qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[PATCH v5 26/60] target/riscv: vector widening saturating scaled multipl


From: LIU Zhiwei
Subject: [PATCH v5 26/60] target/riscv: vector widening saturating scaled multiply-add
Date: Thu, 12 Mar 2020 22:58:26 +0800

Signed-off-by: LIU Zhiwei <address@hidden>
---
 target/riscv/helper.h                   |  22 +++
 target/riscv/insn32.decode              |   7 +
 target/riscv/insn_trans/trans_rvv.inc.c |   9 ++
 target/riscv/vector_helper.c            | 180 ++++++++++++++++++++++++
 4 files changed, 218 insertions(+)

diff --git a/target/riscv/helper.h b/target/riscv/helper.h
index 333eccca57..74c1c695e0 100644
--- a/target/riscv/helper.h
+++ b/target/riscv/helper.h
@@ -733,3 +733,25 @@ DEF_HELPER_6(vsmul_vx_b, void, ptr, ptr, tl, ptr, env, i32)
 DEF_HELPER_6(vsmul_vx_h, void, ptr, ptr, tl, ptr, env, i32)
 DEF_HELPER_6(vsmul_vx_w, void, ptr, ptr, tl, ptr, env, i32)
 DEF_HELPER_6(vsmul_vx_d, void, ptr, ptr, tl, ptr, env, i32)
+
+DEF_HELPER_6(vwsmaccu_vv_b, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccu_vv_h, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccu_vv_w, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vv_b, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vv_h, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vv_w, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vv_b, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vv_h, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vv_w, void, ptr, ptr, ptr, ptr, env, i32)
+DEF_HELPER_6(vwsmaccu_vx_b, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccu_vx_h, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccu_vx_w, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vx_b, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vx_h, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmacc_vx_w, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vx_b, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vx_h, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccsu_vx_w, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccus_vx_b, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccus_vx_h, void, ptr, ptr, tl, ptr, env, i32)
+DEF_HELPER_6(vwsmaccus_vx_w, void, ptr, ptr, tl, ptr, env, i32)
diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
index 99f70924d6..8798919d3e 100644
--- a/target/riscv/insn32.decode
+++ b/target/riscv/insn32.decode
@@ -419,6 +419,13 @@ vasub_vv        100110 . ..... ..... 000 ..... 1010111 
@r_vm
 vasub_vx        100110 . ..... ..... 100 ..... 1010111 @r_vm
 vsmul_vv        100111 . ..... ..... 000 ..... 1010111 @r_vm
 vsmul_vx        100111 . ..... ..... 100 ..... 1010111 @r_vm
+vwsmaccu_vv     111100 . ..... ..... 000 ..... 1010111 @r_vm
+vwsmaccu_vx     111100 . ..... ..... 100 ..... 1010111 @r_vm
+vwsmacc_vv      111101 . ..... ..... 000 ..... 1010111 @r_vm
+vwsmacc_vx      111101 . ..... ..... 100 ..... 1010111 @r_vm
+vwsmaccsu_vv    111110 . ..... ..... 000 ..... 1010111 @r_vm
+vwsmaccsu_vx    111110 . ..... ..... 100 ..... 1010111 @r_vm
+vwsmaccus_vx    111111 . ..... ..... 100 ..... 1010111 @r_vm
 
 vsetvli         0 ........... ..... 111 ..... 1010111  @r2_zimm
 vsetvl          1000000 ..... ..... 111 ..... 1010111  @r
diff --git a/target/riscv/insn_trans/trans_rvv.inc.c 
b/target/riscv/insn_trans/trans_rvv.inc.c
index 60e1e63b7b..68bebd3c37 100644
--- a/target/riscv/insn_trans/trans_rvv.inc.c
+++ b/target/riscv/insn_trans/trans_rvv.inc.c
@@ -1532,3 +1532,12 @@ GEN_OPIVI_TRANS(vaadd_vi, 0, vaadd_vx, opivx_check)
 /* Vector Single-Width Fractional Multiply with Rounding and Saturation */
 GEN_OPIVV_TRANS(vsmul_vv, opivv_check)
 GEN_OPIVX_TRANS(vsmul_vx,  opivx_check)
+
+/* Vector Widening Saturating Scaled Multiply-Add */
+GEN_OPIVV_WIDEN_TRANS(vwsmaccu_vv, opivv_widen_check)
+GEN_OPIVV_WIDEN_TRANS(vwsmacc_vv, opivv_widen_check)
+GEN_OPIVV_WIDEN_TRANS(vwsmaccsu_vv, opivv_widen_check)
+GEN_OPIVX_WIDEN_TRANS(vwsmaccu_vx)
+GEN_OPIVX_WIDEN_TRANS(vwsmacc_vx)
+GEN_OPIVX_WIDEN_TRANS(vwsmaccsu_vx)
+GEN_OPIVX_WIDEN_TRANS(vwsmaccus_vx)
diff --git a/target/riscv/vector_helper.c b/target/riscv/vector_helper.c
index 74ad07743c..90c19577fa 100644
--- a/target/riscv/vector_helper.c
+++ b/target/riscv/vector_helper.c
@@ -2523,3 +2523,183 @@ GEN_VEXT_VX_ENV(vsmul_vx_b, 1, 1, clearb)
 GEN_VEXT_VX_ENV(vsmul_vx_h, 2, 2, clearh)
 GEN_VEXT_VX_ENV(vsmul_vx_w, 4, 4, clearl)
 GEN_VEXT_VX_ENV(vsmul_vx_d, 8, 8, clearq)
+
+/* Vector Widening Saturating Scaled Multiply-Add */
+static uint16_t vwsmaccu8(CPURISCVState *env, uint8_t a, uint8_t b,
+    uint16_t c)
+{
+    uint8_t round;
+    uint16_t res = (uint16_t)a * (uint16_t)b;
+
+    round = get_round(env, res, 4);
+    res   = (res >> 4) + round;
+    return saddu16(env, c, res);
+}
+static uint32_t vwsmaccu16(CPURISCVState *env, uint16_t a, uint16_t b,
+    uint32_t c)
+{
+    uint8_t round;
+    uint32_t res = (uint32_t)a * (uint32_t)b;
+
+    round = get_round(env, res, 8);
+    res   = (res >> 8) + round;
+    return saddu32(env, c, res);
+}
+static uint64_t vwsmaccu32(CPURISCVState *env, uint32_t a, uint32_t b,
+    uint64_t c)
+{
+    uint8_t round;
+    uint64_t res = (uint64_t)a * (uint64_t)b;
+
+    round = get_round(env, res, 16);
+    res   = (res >> 16) + round;
+    return saddu64(env, c, res);
+}
+
+#define OPIVV3_ENV(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)   \
+static void do_##NAME(void *vd, void *vs1, void *vs2, int i,       \
+        CPURISCVState *env)                                        \
+{                                                                  \
+    TX1 s1 = *((T1 *)vs1 + HS1(i));                                \
+    TX2 s2 = *((T2 *)vs2 + HS2(i));                                \
+    TD d = *((TD *)vd + HD(i));                                    \
+    *((TD *)vd + HD(i)) = OP(env, s2, s1, d);                      \
+}
+RVVCALL(OPIVV3_ENV, vwsmaccu_vv_b, WOP_UUU_B, H2, H1, H1, vwsmaccu8)
+RVVCALL(OPIVV3_ENV, vwsmaccu_vv_h, WOP_UUU_H, H4, H2, H2, vwsmaccu16)
+RVVCALL(OPIVV3_ENV, vwsmaccu_vv_w, WOP_UUU_W, H8, H4, H4, vwsmaccu32)
+GEN_VEXT_VV_ENV(vwsmaccu_vv_b, 1, 2, clearh)
+GEN_VEXT_VV_ENV(vwsmaccu_vv_h, 2, 4, clearl)
+GEN_VEXT_VV_ENV(vwsmaccu_vv_w, 4, 8, clearq)
+
+#define OPIVX3_ENV(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)            \
+static void do_##NAME(void *vd, target_ulong s1, void *vs2, int i,     \
+        CPURISCVState *env)                                            \
+{                                                                      \
+    TX2 s2 = *((T2 *)vs2 + HS2(i));                                    \
+    TD d = *((TD *)vd + HD(i));                                        \
+    *((TD *)vd + HD(i)) = OP(env, s2, (TX1)(T1)(target_long)s1, d);    \
+}
+RVVCALL(OPIVX3_ENV, vwsmaccu_vx_b, WOP_UUU_B, H2, H1, vwsmaccu8)
+RVVCALL(OPIVX3_ENV, vwsmaccu_vx_h, WOP_UUU_H, H4, H2, vwsmaccu16)
+RVVCALL(OPIVX3_ENV, vwsmaccu_vx_w, WOP_UUU_W, H8, H4, vwsmaccu32)
+GEN_VEXT_VX_ENV(vwsmaccu_vx_b, 1, 2, clearh)
+GEN_VEXT_VX_ENV(vwsmaccu_vx_h, 2, 4, clearl)
+GEN_VEXT_VX_ENV(vwsmaccu_vx_w, 4, 8, clearq)
+
+static int16_t vwsmacc8(CPURISCVState *env, int8_t a, int8_t b, int16_t c)
+{
+    uint8_t round;
+    int16_t res = (int16_t)a * (int16_t)b;
+
+    round = get_round(env, res, 4);
+    res   = (res >> 4) + round;
+    return sadd16(env, c, res);
+}
+static int32_t vwsmacc16(CPURISCVState *env, int16_t a, int16_t b, int32_t c)
+{
+    uint8_t round;
+    int32_t res = (int32_t)a * (int32_t)b;
+
+    round = get_round(env, res, 8);
+    res   = (res >> 8) + round;
+    return sadd32(env, c, res);
+
+}
+static int64_t vwsmacc32(CPURISCVState *env, int32_t a, int32_t b, int64_t c)
+{
+    uint8_t round;
+    int64_t res = (int64_t)a * (int64_t)b;
+
+    round = get_round(env, res, 16);
+    res   = (res >> 16) + round;
+    return sadd64(env, c, res);
+}
+RVVCALL(OPIVV3_ENV, vwsmacc_vv_b, WOP_SSS_B, H2, H1, H1, vwsmacc8)
+RVVCALL(OPIVV3_ENV, vwsmacc_vv_h, WOP_SSS_H, H4, H2, H2, vwsmacc16)
+RVVCALL(OPIVV3_ENV, vwsmacc_vv_w, WOP_SSS_W, H8, H4, H4, vwsmacc32)
+GEN_VEXT_VV_ENV(vwsmacc_vv_b, 1, 2, clearh)
+GEN_VEXT_VV_ENV(vwsmacc_vv_h, 2, 4, clearl)
+GEN_VEXT_VV_ENV(vwsmacc_vv_w, 4, 8, clearq)
+RVVCALL(OPIVX3_ENV, vwsmacc_vx_b, WOP_SSS_B, H2, H1, vwsmacc8)
+RVVCALL(OPIVX3_ENV, vwsmacc_vx_h, WOP_SSS_H, H4, H2, vwsmacc16)
+RVVCALL(OPIVX3_ENV, vwsmacc_vx_w, WOP_SSS_W, H8, H4, vwsmacc32)
+GEN_VEXT_VX_ENV(vwsmacc_vx_b, 1, 2, clearh)
+GEN_VEXT_VX_ENV(vwsmacc_vx_h, 2, 4, clearl)
+GEN_VEXT_VX_ENV(vwsmacc_vx_w, 4, 8, clearq)
+
+static int16_t vwsmaccsu8(CPURISCVState *env, uint8_t a, int8_t b, int16_t c)
+{
+    uint8_t round;
+    int16_t res = (uint16_t)a * (int16_t)b;
+
+    round = get_round(env, res, 4);
+    res   = (res >> 4) + round;
+    return ssub16(env, c, res);
+}
+static int32_t vwsmaccsu16(CPURISCVState *env, uint16_t a, int16_t b,
+    uint32_t c)
+{
+    uint8_t round;
+    int32_t res = (uint32_t)a * (int32_t)b;
+
+    round = get_round(env, res, 8);
+    res   = (res >> 8) + round;
+    return ssub32(env, c, res);
+}
+static int64_t vwsmaccsu32(CPURISCVState *env, uint32_t a, int32_t b,
+    int64_t c)
+{
+    uint8_t round;
+    int64_t res = (uint64_t)a * (int64_t)b;
+
+    round = get_round(env, res, 16);
+    res   = (res >> 16) + round;
+    return ssub64(env, c, res);
+}
+RVVCALL(OPIVV3_ENV, vwsmaccsu_vv_b, WOP_SSU_B, H2, H1, H1, vwsmaccsu8)
+RVVCALL(OPIVV3_ENV, vwsmaccsu_vv_h, WOP_SSU_H, H4, H2, H2, vwsmaccsu16)
+RVVCALL(OPIVV3_ENV, vwsmaccsu_vv_w, WOP_SSU_W, H8, H4, H4, vwsmaccsu32)
+GEN_VEXT_VV_ENV(vwsmaccsu_vv_b, 1, 2, clearh)
+GEN_VEXT_VV_ENV(vwsmaccsu_vv_h, 2, 4, clearl)
+GEN_VEXT_VV_ENV(vwsmaccsu_vv_w, 4, 8, clearq)
+RVVCALL(OPIVX3_ENV, vwsmaccsu_vx_b, WOP_SSU_B, H2, H1, vwsmaccsu8)
+RVVCALL(OPIVX3_ENV, vwsmaccsu_vx_h, WOP_SSU_H, H4, H2, vwsmaccsu16)
+RVVCALL(OPIVX3_ENV, vwsmaccsu_vx_w, WOP_SSU_W, H8, H4, vwsmaccsu32)
+GEN_VEXT_VX_ENV(vwsmaccsu_vx_b, 1, 2, clearh)
+GEN_VEXT_VX_ENV(vwsmaccsu_vx_h, 2, 4, clearl)
+GEN_VEXT_VX_ENV(vwsmaccsu_vx_w, 4, 8, clearq)
+
+static int16_t vwsmaccus8(CPURISCVState *env, int8_t a, uint8_t b, int16_t c)
+{
+    uint8_t round;
+    int16_t res = (int16_t)a * (uint16_t)b;
+
+    round = get_round(env, res, 4);
+    res   = (res >> 4) + round;
+    return ssub16(env, c, res);
+}
+static int32_t vwsmaccus16(CPURISCVState *env, int16_t a, uint16_t b, int32_t 
c)
+{
+    uint8_t round;
+    int32_t res = (int32_t)a * (uint32_t)b;
+
+    round = get_round(env, res, 8);
+    res   = (res >> 8) + round;
+    return ssub32(env, c, res);
+}
+static int64_t vwsmaccus32(CPURISCVState *env, int32_t a, uint32_t b, int64_t 
c)
+{
+    uint8_t round;
+    int64_t res = (int64_t)a * (uint64_t)b;
+
+    round = get_round(env, res, 16);
+    res   = (res >> 16) + round;
+    return ssub64(env, c, res);
+}
+RVVCALL(OPIVX3_ENV, vwsmaccus_vx_b, WOP_SUS_B, H2, H1, vwsmaccus8)
+RVVCALL(OPIVX3_ENV, vwsmaccus_vx_h, WOP_SUS_H, H4, H2, vwsmaccus16)
+RVVCALL(OPIVX3_ENV, vwsmaccus_vx_w, WOP_SUS_W, H8, H4, vwsmaccus32)
+GEN_VEXT_VX_ENV(vwsmaccus_vx_b, 1, 2, clearh)
+GEN_VEXT_VX_ENV(vwsmaccus_vx_h, 2, 4, clearl)
+GEN_VEXT_VX_ENV(vwsmaccus_vx_w, 4, 8, clearq)
-- 
2.23.0




reply via email to

[Prev in Thread] Current Thread [Next in Thread]