tcg: Add generic helpers for saturating arithmetic

No vector ops as yet.  SSE only has direct support for 8- and 16-bit
saturation; handling 32- and 64-bit saturation is much more expensive.

Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

4 files changed, 390 insertions, 0 deletions

diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c
index 59d7a0a..e6f99ba 100644
--- a/accel/tcg/tcg-runtime-gvec.c
+++ b/accel/tcg/tcg-runtime-gvec.c
@@ -547,3 +547,271 @@ DO_CMP2(64)
 #undef DO_CMP0
 #undef DO_CMP1
 #undef DO_CMP2
+
+void HELPER(gvec_ssadd8)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int8_t)) {
+        int r = *(int8_t *)(a + i) + *(int8_t *)(b + i);
+        if (r > INT8_MAX) {
+            r = INT8_MAX;
+        } else if (r < INT8_MIN) {
+            r = INT8_MIN;
+        }
+        *(int8_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ssadd16)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int16_t)) {
+        int r = *(int16_t *)(a + i) + *(int16_t *)(b + i);
+        if (r > INT16_MAX) {
+            r = INT16_MAX;
+        } else if (r < INT16_MIN) {
+            r = INT16_MIN;
+        }
+        *(int16_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ssadd32)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int32_t)) {
+        int32_t ai = *(int32_t *)(a + i);
+        int32_t bi = *(int32_t *)(b + i);
+        int32_t di = ai + bi;
+        if (((di ^ ai) &~ (ai ^ bi)) < 0) {
+            /* Signed overflow.  */
+            di = (di < 0 ? INT32_MAX : INT32_MIN);
+        }
+        *(int32_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ssadd64)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int64_t)) {
+        int64_t ai = *(int64_t *)(a + i);
+        int64_t bi = *(int64_t *)(b + i);
+        int64_t di = ai + bi;
+        if (((di ^ ai) &~ (ai ^ bi)) < 0) {
+            /* Signed overflow.  */
+            di = (di < 0 ? INT64_MAX : INT64_MIN);
+        }
+        *(int64_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_sssub8)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
+        int r = *(int8_t *)(a + i) - *(int8_t *)(b + i);
+        if (r > INT8_MAX) {
+            r = INT8_MAX;
+        } else if (r < INT8_MIN) {
+            r = INT8_MIN;
+        }
+        *(uint8_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_sssub16)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int16_t)) {
+        int r = *(int16_t *)(a + i) - *(int16_t *)(b + i);
+        if (r > INT16_MAX) {
+            r = INT16_MAX;
+        } else if (r < INT16_MIN) {
+            r = INT16_MIN;
+        }
+        *(int16_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_sssub32)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int32_t)) {
+        int32_t ai = *(int32_t *)(a + i);
+        int32_t bi = *(int32_t *)(b + i);
+        int32_t di = ai - bi;
+        if (((di ^ ai) & (ai ^ bi)) < 0) {
+            /* Signed overflow.  */
+            di = (di < 0 ? INT32_MAX : INT32_MIN);
+        }
+        *(int32_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_sssub64)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(int64_t)) {
+        int64_t ai = *(int64_t *)(a + i);
+        int64_t bi = *(int64_t *)(b + i);
+        int64_t di = ai - bi;
+        if (((di ^ ai) & (ai ^ bi)) < 0) {
+            /* Signed overflow.  */
+            di = (di < 0 ? INT64_MAX : INT64_MIN);
+        }
+        *(int64_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_usadd8)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
+        unsigned r = *(uint8_t *)(a + i) + *(uint8_t *)(b + i);
+        if (r > UINT8_MAX) {
+            r = UINT8_MAX;
+        }
+        *(uint8_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_usadd16)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
+        unsigned r = *(uint16_t *)(a + i) + *(uint16_t *)(b + i);
+        if (r > UINT16_MAX) {
+            r = UINT16_MAX;
+        }
+        *(uint16_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_usadd32)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
+        uint32_t ai = *(uint32_t *)(a + i);
+        uint32_t bi = *(uint32_t *)(b + i);
+        uint32_t di = ai + bi;
+        if (di < ai) {
+            di = UINT32_MAX;
+        }
+        *(uint32_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_usadd64)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
+        uint64_t ai = *(uint64_t *)(a + i);
+        uint64_t bi = *(uint64_t *)(b + i);
+        uint64_t di = ai + bi;
+        if (di < ai) {
+            di = UINT64_MAX;
+        }
+        *(uint64_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ussub8)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
+        int r = *(uint8_t *)(a + i) - *(uint8_t *)(b + i);
+        if (r < 0) {
+            r = 0;
+        }
+        *(uint8_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ussub16)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
+        int r = *(uint16_t *)(a + i) - *(uint16_t *)(b + i);
+        if (r < 0) {
+            r = 0;
+        }
+        *(uint16_t *)(d + i) = r;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ussub32)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
+        uint32_t ai = *(uint32_t *)(a + i);
+        uint32_t bi = *(uint32_t *)(b + i);
+        uint32_t di = ai - bi;
+        if (ai < bi) {
+            di = 0;
+        }
+        *(uint32_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
+
+void HELPER(gvec_ussub64)(void *d, void *a, void *b, uint32_t desc)
+{
+    intptr_t oprsz = simd_oprsz(desc);
+    intptr_t i;
+
+    for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
+        uint64_t ai = *(uint64_t *)(a + i);
+        uint64_t bi = *(uint64_t *)(b + i);
+        uint64_t di = ai - bi;
+        if (ai < bi) {
+            di = 0;
+        }
+        *(uint64_t *)(d + i) = di;
+    }
+    clear_high(d, oprsz, desc);
+}
diff --git a/accel/tcg/tcg-runtime.h b/accel/tcg/tcg-runtime.h
index 54f7e78..f224a97 100644
--- a/accel/tcg/tcg-runtime.h
+++ b/accel/tcg/tcg-runtime.h
@@ -157,6 +157,26 @@ DEF_HELPER_FLAGS_4(gvec_mul16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(gvec_mul32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 DEF_HELPER_FLAGS_4(gvec_mul64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
 
+DEF_HELPER_FLAGS_4(gvec_ssadd8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ssadd16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ssadd32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ssadd64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_sssub8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sssub16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sssub32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_sssub64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_usadd8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_usadd16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_usadd32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_usadd64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
+DEF_HELPER_FLAGS_4(gvec_ussub8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ussub16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ussub32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+DEF_HELPER_FLAGS_4(gvec_ussub64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
+
 DEF_HELPER_FLAGS_3(gvec_neg8, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
 DEF_HELPER_FLAGS_3(gvec_neg16, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
 DEF_HELPER_FLAGS_3(gvec_neg32, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c
index 49c77e3..843bb5b 100644
--- a/tcg/tcg-op-gvec.c
+++ b/tcg/tcg-op-gvec.c
@@ -1309,6 +1309,98 @@ void tcg_gen_gvec_mul(unsigned vece, uint32_t dofs, uint32_t aofs,
     tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
 }
 
+void tcg_gen_gvec_ssadd(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const GVecGen3 g[4] = {
+        { .fno = gen_helper_gvec_ssadd8, .vece = MO_8 },
+        { .fno = gen_helper_gvec_ssadd16, .vece = MO_16 },
+        { .fno = gen_helper_gvec_ssadd32, .vece = MO_32 },
+        { .fno = gen_helper_gvec_ssadd64, .vece = MO_64 }
+    };
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
+void tcg_gen_gvec_sssub(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const GVecGen3 g[4] = {
+        { .fno = gen_helper_gvec_sssub8, .vece = MO_8 },
+        { .fno = gen_helper_gvec_sssub16, .vece = MO_16 },
+        { .fno = gen_helper_gvec_sssub32, .vece = MO_32 },
+        { .fno = gen_helper_gvec_sssub64, .vece = MO_64 }
+    };
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
+static void tcg_gen_vec_usadd32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 max = tcg_const_i32(-1);
+    tcg_gen_add_i32(d, a, b);
+    tcg_gen_movcond_i32(TCG_COND_LTU, d, d, a, max, d);
+    tcg_temp_free_i32(max);
+}
+
+static void tcg_gen_vec_usadd32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 max = tcg_const_i64(-1);
+    tcg_gen_add_i64(d, a, b);
+    tcg_gen_movcond_i64(TCG_COND_LTU, d, d, a, max, d);
+    tcg_temp_free_i64(max);
+}
+
+void tcg_gen_gvec_usadd(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const GVecGen3 g[4] = {
+        { .fno = gen_helper_gvec_usadd8, .vece = MO_8 },
+        { .fno = gen_helper_gvec_usadd16, .vece = MO_16 },
+        { .fni4 = tcg_gen_vec_usadd32_i32,
+          .fno = gen_helper_gvec_usadd32,
+          .vece = MO_32 },
+        { .fni8 = tcg_gen_vec_usadd32_i64,
+          .fno = gen_helper_gvec_usadd64,
+          .vece = MO_64 }
+    };
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
+static void tcg_gen_vec_ussub32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
+{
+    TCGv_i32 min = tcg_const_i32(0);
+    tcg_gen_sub_i32(d, a, b);
+    tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, min, d);
+    tcg_temp_free_i32(min);
+}
+
+static void tcg_gen_vec_ussub32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
+{
+    TCGv_i64 min = tcg_const_i64(0);
+    tcg_gen_sub_i64(d, a, b);
+    tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, min, d);
+    tcg_temp_free_i64(min);
+}
+
+void tcg_gen_gvec_ussub(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
+{
+    static const GVecGen3 g[4] = {
+        { .fno = gen_helper_gvec_ussub8, .vece = MO_8 },
+        { .fno = gen_helper_gvec_ussub16, .vece = MO_16 },
+        { .fni4 = tcg_gen_vec_ussub32_i32,
+          .fno = gen_helper_gvec_ussub32,
+          .vece = MO_32 },
+        { .fni8 = tcg_gen_vec_ussub32_i64,
+          .fno = gen_helper_gvec_ussub64,
+          .vece = MO_64 }
+    };
+    tcg_debug_assert(vece <= MO_64);
+    tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
+}
+
 /* Perform a vector negation using normal negation and a mask.
    Compare gen_subv_mask above.  */
 static void gen_negv_mask(TCGv_i64 d, TCGv_i64 b, TCGv_i64 m)
diff --git a/tcg/tcg-op-gvec.h b/tcg/tcg-op-gvec.h
index abe909d..03ced44 100644
--- a/tcg/tcg-op-gvec.h
+++ b/tcg/tcg-op-gvec.h
@@ -179,6 +179,16 @@ void tcg_gen_gvec_sub(unsigned vece, uint32_t dofs, uint32_t aofs,
 void tcg_gen_gvec_mul(unsigned vece, uint32_t dofs, uint32_t aofs,
                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
 
+/* Saturated arithmetic.  */
+void tcg_gen_gvec_ssadd(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_sssub(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_usadd(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+void tcg_gen_gvec_ussub(unsigned vece, uint32_t dofs, uint32_t aofs,
+                        uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
+
 void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs,
                       uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
 void tcg_gen_gvec_or(unsigned vece, uint32_t dofs, uint32_t aofs,