[PATCH v9 69/76] target/riscv: rvv-1.0: floating-point reciprocal square-root estimate instruction

From: frank.chang@sifive.com
To: qemu-devel@nongnu.org
Cc: qemu-riscv@nongnu.org, Frank Chang <frank.chang@sifive.com>,
	Bin Meng <bin.meng@windriver.com>,
	Richard Henderson <richard.henderson@linaro.org>,
	Palmer Dabbelt <palmer@dabbelt.com>,
	Alistair Francis <alistair.francis@wdc.com>,
	LIU Zhiwei <zhiwei_liu@c-sky.com>
Subject: [PATCH v9 69/76] target/riscv: rvv-1.0: floating-point reciprocal square-root estimate instruction
Date: Fri, 29 Oct 2021 16:59:14 +0800	[thread overview]
Message-ID: <20211029085922.255197-70-frank.chang@sifive.com> (raw)
In-Reply-To: <20211029085922.255197-1-frank.chang@sifive.com>

From: Frank Chang <frank.chang@sifive.com>

Implement the floating-point reciprocal square-root estimate to 7 bits
instruction.

Signed-off-by: Frank Chang <frank.chang@sifive.com>
---
 target/riscv/helper.h                   |   4 +
 target/riscv/insn32.decode              |   1 +
 target/riscv/insn_trans/trans_rvv.c.inc |   1 +
 target/riscv/vector_helper.c            | 183 ++++++++++++++++++++++++
 4 files changed, 189 insertions(+)

diff --git a/target/riscv/helper.h b/target/riscv/helper.h
index a717a87a0e0..bdf06dfb24d 100644
--- a/target/riscv/helper.h
+++ b/target/riscv/helper.h
@@ -841,6 +841,10 @@ DEF_HELPER_5(vfsqrt_v_h, void, ptr, ptr, ptr, env, i32)
 DEF_HELPER_5(vfsqrt_v_w, void, ptr, ptr, ptr, env, i32)
 DEF_HELPER_5(vfsqrt_v_d, void, ptr, ptr, ptr, env, i32)
 
+DEF_HELPER_5(vfrsqrt7_v_h, void, ptr, ptr, ptr, env, i32)
+DEF_HELPER_5(vfrsqrt7_v_w, void, ptr, ptr, ptr, env, i32)
+DEF_HELPER_5(vfrsqrt7_v_d, void, ptr, ptr, ptr, env, i32)
+
 DEF_HELPER_6(vfmin_vv_h, void, ptr, ptr, ptr, ptr, env, i32)
 DEF_HELPER_6(vfmin_vv_w, void, ptr, ptr, ptr, ptr, env, i32)
 DEF_HELPER_6(vfmin_vv_d, void, ptr, ptr, ptr, ptr, env, i32)
diff --git a/target/riscv/insn32.decode b/target/riscv/insn32.decode
index c4fdc76a269..6e5f288943a 100644
--- a/target/riscv/insn32.decode
+++ b/target/riscv/insn32.decode
@@ -560,6 +560,7 @@ vfwmsac_vf      111110 . ..... ..... 101 ..... 1010111 @r_vm
 vfwnmsac_vv     111111 . ..... ..... 001 ..... 1010111 @r_vm
 vfwnmsac_vf     111111 . ..... ..... 101 ..... 1010111 @r_vm
 vfsqrt_v        010011 . ..... 00000 001 ..... 1010111 @r2_vm
+vfrsqrt7_v      010011 . ..... 00100 001 ..... 1010111 @r2_vm
 vfmin_vv        000100 . ..... ..... 001 ..... 1010111 @r_vm
 vfmin_vf        000100 . ..... ..... 101 ..... 1010111 @r_vm
 vfmax_vv        000110 . ..... ..... 001 ..... 1010111 @r_vm
diff --git a/target/riscv/insn_trans/trans_rvv.c.inc b/target/riscv/insn_trans/trans_rvv.c.inc
index 53c8573f117..8fe718610a9 100644
--- a/target/riscv/insn_trans/trans_rvv.c.inc
+++ b/target/riscv/insn_trans/trans_rvv.c.inc
@@ -2407,6 +2407,7 @@ static bool trans_##NAME(DisasContext *s, arg_rmr *a)  \
 }
 
 GEN_OPFV_TRANS(vfsqrt_v, opfv_check, RISCV_FRM_DYN)
+GEN_OPFV_TRANS(vfrsqrt7_v, opfv_check, RISCV_FRM_DYN)
 
 /* Vector Floating-Point MIN/MAX Instructions */
 GEN_OPFVV_TRANS(vfmin_vv, opfvv_check)
diff --git a/target/riscv/vector_helper.c b/target/riscv/vector_helper.c
index 22848d6b683..d5f3229bcb4 100644
--- a/target/riscv/vector_helper.c
+++ b/target/riscv/vector_helper.c
@@ -18,6 +18,7 @@
 
 #include "qemu/osdep.h"
 #include "qemu/host-utils.h"
+#include "qemu/bitops.h"
 #include "cpu.h"
 #include "exec/memop.h"
 #include "exec/exec-all.h"
@@ -3404,6 +3405,188 @@ GEN_VEXT_V_ENV(vfsqrt_v_h, 2, 2)
 GEN_VEXT_V_ENV(vfsqrt_v_w, 4, 4)
 GEN_VEXT_V_ENV(vfsqrt_v_d, 8, 8)
 
+/*
+ * Vector Floating-Point Reciprocal Square-Root Estimate Instruction
+ *
+ * Adapted from riscv-v-spec recip.c:
+ * https://github.com/riscv/riscv-v-spec/blob/master/recip.c
+ */
+static uint64_t frsqrt7(uint64_t f, int exp_size, int frac_size)
+{
+    uint64_t sign = extract64(f, frac_size + exp_size, 1);
+    uint64_t exp = extract64(f, frac_size, exp_size);
+    uint64_t frac = extract64(f, 0, frac_size);
+
+    const uint8_t lookup_table[] = {
+        52, 51, 50, 48, 47, 46, 44, 43,
+        42, 41, 40, 39, 38, 36, 35, 34,
+        33, 32, 31, 30, 30, 29, 28, 27,
+        26, 25, 24, 23, 23, 22, 21, 20,
+        19, 19, 18, 17, 16, 16, 15, 14,
+        14, 13, 12, 12, 11, 10, 10, 9,
+        9, 8, 7, 7, 6, 6, 5, 4,
+        4, 3, 3, 2, 2, 1, 1, 0,
+        127, 125, 123, 121, 119, 118, 116, 114,
+        113, 111, 109, 108, 106, 105, 103, 102,
+        100, 99, 97, 96, 95, 93, 92, 91,
+        90, 88, 87, 86, 85, 84, 83, 82,
+        80, 79, 78, 77, 76, 75, 74, 73,
+        72, 71, 70, 70, 69, 68, 67, 66,
+        65, 64, 63, 63, 62, 61, 60, 59,
+        59, 58, 57, 56, 56, 55, 54, 53
+    };
+    const int precision = 7;
+
+    if (exp == 0 && frac != 0) { /* subnormal */
+        /* Normalize the subnormal. */
+        while (extract64(frac, frac_size - 1, 1) == 0) {
+            exp--;
+            frac <<= 1;
+        }
+
+        frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size);
+    }
+
+    int idx = ((exp & 1) << (precision - 1)) |
+                (frac >> (frac_size - precision + 1));
+    uint64_t out_frac = (uint64_t)(lookup_table[idx]) <<
+                            (frac_size - precision);
+    uint64_t out_exp = (3 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp) / 2;
+
+    uint64_t val = 0;
+    val = deposit64(val, 0, frac_size, out_frac);
+    val = deposit64(val, frac_size, exp_size, out_exp);
+    val = deposit64(val, frac_size + exp_size, 1, sign);
+    return val;
+}
+
+static float16 frsqrt7_h(float16 f, float_status *s)
+{
+    int exp_size = 5, frac_size = 10;
+    bool sign = float16_is_neg(f);
+
+    /*
+     * frsqrt7(sNaN) = canonical NaN
+     * frsqrt7(-inf) = canonical NaN
+     * frsqrt7(-normal) = canonical NaN
+     * frsqrt7(-subnormal) = canonical NaN
+     */
+    if (float16_is_signaling_nan(f, s) ||
+            (float16_is_infinity(f) && sign) ||
+            (float16_is_normal(f) && sign) ||
+            (float16_is_zero_or_denormal(f) && !float16_is_zero(f) && sign)) {
+        s->float_exception_flags |= float_flag_invalid;
+        return float16_default_nan(s);
+    }
+
+    /* frsqrt7(qNaN) = canonical NaN */
+    if (float16_is_quiet_nan(f, s)) {
+        return float16_default_nan(s);
+    }
+
+    /* frsqrt7(+-0) = +-inf */
+    if (float16_is_zero(f)) {
+        s->float_exception_flags |= float_flag_divbyzero;
+        return float16_set_sign(float16_infinity, sign);
+    }
+
+    /* frsqrt7(+inf) = +0 */
+    if (float16_is_infinity(f) && !sign) {
+        return float16_set_sign(float16_zero, sign);
+    }
+
+    /* +normal, +subnormal */
+    uint64_t val = frsqrt7(f, exp_size, frac_size);
+    return make_float16(val);
+}
+
+static float32 frsqrt7_s(float32 f, float_status *s)
+{
+    int exp_size = 8, frac_size = 23;
+    bool sign = float32_is_neg(f);
+
+    /*
+     * frsqrt7(sNaN) = canonical NaN
+     * frsqrt7(-inf) = canonical NaN
+     * frsqrt7(-normal) = canonical NaN
+     * frsqrt7(-subnormal) = canonical NaN
+     */
+    if (float32_is_signaling_nan(f, s) ||
+            (float32_is_infinity(f) && sign) ||
+            (float32_is_normal(f) && sign) ||
+            (float32_is_zero_or_denormal(f) && !float32_is_zero(f) && sign)) {
+        s->float_exception_flags |= float_flag_invalid;
+        return float32_default_nan(s);
+    }
+
+    /* frsqrt7(qNaN) = canonical NaN */
+    if (float32_is_quiet_nan(f, s)) {
+        return float32_default_nan(s);
+    }
+
+    /* frsqrt7(+-0) = +-inf */
+    if (float32_is_zero(f)) {
+        s->float_exception_flags |= float_flag_divbyzero;
+        return float32_set_sign(float32_infinity, sign);
+    }
+
+    /* frsqrt7(+inf) = +0 */
+    if (float32_is_infinity(f) && !sign) {
+        return float32_set_sign(float32_zero, sign);
+    }
+
+    /* +normal, +subnormal */
+    uint64_t val = frsqrt7(f, exp_size, frac_size);
+    return make_float32(val);
+}
+
+static float64 frsqrt7_d(float64 f, float_status *s)
+{
+    int exp_size = 11, frac_size = 52;
+    bool sign = float64_is_neg(f);
+
+    /*
+     * frsqrt7(sNaN) = canonical NaN
+     * frsqrt7(-inf) = canonical NaN
+     * frsqrt7(-normal) = canonical NaN
+     * frsqrt7(-subnormal) = canonical NaN
+     */
+    if (float64_is_signaling_nan(f, s) ||
+            (float64_is_infinity(f) && sign) ||
+            (float64_is_normal(f) && sign) ||
+            (float64_is_zero_or_denormal(f) && !float64_is_zero(f) && sign)) {
+        s->float_exception_flags |= float_flag_invalid;
+        return float64_default_nan(s);
+    }
+
+    /* frsqrt7(qNaN) = canonical NaN */
+    if (float64_is_quiet_nan(f, s)) {
+        return float64_default_nan(s);
+    }
+
+    /* frsqrt7(+-0) = +-inf */
+    if (float64_is_zero(f)) {
+        s->float_exception_flags |= float_flag_divbyzero;
+        return float64_set_sign(float64_infinity, sign);
+    }
+
+    /* frsqrt7(+inf) = +0 */
+    if (float64_is_infinity(f) && !sign) {
+        return float64_set_sign(float64_zero, sign);
+    }
+
+    /* +normal, +subnormal */
+    uint64_t val = frsqrt7(f, exp_size, frac_size);
+    return make_float64(val);
+}
+
+RVVCALL(OPFVV1, vfrsqrt7_v_h, OP_UU_H, H2, H2, frsqrt7_h)
+RVVCALL(OPFVV1, vfrsqrt7_v_w, OP_UU_W, H4, H4, frsqrt7_s)
+RVVCALL(OPFVV1, vfrsqrt7_v_d, OP_UU_D, H8, H8, frsqrt7_d)
+GEN_VEXT_V_ENV(vfrsqrt7_v_h, 2, 2)
+GEN_VEXT_V_ENV(vfrsqrt7_v_w, 4, 4)
+GEN_VEXT_V_ENV(vfrsqrt7_v_d, 8, 8)
+
 /* Vector Floating-Point MIN/MAX Instructions */
 RVVCALL(OPFVV2, vfmin_vv_h, OP_UUU_H, H2, H2, H2, float16_minimum_number)
 RVVCALL(OPFVV2, vfmin_vv_w, OP_UUU_W, H4, H4, H4, float32_minimum_number)
-- 
2.25.1


