+xvmulwod_w_h     0111 01001001 00101 ..... ..... .....    @vvv
+xvmulwod_d_w     0111 01001001 00110 ..... ..... .....    @vvv
+xvmulwod_q_d     0111 01001001 00111 ..... ..... .....    @vvv
+xvmulwev_h_bu    0111 01001001 10000 ..... ..... .....    @vvv
+xvmulwev_w_hu    0111 01001001 10001 ..... ..... .....    @vvv
+xvmulwev_d_wu    0111 01001001 10010 ..... ..... .....    @vvv
+xvmulwev_q_du    0111 01001001 10011 ..... ..... .....    @vvv
+xvmulwod_h_bu    0111 01001001 10100 ..... ..... .....    @vvv
+xvmulwod_w_hu    0111 01001001 10101 ..... ..... .....    @vvv
+xvmulwod_d_wu    0111 01001001 10110 ..... ..... .....    @vvv
+xvmulwod_q_du    0111 01001001 10111 ..... ..... .....    @vvv
+xvmulwev_h_bu_b  0111 01001010 00000 ..... ..... .....    @vvv
+xvmulwev_w_hu_h  0111 01001010 00001 ..... ..... .....    @vvv
+xvmulwev_d_wu_w  0111 01001010 00010 ..... ..... .....    @vvv
+xvmulwev_q_du_d  0111 01001010 00011 ..... ..... .....    @vvv
+xvmulwod_h_bu_b  0111 01001010 00100 ..... ..... .....    @vvv
+xvmulwod_w_hu_h  0111 01001010 00101 ..... ..... .....    @vvv
+xvmulwod_d_wu_w  0111 01001010 00110 ..... ..... .....    @vvv
+xvmulwod_q_du_d  0111 01001010 00111 ..... ..... .....    @vvv
+
 xvreplgr2vr_b    0111 01101001 11110 00000 ..... .....    @vr
 xvreplgr2vr_h    0111 01101001 11110 00001 ..... .....    @vr
 xvreplgr2vr_w    0111 01101001 11110 00010 ..... .....    @vr
diff --git a/target/loongarch/disas.c b/target/loongarch/disas.c
index 63c1dc757f..e5f9a6bcdf 100644
--- a/target/loongarch/disas.c
+++ b/target/loongarch/disas.c
@@ -1890,6 +1890,44 @@  INSN_LASX(xvmini_hu,         vv_i)
 INSN_LASX(xvmini_wu,         vv_i)
 INSN_LASX(xvmini_du,         vv_i)
 
+INSN_LASX(xvmul_b,           vvv)
+INSN_LASX(xvmul_h,           vvv)
+INSN_LASX(xvmul_w,           vvv)
+INSN_LASX(xvmul_d,           vvv)
+INSN_LASX(xvmuh_b,           vvv)
+INSN_LASX(xvmuh_h,           vvv)
+INSN_LASX(xvmuh_w,           vvv)
+INSN_LASX(xvmuh_d,           vvv)
+INSN_LASX(xvmuh_bu,          vvv)
+INSN_LASX(xvmuh_hu,          vvv)
+INSN_LASX(xvmuh_wu,          vvv)
+INSN_LASX(xvmuh_du,          vvv)
+
+INSN_LASX(xvmulwev_h_b,      vvv)
+INSN_LASX(xvmulwev_w_h,      vvv)
+INSN_LASX(xvmulwev_d_w,      vvv)
+INSN_LASX(xvmulwev_q_d,      vvv)
+INSN_LASX(xvmulwod_h_b,      vvv)
+INSN_LASX(xvmulwod_w_h,      vvv)
+INSN_LASX(xvmulwod_d_w,      vvv)
+INSN_LASX(xvmulwod_q_d,      vvv)
+INSN_LASX(xvmulwev_h_bu,     vvv)
+INSN_LASX(xvmulwev_w_hu,     vvv)
+INSN_LASX(xvmulwev_d_wu,     vvv)
+INSN_LASX(xvmulwev_q_du,     vvv)
+INSN_LASX(xvmulwod_h_bu,     vvv)
+INSN_LASX(xvmulwod_w_hu,     vvv)
+INSN_LASX(xvmulwod_d_wu,     vvv)
+INSN_LASX(xvmulwod_q_du,     vvv)
+INSN_LASX(xvmulwev_h_bu_b,   vvv)
+INSN_LASX(xvmulwev_w_hu_h,   vvv)
+INSN_LASX(xvmulwev_d_wu_w,   vvv)
+INSN_LASX(xvmulwev_q_du_d,   vvv)
+INSN_LASX(xvmulwod_h_bu_b,   vvv)
+INSN_LASX(xvmulwod_w_hu_h,   vvv)
+INSN_LASX(xvmulwod_d_wu_w,   vvv)
+INSN_LASX(xvmulwod_q_du_d,   vvv)
+
 INSN_LASX(xvreplgr2vr_b,     vr)
 INSN_LASX(xvreplgr2vr_h,     vr)
 INSN_LASX(xvreplgr2vr_w,     vr)
diff --git a/target/loongarch/vec_helper.c b/target/loongarch/vec_helper.c
index 0c641d80c7..f641950cbe 100644
--- a/target/loongarch/vec_helper.c
+++ b/target/loongarch/vec_helper.c
@@ -434,58 +434,59 @@  VMINMAXI(vmaxi_hu, 16, UH, DO_MAX)
 VMINMAXI(vmaxi_wu, 32, UW, DO_MAX)
 VMINMAXI(vmaxi_du, 64, UD, DO_MAX)
 
-#define DO_VMUH(NAME, BIT, E1, E2, DO_OP)                   \
-void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t v) \
-{                                                           \
-    int i;                                                  \
-    VReg *Vd = (VReg *)vd;                                  \
-    VReg *Vj = (VReg *)vj;                                  \
-    VReg *Vk = (VReg *)vk;                                  \
-    typedef __typeof(Vd->E1(0)) T;                          \
-                                                            \
-    for (i = 0; i < LSX_LEN/BIT; i++) {                     \
-        Vd->E2(i) = ((T)Vj->E2(i)) * ((T)Vk->E2(i)) >> BIT; \
-    }                                                       \
+#define DO_VMUH(NAME, BIT, E1, E2, DO_OP)                      \
+void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
+{                                                              \
+    int i;                                                     \
+    VReg *Vd = (VReg *)vd;                                     \
+    VReg *Vj = (VReg *)vj;                                     \
+    VReg *Vk = (VReg *)vk;                                     \
+    typedef __typeof(Vd->E1(0)) T;                             \
+    int oprsz = simd_oprsz(desc);                              \
+                                                               \
+    for (i = 0; i < oprsz / (BIT / 8); i++) {                  \
+        Vd->E2(i) = ((T)Vj->E2(i)) * ((T)Vk->E2(i)) >> BIT;    \
+    }                                                          \
 }
 
-void HELPER(vmuh_d)(void *vd, void *vj, void *vk, uint32_t v)
+void HELPER(vmuh_d)(void *vd, void *vj, void *vk, uint32_t desc)
 {
-    uint64_t l, h1, h2;
+    int i;
+    uint64_t l, h;
     VReg *Vd = (VReg *)vd;
     VReg *Vj = (VReg *)vj;
     VReg *Vk = (VReg *)vk;
+    int oprsz = simd_oprsz(desc);
 
-    muls64(&l, &h1, Vj->D(0), Vk->D(0));
-    muls64(&l, &h2, Vj->D(1), Vk->D(1));
-
-    Vd->D(0) = h1;
-    Vd->D(1) = h2;
+    for (i = 0; i < oprsz / 8; i++) {
+        muls64(&l, &h, Vj->D(i), Vk->D(i));
+        Vd->D(i) = h;
+    }
 }
 
 DO_VMUH(vmuh_b, 8, H, B, DO_MUH)
 DO_VMUH(vmuh_h, 16, W, H, DO_MUH)
 DO_VMUH(vmuh_w, 32, D, W, DO_MUH)
 
-void HELPER(vmuh_du)(void *vd, void *vj, void *vk, uint32_t v)
+void HELPER(vmuh_du)(void *vd, void *vj, void *vk, uint32_t desc)
 {
-    uint64_t l, h1, h2;
+    int i;
+    uint64_t l, h;
     VReg *Vd = (VReg *)vd;
     VReg *Vj = (VReg *)vj;
     VReg *Vk = (VReg *)vk;
+    int oprsz = simd_oprsz(desc);
 
-    mulu64(&l, &h1, Vj->D(0), Vk->D(0));
-    mulu64(&l, &h2, Vj->D(1), Vk->D(1));
-
-    Vd->D(0) = h1;
-    Vd->D(1) = h2;
+    for (i = 0; i < oprsz / 8; i++) {
+        mulu64(&l, &h, Vj->D(i), Vk->D(i));
+        Vd->D(i) = h;
+    }
 }
 
 DO_VMUH(vmuh_bu, 8, UH, UB, DO_MUH)
 DO_VMUH(vmuh_hu, 16, UW, UH, DO_MUH)
 DO_VMUH(vmuh_wu, 32, UD, UW, DO_MUH)
 
-#define DO_MUL(a, b) (a * b)
-
 DO_EVEN(vmulwev_h_b, 16, H, B, DO_MUL)
 DO_EVEN(vmulwev_w_h, 32, W, H, DO_MUL)
 DO_EVEN(vmulwev_d_w, 64, D, W, DO_MUL)
diff --git a/target/loongarch/insn_trans/trans_lasx.c.inc b/target/loongarch/insn_trans/trans_lasx.c.inc
index ba31da6578..ca9361782e 100644
--- a/target/loongarch/insn_trans/trans_lasx.c.inc
+++ b/target/loongarch/insn_trans/trans_lasx.c.inc
@@ -211,6 +211,48 @@  TRANS(xvmaxi_hu, LASX, gvec_vv_i, 32, MO_16, do_vmaxi_u)
 TRANS(xvmaxi_wu, LASX, gvec_vv_i, 32, MO_32, do_vmaxi_u)
 TRANS(xvmaxi_du, LASX, gvec_vv_i, 32, MO_64, do_vmaxi_u)
 
+TRANS(xvmul_b, LASX, gvec_vvv, 32, MO_8, tcg_gen_gvec_mul)
+TRANS(xvmul_h, LASX, gvec_vvv, 32, MO_16, tcg_gen_gvec_mul)
+TRANS(xvmul_w, LASX, gvec_vvv, 32, MO_32, tcg_gen_gvec_mul)
+TRANS(xvmul_d, LASX, gvec_vvv, 32, MO_64, tcg_gen_gvec_mul)
+TRANS(xvmuh_b, LASX, gvec_vvv, 32, MO_8, do_vmuh_s)
+TRANS(xvmuh_h, LASX, gvec_vvv, 32, MO_16, do_vmuh_s)
+TRANS(xvmuh_w, LASX, gvec_vvv, 32, MO_32, do_vmuh_s)
+TRANS(xvmuh_d, LASX, gvec_vvv, 32, MO_64, do_vmuh_s)
+TRANS(xvmuh_bu, LASX, gvec_vvv, 32, MO_8, do_vmuh_u)
+TRANS(xvmuh_hu, LASX, gvec_vvv, 32, MO_16, do_vmuh_u)
+TRANS(xvmuh_wu, LASX, gvec_vvv, 32, MO_32, do_vmuh_u)
+TRANS(xvmuh_du, LASX, gvec_vvv, 32, MO_64, do_vmuh_u)
+
+TRANS(xvmulwev_h_b, LASX, gvec_vvv, 32, MO_8, do_vmulwev_s)
+TRANS(xvmulwev_w_h, LASX, gvec_vvv, 32, MO_16, do_vmulwev_s)
+TRANS(xvmulwev_d_w, LASX, gvec_vvv, 32, MO_32, do_vmulwev_s)
+
+TRANS(xvmulwev_q_d, LASX, gen_vmul_q, 32, 0, 0, tcg_gen_muls2_i64)
+TRANS(xvmulwod_q_d, LASX, gen_vmul_q, 32, 1, 1, tcg_gen_muls2_i64)
+TRANS(xvmulwev_q_du, LASX, gen_vmul_q, 32, 0, 0, tcg_gen_mulu2_i64)
+TRANS(xvmulwod_q_du, LASX, gen_vmul_q, 32, 1, 1, tcg_gen_mulu2_i64)
+TRANS(xvmulwev_q_du_d, LASX, gen_vmul_q, 32, 0, 0, tcg_gen_mulus2_i64)
+TRANS(xvmulwod_q_du_d, LASX, gen_vmul_q, 32, 1, 1, tcg_gen_mulus2_i64)
+
+TRANS(xvmulwod_h_b, LASX, gvec_vvv, 32, MO_8, do_vmulwod_s)
+TRANS(xvmulwod_w_h, LASX, gvec_vvv, 32, MO_16, do_vmulwod_s)
+TRANS(xvmulwod_d_w, LASX, gvec_vvv, 32, MO_32, do_vmulwod_s)
+
+TRANS(xvmulwev_h_bu, LASX, gvec_vvv, 32, MO_8, do_vmulwev_u)
+TRANS(xvmulwev_w_hu, LASX, gvec_vvv, 32, MO_16, do_vmulwev_u)
+TRANS(xvmulwev_d_wu, LASX, gvec_vvv, 32, MO_32, do_vmulwev_u)
+TRANS(xvmulwod_h_bu, LASX, gvec_vvv, 32, MO_8, do_vmulwod_u)
+TRANS(xvmulwod_w_hu, LASX, gvec_vvv, 32, MO_16, do_vmulwod_u)
+TRANS(xvmulwod_d_wu, LASX, gvec_vvv, 32, MO_32, do_vmulwod_u)
+
+TRANS(xvmulwev_h_bu_b, LASX, gvec_vvv, 32, MO_8, do_vmulwev_u_s)
+TRANS(xvmulwev_w_hu_h, LASX, gvec_vvv, 32, MO_16, do_vmulwev_u_s)
+TRANS(xvmulwev_d_wu_w, LASX, gvec_vvv, 32, MO_32, do_vmulwev_u_s)
+TRANS(xvmulwod_h_bu_b, LASX, gvec_vvv, 32, MO_8, do_vmulwod_u_s)
+TRANS(xvmulwod_w_hu_h, LASX, gvec_vvv, 32, MO_16, do_vmulwod_u_s)
+TRANS(xvmulwod_d_wu_w, LASX, gvec_vvv, 32, MO_32, do_vmulwod_u_s)
+
 TRANS(xvreplgr2vr_b, LASX, gvec_dup, 32, MO_8)
 TRANS(xvreplgr2vr_h, LASX, gvec_dup, 32, MO_16)
 TRANS(xvreplgr2vr_w, LASX, gvec_dup, 32, MO_32)
diff --git a/target/loongarch/insn_trans/trans_lsx.c.inc b/target/loongarch/insn_trans/trans_lsx.c.inc
index 5653a556bf..d25f89a6a4 100644
--- a/target/loongarch/insn_trans/trans_lsx.c.inc
+++ b/target/loongarch/insn_trans/trans_lsx.c.inc
@@ -1764,37 +1764,39 @@  static void tcg_gen_mulus2_i64(TCGv_i64 rl, TCGv_i64 rh,
     tcg_gen_mulsu2_i64(rl, rh, arg2, arg1);
 }
 
-#define VMUL_Q(NAME, FN, idx1, idx2)                      \
-static bool trans_## NAME (DisasContext *ctx, arg_vvv *a) \
-{                                                         \
-    TCGv_i64 rh, rl, arg1, arg2;                          \
-                                                          \
-    if (!avail_LSX(ctx)) {                                \
-        return false;                                     \
-    }                                                     \
-                                                          \
-    rh = tcg_temp_new_i64();                              \
-    rl = tcg_temp_new_i64();                              \
-    arg1 = tcg_temp_new_i64();                            \
-    arg2 = tcg_temp_new_i64();                            \
-                                                          \
-    get_vreg64(arg1, a->vj, idx1);                        \
-    get_vreg64(arg2, a->vk, idx2);                        \
-                                                          \
-    tcg_gen_## FN ##_i64(rl, rh, arg1, arg2);             \
-                                                          \
-    set_vreg64(rh, a->vd, 1);                             \
-    set_vreg64(rl, a->vd, 0);                             \
-                                                          \
-    return true;                                          \
+static bool gen_vmul_q(DisasContext *ctx,
+                       arg_vvv *a, int oprsz, int idx1, int idx2,
+                       void (*func)(TCGv_i64, TCGv_i64, TCGv_i64, TCGv_i64))
+{
+    TCGv_i64 rh, rl, arg1, arg2;
+    int i;
+
+    CHECK_VEC;
+
+    rh = tcg_temp_new_i64();
+    rl = tcg_temp_new_i64();
+    arg1 = tcg_temp_new_i64();
+    arg2 = tcg_temp_new_i64();
+
+    for (i = 0; i < oprsz / 16; i++) {
+        get_vreg64(arg1, a->vj, 2 * i + idx1);
+        get_vreg64(arg2, a->vk, 2 * i + idx2);
+
+        func(rl, rh, arg1, arg2);
+
+        set_vreg64(rh, a->vd, 2 * i + 1);
+        set_vreg64(rl, a->vd, 2 * i);
+    }
+
+    return true;
 }
 
-VMUL_Q(vmulwev_q_d, muls2, 0, 0)
-VMUL_Q(vmulwod_q_d, muls2, 1, 1)
-VMUL_Q(vmulwev_q_du, mulu2, 0, 0)
-VMUL_Q(vmulwod_q_du, mulu2, 1, 1)
-VMUL_Q(vmulwev_q_du_d, mulus2, 0, 0)
-VMUL_Q(vmulwod_q_du_d, mulus2, 1, 1)
+TRANS(vmulwev_q_d, LSX, gen_vmul_q, 16, 0, 0, tcg_gen_muls2_i64)
+TRANS(vmulwod_q_d, LSX, gen_vmul_q, 16, 1, 1, tcg_gen_muls2_i64)
+TRANS(vmulwev_q_du, LSX, gen_vmul_q, 16, 0, 0, tcg_gen_mulu2_i64)
+TRANS(vmulwod_q_du, LSX, gen_vmul_q, 16, 1, 1, tcg_gen_mulu2_i64)
+TRANS(vmulwev_q_du_d, LSX, gen_vmul_q, 16, 0, 0, tcg_gen_mulus2_i64)
+TRANS(vmulwod_q_du_d, LSX, gen_vmul_q, 16, 1, 1, tcg_gen_mulus2_i64)
 
 static void gen_vmulwod_s(unsigned vece, TCGv_vec t, TCGv_vec a, TCGv_vec b)
 {