1 | The following changes since commit e18e5501d8ac692d32657a3e1ef545b14e72b730: | 1 | The following changes since commit 40c67636f67c2a89745f2e698522fe917326a952: |
---|---|---|---|
2 | 2 | ||
3 | Merge remote-tracking branch 'remotes/dgilbert-gitlab/tags/pull-virtiofs-20200210' into staging (2020-02-10 18:09:14 +0000) | 3 | Merge remote-tracking branch 'remotes/kraxel/tags/usb-20200317-pull-request' into staging (2020-03-17 14:00:56 +0000) |
4 | 4 | ||
5 | are available in the Git repository at: | 5 | are available in the Git repository at: |
6 | 6 | ||
7 | https://github.com/rth7680/qemu.git tags/pull-tcg-20200212 | 7 | https://github.com/rth7680/qemu.git tags/pull-tcg-20200317 |
8 | 8 | ||
9 | for you to fetch changes up to 2445971604c1cfd3ec484457159f4ac300fb04d2: | 9 | for you to fetch changes up to 0270bd503e3699b7202200a2d693ad1feb57473f: |
10 | 10 | ||
11 | tcg: Add tcg_gen_gvec_5_ptr (2020-02-12 14:58:36 -0800) | 11 | tcg: Remove tcg-runtime-gvec.c DO_CMP0 (2020-03-17 08:41:07 -0700) |
12 | 12 | ||
13 | ---------------------------------------------------------------- | 13 | ---------------------------------------------------------------- |
14 | Fix breakpoint invalidation. | 14 | Fix tcg/i386 bug vs sari_vec. |
15 | Add support for tcg helpers with 7 arguments. | 15 | Fix tcg-runtime-gvec.c vs i386 without avx. |
16 | Add support for gvec helpers with 5 arguments. | ||
17 | 16 | ||
18 | ---------------------------------------------------------------- | 17 | ---------------------------------------------------------------- |
19 | Max Filippov (1): | 18 | Richard Henderson (5): |
20 | exec: flush CPU TB cache in breakpoint_invalidate | 19 | tcg/i386: Bound shift count expanding sari_vec |
20 | tcg: Remove CONFIG_VECTOR16 | ||
21 | tcg: Tidy tcg-runtime-gvec.c types | ||
22 | tcg: Tidy tcg-runtime-gvec.c DUP* | ||
23 | tcg: Remove tcg-runtime-gvec.c DO_CMP0 | ||
21 | 24 | ||
22 | Richard Henderson (1): | 25 | configure | 56 -------- |
23 | tcg: Add tcg_gen_gvec_5_ptr | 26 | accel/tcg/tcg-runtime-gvec.c | 298 +++++++++++++++++-------------------------- |
27 | tcg/i386/tcg-target.inc.c | 9 +- | ||
28 | 3 files changed, 122 insertions(+), 241 deletions(-) | ||
24 | 29 | ||
25 | Taylor Simpson (1): | ||
26 | tcg: Add support for a helper with 7 arguments | ||
27 | |||
28 | include/exec/helper-gen.h | 13 +++++++++++++ | ||
29 | include/exec/helper-head.h | 2 ++ | ||
30 | include/exec/helper-proto.h | 6 ++++++ | ||
31 | include/exec/helper-tcg.h | 7 +++++++ | ||
32 | include/tcg/tcg-op-gvec.h | 7 +++++++ | ||
33 | exec.c | 15 +++++++-------- | ||
34 | tcg/tcg-op-gvec.c | 32 ++++++++++++++++++++++++++++++++ | ||
35 | 7 files changed, 74 insertions(+), 8 deletions(-) | ||
36 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | A given RISU testcase for SVE can produce | ||
1 | 2 | ||
3 | tcg-op-vec.c:511: do_shifti: Assertion `i >= 0 && i < (8 << vece)' failed. | ||
4 | |||
5 | because expand_vec_sari gave a shift count of 32 to a MO_32 | ||
6 | vector shift. | ||
7 | |||
8 | In 44f1441dbe1, we changed from direct expansion of vector opcodes | ||
9 | to re-use of the tcg expanders. So while the comment correctly notes | ||
10 | that the hw will handle such a shift count, we now have to take our | ||
11 | own sanity checks into account. Which is easy in this particular case. | ||
12 | |||
13 | Fixes: 44f1441dbe1 | ||
14 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
15 | --- | ||
16 | tcg/i386/tcg-target.inc.c | 9 ++++++--- | ||
17 | 1 file changed, 6 insertions(+), 3 deletions(-) | ||
18 | |||
19 | diff --git a/tcg/i386/tcg-target.inc.c b/tcg/i386/tcg-target.inc.c | ||
20 | index XXXXXXX..XXXXXXX 100644 | ||
21 | --- a/tcg/i386/tcg-target.inc.c | ||
22 | +++ b/tcg/i386/tcg-target.inc.c | ||
23 | @@ -XXX,XX +XXX,XX @@ static void expand_vec_sari(TCGType type, unsigned vece, | ||
24 | |||
25 | case MO_64: | ||
26 | if (imm <= 32) { | ||
27 | - /* We can emulate a small sign extend by performing an arithmetic | ||
28 | + /* | ||
29 | + * We can emulate a small sign extend by performing an arithmetic | ||
30 | * 32-bit shift and overwriting the high half of a 64-bit logical | ||
31 | - * shift (note that the ISA says shift of 32 is valid). | ||
32 | + * shift. Note that the ISA says shift of 32 is valid, but TCG | ||
33 | + * does not, so we have to bound the smaller shift -- we get the | ||
34 | + * same result in the high half either way. | ||
35 | */ | ||
36 | t1 = tcg_temp_new_vec(type); | ||
37 | - tcg_gen_sari_vec(MO_32, t1, v1, imm); | ||
38 | + tcg_gen_sari_vec(MO_32, t1, v1, MIN(imm, 31)); | ||
39 | tcg_gen_shri_vec(MO_64, v0, v1, imm); | ||
40 | vec_gen_4(INDEX_op_x86_blend_vec, type, MO_32, | ||
41 | tcgv_vec_arg(v0), tcgv_vec_arg(v0), | ||
42 | -- | ||
43 | 2.20.1 | ||
44 | |||
45 | diff view generated by jsdifflib |
1 | From: Taylor Simpson <tsimpson@quicinc.com> | 1 | The comment in tcg-runtime-gvec.c about CONFIG_VECTOR16 says that |
---|---|---|---|
2 | tcg-op-gvec.c has eliminated size 8 vectors, and only passes on | ||
3 | multiples of 16. This may have been true of the first few operations, | ||
4 | but is not true of all operations. | ||
2 | 5 | ||
3 | Currently, helpers can only take up to 6 arguments. This patch adds the | 6 | In particular, multiply, shift by scalar, and compare of 8- and 16-bit |
4 | capability for up to 7 arguments. I have tested it with the Hexagon port | 7 | elements are not expanded inline if host vector operations are not |
5 | that I am preparing for submission. | 8 | supported. |
6 | 9 | ||
7 | Signed-off-by: Taylor Simpson <tsimpson@quicinc.com> | 10 | For an x86_64 host that does not support AVX, this means that we will |
8 | Message-Id: <1580942510-2820-1-git-send-email-tsimpson@quicinc.com> | 11 | fall back to the helper, which will attempt to use SSE instructions, |
12 | which will SEGV on an invalid 8-byte aligned memory operation. | ||
13 | |||
14 | This patch simply removes the CONFIG_VECTOR16 code and configuration | ||
15 | without further simplification. | ||
16 | |||
17 | Buglink: https://bugs.launchpad.net/bugs/1863508 | ||
9 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 18 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
10 | --- | 19 | --- |
11 | include/exec/helper-gen.h | 13 +++++++++++++ | 20 | configure | 56 ------------------------------------ |
12 | include/exec/helper-head.h | 2 ++ | 21 | accel/tcg/tcg-runtime-gvec.c | 35 +--------------------- |
13 | include/exec/helper-proto.h | 6 ++++++ | 22 | 2 files changed, 1 insertion(+), 90 deletions(-) |
14 | include/exec/helper-tcg.h | 7 +++++++ | ||
15 | 4 files changed, 28 insertions(+) | ||
16 | 23 | ||
17 | diff --git a/include/exec/helper-gen.h b/include/exec/helper-gen.h | 24 | diff --git a/configure b/configure |
25 | index XXXXXXX..XXXXXXX 100755 | ||
26 | --- a/configure | ||
27 | +++ b/configure | ||
28 | @@ -XXX,XX +XXX,XX @@ if test "$plugins" = "yes" && | ||
29 | "for this purpose. You can't build with --static." | ||
30 | fi | ||
31 | |||
32 | -######################################## | ||
33 | -# See if 16-byte vector operations are supported. | ||
34 | -# Even without a vector unit the compiler may expand these. | ||
35 | -# There is a bug in old GCC for PPC that crashes here. | ||
36 | -# Unfortunately it's the system compiler for Centos 7. | ||
37 | - | ||
38 | -cat > $TMPC << EOF | ||
39 | -typedef unsigned char U1 __attribute__((vector_size(16))); | ||
40 | -typedef unsigned short U2 __attribute__((vector_size(16))); | ||
41 | -typedef unsigned int U4 __attribute__((vector_size(16))); | ||
42 | -typedef unsigned long long U8 __attribute__((vector_size(16))); | ||
43 | -typedef signed char S1 __attribute__((vector_size(16))); | ||
44 | -typedef signed short S2 __attribute__((vector_size(16))); | ||
45 | -typedef signed int S4 __attribute__((vector_size(16))); | ||
46 | -typedef signed long long S8 __attribute__((vector_size(16))); | ||
47 | -static U1 a1, b1; | ||
48 | -static U2 a2, b2; | ||
49 | -static U4 a4, b4; | ||
50 | -static U8 a8, b8; | ||
51 | -static S1 c1; | ||
52 | -static S2 c2; | ||
53 | -static S4 c4; | ||
54 | -static S8 c8; | ||
55 | -static int i; | ||
56 | -void helper(void *d, void *a, int shift, int i); | ||
57 | -void helper(void *d, void *a, int shift, int i) | ||
58 | -{ | ||
59 | - *(U1 *)(d + i) = *(U1 *)(a + i) << shift; | ||
60 | - *(U2 *)(d + i) = *(U2 *)(a + i) << shift; | ||
61 | - *(U4 *)(d + i) = *(U4 *)(a + i) << shift; | ||
62 | - *(U8 *)(d + i) = *(U8 *)(a + i) << shift; | ||
63 | -} | ||
64 | -int main(void) | ||
65 | -{ | ||
66 | - a1 += b1; a2 += b2; a4 += b4; a8 += b8; | ||
67 | - a1 -= b1; a2 -= b2; a4 -= b4; a8 -= b8; | ||
68 | - a1 *= b1; a2 *= b2; a4 *= b4; a8 *= b8; | ||
69 | - a1 &= b1; a2 &= b2; a4 &= b4; a8 &= b8; | ||
70 | - a1 |= b1; a2 |= b2; a4 |= b4; a8 |= b8; | ||
71 | - a1 ^= b1; a2 ^= b2; a4 ^= b4; a8 ^= b8; | ||
72 | - a1 <<= i; a2 <<= i; a4 <<= i; a8 <<= i; | ||
73 | - a1 >>= i; a2 >>= i; a4 >>= i; a8 >>= i; | ||
74 | - c1 >>= i; c2 >>= i; c4 >>= i; c8 >>= i; | ||
75 | - return 0; | ||
76 | -} | ||
77 | -EOF | ||
78 | - | ||
79 | -vector16=no | ||
80 | -if compile_prog "" "" ; then | ||
81 | - vector16=yes | ||
82 | -fi | ||
83 | - | ||
84 | ######################################## | ||
85 | # See if __attribute__((alias)) is supported. | ||
86 | # This false for Xcode 9, but has been remedied for Xcode 10. | ||
87 | @@ -XXX,XX +XXX,XX @@ if test "$atomic64" = "yes" ; then | ||
88 | echo "CONFIG_ATOMIC64=y" >> $config_host_mak | ||
89 | fi | ||
90 | |||
91 | -if test "$vector16" = "yes" ; then | ||
92 | - echo "CONFIG_VECTOR16=y" >> $config_host_mak | ||
93 | -fi | ||
94 | - | ||
95 | if test "$attralias" = "yes" ; then | ||
96 | echo "CONFIG_ATTRIBUTE_ALIAS=y" >> $config_host_mak | ||
97 | fi | ||
98 | diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c | ||
18 | index XXXXXXX..XXXXXXX 100644 | 99 | index XXXXXXX..XXXXXXX 100644 |
19 | --- a/include/exec/helper-gen.h | 100 | --- a/accel/tcg/tcg-runtime-gvec.c |
20 | +++ b/include/exec/helper-gen.h | 101 | +++ b/accel/tcg/tcg-runtime-gvec.c |
21 | @@ -XXX,XX +XXX,XX @@ static inline void glue(gen_helper_, name)(dh_retvar_decl(ret) \ | 102 | @@ -XXX,XX +XXX,XX @@ |
22 | tcg_gen_callN(HELPER(name), dh_retvar(ret), 6, args); \ | 103 | #include "tcg/tcg-gvec-desc.h" |
104 | |||
105 | |||
106 | -/* Virtually all hosts support 16-byte vectors. Those that don't can emulate | ||
107 | - * them via GCC's generic vector extension. This turns out to be simpler and | ||
108 | - * more reliable than getting the compiler to autovectorize. | ||
109 | - * | ||
110 | - * In tcg-op-gvec.c, we asserted that both the size and alignment of the data | ||
111 | - * are multiples of 16. | ||
112 | - * | ||
113 | - * When the compiler does not support all of the operations we require, the | ||
114 | - * loops are written so that we can always fall back on the base types. | ||
115 | - */ | ||
116 | -#ifdef CONFIG_VECTOR16 | ||
117 | -typedef uint8_t vec8 __attribute__((vector_size(16))); | ||
118 | -typedef uint16_t vec16 __attribute__((vector_size(16))); | ||
119 | -typedef uint32_t vec32 __attribute__((vector_size(16))); | ||
120 | -typedef uint64_t vec64 __attribute__((vector_size(16))); | ||
121 | - | ||
122 | -typedef int8_t svec8 __attribute__((vector_size(16))); | ||
123 | -typedef int16_t svec16 __attribute__((vector_size(16))); | ||
124 | -typedef int32_t svec32 __attribute__((vector_size(16))); | ||
125 | -typedef int64_t svec64 __attribute__((vector_size(16))); | ||
126 | - | ||
127 | -#define DUP16(X) { X, X, X, X, X, X, X, X, X, X, X, X, X, X, X, X } | ||
128 | -#define DUP8(X) { X, X, X, X, X, X, X, X } | ||
129 | -#define DUP4(X) { X, X, X, X } | ||
130 | -#define DUP2(X) { X, X } | ||
131 | -#else | ||
132 | typedef uint8_t vec8; | ||
133 | typedef uint16_t vec16; | ||
134 | typedef uint32_t vec32; | ||
135 | @@ -XXX,XX +XXX,XX @@ typedef int64_t svec64; | ||
136 | #define DUP8(X) X | ||
137 | #define DUP4(X) X | ||
138 | #define DUP2(X) X | ||
139 | -#endif /* CONFIG_VECTOR16 */ | ||
140 | |||
141 | static inline void clear_high(void *d, intptr_t oprsz, uint32_t desc) | ||
142 | { | ||
143 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar64v)(void *d, void *a, void *b, uint32_t desc) | ||
144 | clear_high(d, oprsz, desc); | ||
23 | } | 145 | } |
24 | 146 | ||
25 | +#define DEF_HELPER_FLAGS_7(name, flags, ret, t1, t2, t3, t4, t5, t6, t7)\ | 147 | -/* If vectors are enabled, the compiler fills in -1 for true. |
26 | +static inline void glue(gen_helper_, name)(dh_retvar_decl(ret) \ | 148 | - Otherwise, we must take care of this by hand. */ |
27 | + dh_arg_decl(t1, 1), dh_arg_decl(t2, 2), dh_arg_decl(t3, 3), \ | 149 | -#ifdef CONFIG_VECTOR16 |
28 | + dh_arg_decl(t4, 4), dh_arg_decl(t5, 5), dh_arg_decl(t6, 6), \ | 150 | -# define DO_CMP0(X) X |
29 | + dh_arg_decl(t7, 7)) \ | 151 | -#else |
30 | +{ \ | 152 | -# define DO_CMP0(X) -(X) |
31 | + TCGTemp *args[7] = { dh_arg(t1, 1), dh_arg(t2, 2), dh_arg(t3, 3), \ | 153 | -#endif |
32 | + dh_arg(t4, 4), dh_arg(t5, 5), dh_arg(t6, 6), \ | 154 | +#define DO_CMP0(X) -(X) |
33 | + dh_arg(t7, 7) }; \ | 155 | |
34 | + tcg_gen_callN(HELPER(name), dh_retvar(ret), 7, args); \ | 156 | #define DO_CMP1(NAME, TYPE, OP) \ |
35 | +} | 157 | void HELPER(NAME)(void *d, void *a, void *b, uint32_t desc) \ |
36 | + | ||
37 | #include "helper.h" | ||
38 | #include "trace/generated-helpers.h" | ||
39 | #include "trace/generated-helpers-wrappers.h" | ||
40 | @@ -XXX,XX +XXX,XX @@ static inline void glue(gen_helper_, name)(dh_retvar_decl(ret) \ | ||
41 | #undef DEF_HELPER_FLAGS_4 | ||
42 | #undef DEF_HELPER_FLAGS_5 | ||
43 | #undef DEF_HELPER_FLAGS_6 | ||
44 | +#undef DEF_HELPER_FLAGS_7 | ||
45 | #undef GEN_HELPER | ||
46 | |||
47 | #endif /* HELPER_GEN_H */ | ||
48 | diff --git a/include/exec/helper-head.h b/include/exec/helper-head.h | ||
49 | index XXXXXXX..XXXXXXX 100644 | ||
50 | --- a/include/exec/helper-head.h | ||
51 | +++ b/include/exec/helper-head.h | ||
52 | @@ -XXX,XX +XXX,XX @@ | ||
53 | DEF_HELPER_FLAGS_5(name, 0, ret, t1, t2, t3, t4, t5) | ||
54 | #define DEF_HELPER_6(name, ret, t1, t2, t3, t4, t5, t6) \ | ||
55 | DEF_HELPER_FLAGS_6(name, 0, ret, t1, t2, t3, t4, t5, t6) | ||
56 | +#define DEF_HELPER_7(name, ret, t1, t2, t3, t4, t5, t6, t7) \ | ||
57 | + DEF_HELPER_FLAGS_7(name, 0, ret, t1, t2, t3, t4, t5, t6, t7) | ||
58 | |||
59 | /* MAX_OPC_PARAM_IARGS must be set to n if last entry is DEF_HELPER_FLAGS_n. */ | ||
60 | |||
61 | diff --git a/include/exec/helper-proto.h b/include/exec/helper-proto.h | ||
62 | index XXXXXXX..XXXXXXX 100644 | ||
63 | --- a/include/exec/helper-proto.h | ||
64 | +++ b/include/exec/helper-proto.h | ||
65 | @@ -XXX,XX +XXX,XX @@ dh_ctype(ret) HELPER(name) (dh_ctype(t1), dh_ctype(t2), dh_ctype(t3), \ | ||
66 | dh_ctype(ret) HELPER(name) (dh_ctype(t1), dh_ctype(t2), dh_ctype(t3), \ | ||
67 | dh_ctype(t4), dh_ctype(t5), dh_ctype(t6)); | ||
68 | |||
69 | +#define DEF_HELPER_FLAGS_7(name, flags, ret, t1, t2, t3, t4, t5, t6, t7) \ | ||
70 | +dh_ctype(ret) HELPER(name) (dh_ctype(t1), dh_ctype(t2), dh_ctype(t3), \ | ||
71 | + dh_ctype(t4), dh_ctype(t5), dh_ctype(t6), \ | ||
72 | + dh_ctype(t7)); | ||
73 | + | ||
74 | #include "helper.h" | ||
75 | #include "trace/generated-helpers.h" | ||
76 | #include "tcg-runtime.h" | ||
77 | @@ -XXX,XX +XXX,XX @@ dh_ctype(ret) HELPER(name) (dh_ctype(t1), dh_ctype(t2), dh_ctype(t3), \ | ||
78 | #undef DEF_HELPER_FLAGS_4 | ||
79 | #undef DEF_HELPER_FLAGS_5 | ||
80 | #undef DEF_HELPER_FLAGS_6 | ||
81 | +#undef DEF_HELPER_FLAGS_7 | ||
82 | |||
83 | #endif /* HELPER_PROTO_H */ | ||
84 | diff --git a/include/exec/helper-tcg.h b/include/exec/helper-tcg.h | ||
85 | index XXXXXXX..XXXXXXX 100644 | ||
86 | --- a/include/exec/helper-tcg.h | ||
87 | +++ b/include/exec/helper-tcg.h | ||
88 | @@ -XXX,XX +XXX,XX @@ | ||
89 | | dh_sizemask(t2, 2) | dh_sizemask(t3, 3) | dh_sizemask(t4, 4) \ | ||
90 | | dh_sizemask(t5, 5) | dh_sizemask(t6, 6) }, | ||
91 | |||
92 | +#define DEF_HELPER_FLAGS_7(NAME, FLAGS, ret, t1, t2, t3, t4, t5, t6, t7) \ | ||
93 | + { .func = HELPER(NAME), .name = str(NAME), .flags = FLAGS, \ | ||
94 | + .sizemask = dh_sizemask(ret, 0) | dh_sizemask(t1, 1) \ | ||
95 | + | dh_sizemask(t2, 2) | dh_sizemask(t3, 3) | dh_sizemask(t4, 4) \ | ||
96 | + | dh_sizemask(t5, 5) | dh_sizemask(t6, 6) | dh_sizemask(t7, 7) }, | ||
97 | + | ||
98 | #include "helper.h" | ||
99 | #include "trace/generated-helpers.h" | ||
100 | #include "tcg-runtime.h" | ||
101 | @@ -XXX,XX +XXX,XX @@ | ||
102 | #undef DEF_HELPER_FLAGS_4 | ||
103 | #undef DEF_HELPER_FLAGS_5 | ||
104 | #undef DEF_HELPER_FLAGS_6 | ||
105 | +#undef DEF_HELPER_FLAGS_7 | ||
106 | |||
107 | #endif /* HELPER_TCG_H */ | ||
108 | -- | 158 | -- |
109 | 2.20.1 | 159 | 2.20.1 |
110 | 160 | ||
111 | 161 | diff view generated by jsdifflib |
New patch | |||
---|---|---|---|
1 | Partial cleanup from the CONFIG_VECTOR16 removal. | ||
2 | Replace the vec* types with their scalar expansions. | ||
1 | 3 | ||
4 | Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> | ||
5 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | ||
6 | --- | ||
7 | accel/tcg/tcg-runtime-gvec.c | 270 +++++++++++++++++------------------ | ||
8 | 1 file changed, 130 insertions(+), 140 deletions(-) | ||
9 | |||
10 | diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c | ||
11 | index XXXXXXX..XXXXXXX 100644 | ||
12 | --- a/accel/tcg/tcg-runtime-gvec.c | ||
13 | +++ b/accel/tcg/tcg-runtime-gvec.c | ||
14 | @@ -XXX,XX +XXX,XX @@ | ||
15 | #include "tcg/tcg-gvec-desc.h" | ||
16 | |||
17 | |||
18 | -typedef uint8_t vec8; | ||
19 | -typedef uint16_t vec16; | ||
20 | -typedef uint32_t vec32; | ||
21 | -typedef uint64_t vec64; | ||
22 | - | ||
23 | -typedef int8_t svec8; | ||
24 | -typedef int16_t svec16; | ||
25 | -typedef int32_t svec32; | ||
26 | -typedef int64_t svec64; | ||
27 | - | ||
28 | #define DUP16(X) X | ||
29 | #define DUP8(X) X | ||
30 | #define DUP4(X) X | ||
31 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add8)(void *d, void *a, void *b, uint32_t desc) | ||
32 | intptr_t oprsz = simd_oprsz(desc); | ||
33 | intptr_t i; | ||
34 | |||
35 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
36 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) + *(vec8 *)(b + i); | ||
37 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
38 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) + *(uint8_t *)(b + i); | ||
39 | } | ||
40 | clear_high(d, oprsz, desc); | ||
41 | } | ||
42 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add16)(void *d, void *a, void *b, uint32_t desc) | ||
43 | intptr_t oprsz = simd_oprsz(desc); | ||
44 | intptr_t i; | ||
45 | |||
46 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
47 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) + *(vec16 *)(b + i); | ||
48 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
49 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) + *(uint16_t *)(b + i); | ||
50 | } | ||
51 | clear_high(d, oprsz, desc); | ||
52 | } | ||
53 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add32)(void *d, void *a, void *b, uint32_t desc) | ||
54 | intptr_t oprsz = simd_oprsz(desc); | ||
55 | intptr_t i; | ||
56 | |||
57 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
58 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) + *(vec32 *)(b + i); | ||
59 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
60 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) + *(uint32_t *)(b + i); | ||
61 | } | ||
62 | clear_high(d, oprsz, desc); | ||
63 | } | ||
64 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add64)(void *d, void *a, void *b, uint32_t desc) | ||
65 | intptr_t oprsz = simd_oprsz(desc); | ||
66 | intptr_t i; | ||
67 | |||
68 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
69 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) + *(vec64 *)(b + i); | ||
70 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
71 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) + *(uint64_t *)(b + i); | ||
72 | } | ||
73 | clear_high(d, oprsz, desc); | ||
74 | } | ||
75 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add64)(void *d, void *a, void *b, uint32_t desc) | ||
76 | void HELPER(gvec_adds8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
77 | { | ||
78 | intptr_t oprsz = simd_oprsz(desc); | ||
79 | - vec8 vecb = (vec8)DUP16(b); | ||
80 | + uint8_t vecb = (uint8_t)DUP16(b); | ||
81 | intptr_t i; | ||
82 | |||
83 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
84 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) + vecb; | ||
85 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
86 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) + vecb; | ||
87 | } | ||
88 | clear_high(d, oprsz, desc); | ||
89 | } | ||
90 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
91 | void HELPER(gvec_adds16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
92 | { | ||
93 | intptr_t oprsz = simd_oprsz(desc); | ||
94 | - vec16 vecb = (vec16)DUP8(b); | ||
95 | + uint16_t vecb = (uint16_t)DUP8(b); | ||
96 | intptr_t i; | ||
97 | |||
98 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
99 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) + vecb; | ||
100 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
101 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) + vecb; | ||
102 | } | ||
103 | clear_high(d, oprsz, desc); | ||
104 | } | ||
105 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
106 | void HELPER(gvec_adds32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
107 | { | ||
108 | intptr_t oprsz = simd_oprsz(desc); | ||
109 | - vec32 vecb = (vec32)DUP4(b); | ||
110 | + uint32_t vecb = (uint32_t)DUP4(b); | ||
111 | intptr_t i; | ||
112 | |||
113 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
114 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) + vecb; | ||
115 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
116 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) + vecb; | ||
117 | } | ||
118 | clear_high(d, oprsz, desc); | ||
119 | } | ||
120 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
121 | void HELPER(gvec_adds64)(void *d, void *a, uint64_t b, uint32_t desc) | ||
122 | { | ||
123 | intptr_t oprsz = simd_oprsz(desc); | ||
124 | - vec64 vecb = (vec64)DUP2(b); | ||
125 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
126 | intptr_t i; | ||
127 | |||
128 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
129 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) + vecb; | ||
130 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
131 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) + vecb; | ||
132 | } | ||
133 | clear_high(d, oprsz, desc); | ||
134 | } | ||
135 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub8)(void *d, void *a, void *b, uint32_t desc) | ||
136 | intptr_t oprsz = simd_oprsz(desc); | ||
137 | intptr_t i; | ||
138 | |||
139 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
140 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) - *(vec8 *)(b + i); | ||
141 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
142 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) - *(uint8_t *)(b + i); | ||
143 | } | ||
144 | clear_high(d, oprsz, desc); | ||
145 | } | ||
146 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub16)(void *d, void *a, void *b, uint32_t desc) | ||
147 | intptr_t oprsz = simd_oprsz(desc); | ||
148 | intptr_t i; | ||
149 | |||
150 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
151 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) - *(vec16 *)(b + i); | ||
152 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
153 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) - *(uint16_t *)(b + i); | ||
154 | } | ||
155 | clear_high(d, oprsz, desc); | ||
156 | } | ||
157 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub32)(void *d, void *a, void *b, uint32_t desc) | ||
158 | intptr_t oprsz = simd_oprsz(desc); | ||
159 | intptr_t i; | ||
160 | |||
161 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
162 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) - *(vec32 *)(b + i); | ||
163 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
164 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) - *(uint32_t *)(b + i); | ||
165 | } | ||
166 | clear_high(d, oprsz, desc); | ||
167 | } | ||
168 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub64)(void *d, void *a, void *b, uint32_t desc) | ||
169 | intptr_t oprsz = simd_oprsz(desc); | ||
170 | intptr_t i; | ||
171 | |||
172 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
173 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) - *(vec64 *)(b + i); | ||
174 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
175 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) - *(uint64_t *)(b + i); | ||
176 | } | ||
177 | clear_high(d, oprsz, desc); | ||
178 | } | ||
179 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub64)(void *d, void *a, void *b, uint32_t desc) | ||
180 | void HELPER(gvec_subs8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
181 | { | ||
182 | intptr_t oprsz = simd_oprsz(desc); | ||
183 | - vec8 vecb = (vec8)DUP16(b); | ||
184 | + uint8_t vecb = (uint8_t)DUP16(b); | ||
185 | intptr_t i; | ||
186 | |||
187 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
188 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) - vecb; | ||
189 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
190 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) - vecb; | ||
191 | } | ||
192 | clear_high(d, oprsz, desc); | ||
193 | } | ||
194 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
195 | void HELPER(gvec_subs16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
196 | { | ||
197 | intptr_t oprsz = simd_oprsz(desc); | ||
198 | - vec16 vecb = (vec16)DUP8(b); | ||
199 | + uint16_t vecb = (uint16_t)DUP8(b); | ||
200 | intptr_t i; | ||
201 | |||
202 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
203 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) - vecb; | ||
204 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
205 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) - vecb; | ||
206 | } | ||
207 | clear_high(d, oprsz, desc); | ||
208 | } | ||
209 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
210 | void HELPER(gvec_subs32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
211 | { | ||
212 | intptr_t oprsz = simd_oprsz(desc); | ||
213 | - vec32 vecb = (vec32)DUP4(b); | ||
214 | + uint32_t vecb = (uint32_t)DUP4(b); | ||
215 | intptr_t i; | ||
216 | |||
217 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
218 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) - vecb; | ||
219 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
220 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) - vecb; | ||
221 | } | ||
222 | clear_high(d, oprsz, desc); | ||
223 | } | ||
224 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
225 | void HELPER(gvec_subs64)(void *d, void *a, uint64_t b, uint32_t desc) | ||
226 | { | ||
227 | intptr_t oprsz = simd_oprsz(desc); | ||
228 | - vec64 vecb = (vec64)DUP2(b); | ||
229 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
230 | intptr_t i; | ||
231 | |||
232 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
233 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) - vecb; | ||
234 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
235 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) - vecb; | ||
236 | } | ||
237 | clear_high(d, oprsz, desc); | ||
238 | } | ||
239 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul8)(void *d, void *a, void *b, uint32_t desc) | ||
240 | intptr_t oprsz = simd_oprsz(desc); | ||
241 | intptr_t i; | ||
242 | |||
243 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
244 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) * *(vec8 *)(b + i); | ||
245 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
246 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) * *(uint8_t *)(b + i); | ||
247 | } | ||
248 | clear_high(d, oprsz, desc); | ||
249 | } | ||
250 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul16)(void *d, void *a, void *b, uint32_t desc) | ||
251 | intptr_t oprsz = simd_oprsz(desc); | ||
252 | intptr_t i; | ||
253 | |||
254 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
255 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) * *(vec16 *)(b + i); | ||
256 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
257 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) * *(uint16_t *)(b + i); | ||
258 | } | ||
259 | clear_high(d, oprsz, desc); | ||
260 | } | ||
261 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul32)(void *d, void *a, void *b, uint32_t desc) | ||
262 | intptr_t oprsz = simd_oprsz(desc); | ||
263 | intptr_t i; | ||
264 | |||
265 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
266 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) * *(vec32 *)(b + i); | ||
267 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
268 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) * *(uint32_t *)(b + i); | ||
269 | } | ||
270 | clear_high(d, oprsz, desc); | ||
271 | } | ||
272 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul64)(void *d, void *a, void *b, uint32_t desc) | ||
273 | intptr_t oprsz = simd_oprsz(desc); | ||
274 | intptr_t i; | ||
275 | |||
276 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
277 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) * *(vec64 *)(b + i); | ||
278 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
279 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) * *(uint64_t *)(b + i); | ||
280 | } | ||
281 | clear_high(d, oprsz, desc); | ||
282 | } | ||
283 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul64)(void *d, void *a, void *b, uint32_t desc) | ||
284 | void HELPER(gvec_muls8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
285 | { | ||
286 | intptr_t oprsz = simd_oprsz(desc); | ||
287 | - vec8 vecb = (vec8)DUP16(b); | ||
288 | + uint8_t vecb = (uint8_t)DUP16(b); | ||
289 | intptr_t i; | ||
290 | |||
291 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
292 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) * vecb; | ||
293 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
294 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) * vecb; | ||
295 | } | ||
296 | clear_high(d, oprsz, desc); | ||
297 | } | ||
298 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
299 | void HELPER(gvec_muls16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
300 | { | ||
301 | intptr_t oprsz = simd_oprsz(desc); | ||
302 | - vec16 vecb = (vec16)DUP8(b); | ||
303 | + uint16_t vecb = (uint16_t)DUP8(b); | ||
304 | intptr_t i; | ||
305 | |||
306 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
307 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) * vecb; | ||
308 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
309 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) * vecb; | ||
310 | } | ||
311 | clear_high(d, oprsz, desc); | ||
312 | } | ||
313 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
314 | void HELPER(gvec_muls32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
315 | { | ||
316 | intptr_t oprsz = simd_oprsz(desc); | ||
317 | - vec32 vecb = (vec32)DUP4(b); | ||
318 | + uint32_t vecb = (uint32_t)DUP4(b); | ||
319 | intptr_t i; | ||
320 | |||
321 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
322 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) * vecb; | ||
323 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
324 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) * vecb; | ||
325 | } | ||
326 | clear_high(d, oprsz, desc); | ||
327 | } | ||
328 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
329 | void HELPER(gvec_muls64)(void *d, void *a, uint64_t b, uint32_t desc) | ||
330 | { | ||
331 | intptr_t oprsz = simd_oprsz(desc); | ||
332 | - vec64 vecb = (vec64)DUP2(b); | ||
333 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
334 | intptr_t i; | ||
335 | |||
336 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
337 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) * vecb; | ||
338 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
339 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) * vecb; | ||
340 | } | ||
341 | clear_high(d, oprsz, desc); | ||
342 | } | ||
343 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_neg8)(void *d, void *a, uint32_t desc) | ||
344 | intptr_t oprsz = simd_oprsz(desc); | ||
345 | intptr_t i; | ||
346 | |||
347 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
348 | - *(vec8 *)(d + i) = -*(vec8 *)(a + i); | ||
349 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
350 | + *(uint8_t *)(d + i) = -*(uint8_t *)(a + i); | ||
351 | } | ||
352 | clear_high(d, oprsz, desc); | ||
353 | } | ||
354 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_neg16)(void *d, void *a, uint32_t desc) | ||
355 | intptr_t oprsz = simd_oprsz(desc); | ||
356 | intptr_t i; | ||
357 | |||
358 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
359 | - *(vec16 *)(d + i) = -*(vec16 *)(a + i); | ||
360 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
361 | + *(uint16_t *)(d + i) = -*(uint16_t *)(a + i); | ||
362 | } | ||
363 | clear_high(d, oprsz, desc); | ||
364 | } | ||
365 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_neg32)(void *d, void *a, uint32_t desc) | ||
366 | intptr_t oprsz = simd_oprsz(desc); | ||
367 | intptr_t i; | ||
368 | |||
369 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
370 | - *(vec32 *)(d + i) = -*(vec32 *)(a + i); | ||
371 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
372 | + *(uint32_t *)(d + i) = -*(uint32_t *)(a + i); | ||
373 | } | ||
374 | clear_high(d, oprsz, desc); | ||
375 | } | ||
376 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_neg64)(void *d, void *a, uint32_t desc) | ||
377 | intptr_t oprsz = simd_oprsz(desc); | ||
378 | intptr_t i; | ||
379 | |||
380 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
381 | - *(vec64 *)(d + i) = -*(vec64 *)(a + i); | ||
382 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
383 | + *(uint64_t *)(d + i) = -*(uint64_t *)(a + i); | ||
384 | } | ||
385 | clear_high(d, oprsz, desc); | ||
386 | } | ||
387 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_not)(void *d, void *a, uint32_t desc) | ||
388 | intptr_t oprsz = simd_oprsz(desc); | ||
389 | intptr_t i; | ||
390 | |||
391 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
392 | - *(vec64 *)(d + i) = ~*(vec64 *)(a + i); | ||
393 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
394 | + *(uint64_t *)(d + i) = ~*(uint64_t *)(a + i); | ||
395 | } | ||
396 | clear_high(d, oprsz, desc); | ||
397 | } | ||
398 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_and)(void *d, void *a, void *b, uint32_t desc) | ||
399 | intptr_t oprsz = simd_oprsz(desc); | ||
400 | intptr_t i; | ||
401 | |||
402 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
403 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) & *(vec64 *)(b + i); | ||
404 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
405 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) & *(uint64_t *)(b + i); | ||
406 | } | ||
407 | clear_high(d, oprsz, desc); | ||
408 | } | ||
409 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_or)(void *d, void *a, void *b, uint32_t desc) | ||
410 | intptr_t oprsz = simd_oprsz(desc); | ||
411 | intptr_t i; | ||
412 | |||
413 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
414 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) | *(vec64 *)(b + i); | ||
415 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
416 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) | *(uint64_t *)(b + i); | ||
417 | } | ||
418 | clear_high(d, oprsz, desc); | ||
419 | } | ||
420 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_xor)(void *d, void *a, void *b, uint32_t desc) | ||
421 | intptr_t oprsz = simd_oprsz(desc); | ||
422 | intptr_t i; | ||
423 | |||
424 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
425 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) ^ *(vec64 *)(b + i); | ||
426 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
427 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) ^ *(uint64_t *)(b + i); | ||
428 | } | ||
429 | clear_high(d, oprsz, desc); | ||
430 | } | ||
431 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_andc)(void *d, void *a, void *b, uint32_t desc) | ||
432 | intptr_t oprsz = simd_oprsz(desc); | ||
433 | intptr_t i; | ||
434 | |||
435 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
436 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) &~ *(vec64 *)(b + i); | ||
437 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
438 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) &~ *(uint64_t *)(b + i); | ||
439 | } | ||
440 | clear_high(d, oprsz, desc); | ||
441 | } | ||
442 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_orc)(void *d, void *a, void *b, uint32_t desc) | ||
443 | intptr_t oprsz = simd_oprsz(desc); | ||
444 | intptr_t i; | ||
445 | |||
446 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
447 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) |~ *(vec64 *)(b + i); | ||
448 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
449 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) |~ *(uint64_t *)(b + i); | ||
450 | } | ||
451 | clear_high(d, oprsz, desc); | ||
452 | } | ||
453 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_nand)(void *d, void *a, void *b, uint32_t desc) | ||
454 | intptr_t oprsz = simd_oprsz(desc); | ||
455 | intptr_t i; | ||
456 | |||
457 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
458 | - *(vec64 *)(d + i) = ~(*(vec64 *)(a + i) & *(vec64 *)(b + i)); | ||
459 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
460 | + *(uint64_t *)(d + i) = ~(*(uint64_t *)(a + i) & *(uint64_t *)(b + i)); | ||
461 | } | ||
462 | clear_high(d, oprsz, desc); | ||
463 | } | ||
464 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_nor)(void *d, void *a, void *b, uint32_t desc) | ||
465 | intptr_t oprsz = simd_oprsz(desc); | ||
466 | intptr_t i; | ||
467 | |||
468 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
469 | - *(vec64 *)(d + i) = ~(*(vec64 *)(a + i) | *(vec64 *)(b + i)); | ||
470 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
471 | + *(uint64_t *)(d + i) = ~(*(uint64_t *)(a + i) | *(uint64_t *)(b + i)); | ||
472 | } | ||
473 | clear_high(d, oprsz, desc); | ||
474 | } | ||
475 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_eqv)(void *d, void *a, void *b, uint32_t desc) | ||
476 | intptr_t oprsz = simd_oprsz(desc); | ||
477 | intptr_t i; | ||
478 | |||
479 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
480 | - *(vec64 *)(d + i) = ~(*(vec64 *)(a + i) ^ *(vec64 *)(b + i)); | ||
481 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
482 | + *(uint64_t *)(d + i) = ~(*(uint64_t *)(a + i) ^ *(uint64_t *)(b + i)); | ||
483 | } | ||
484 | clear_high(d, oprsz, desc); | ||
485 | } | ||
486 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_eqv)(void *d, void *a, void *b, uint32_t desc) | ||
487 | void HELPER(gvec_ands)(void *d, void *a, uint64_t b, uint32_t desc) | ||
488 | { | ||
489 | intptr_t oprsz = simd_oprsz(desc); | ||
490 | - vec64 vecb = (vec64)DUP2(b); | ||
491 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
492 | intptr_t i; | ||
493 | |||
494 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
495 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) & vecb; | ||
496 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
497 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) & vecb; | ||
498 | } | ||
499 | clear_high(d, oprsz, desc); | ||
500 | } | ||
501 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_ands)(void *d, void *a, uint64_t b, uint32_t desc) | ||
502 | void HELPER(gvec_xors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
503 | { | ||
504 | intptr_t oprsz = simd_oprsz(desc); | ||
505 | - vec64 vecb = (vec64)DUP2(b); | ||
506 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
507 | intptr_t i; | ||
508 | |||
509 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
510 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) ^ vecb; | ||
511 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
512 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) ^ vecb; | ||
513 | } | ||
514 | clear_high(d, oprsz, desc); | ||
515 | } | ||
516 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_xors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
517 | void HELPER(gvec_ors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
518 | { | ||
519 | intptr_t oprsz = simd_oprsz(desc); | ||
520 | - vec64 vecb = (vec64)DUP2(b); | ||
521 | + uint64_t vecb = (uint64_t)DUP2(b); | ||
522 | intptr_t i; | ||
523 | |||
524 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
525 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) | vecb; | ||
526 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
527 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) | vecb; | ||
528 | } | ||
529 | clear_high(d, oprsz, desc); | ||
530 | } | ||
531 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shl8i)(void *d, void *a, uint32_t desc) | ||
532 | int shift = simd_data(desc); | ||
533 | intptr_t i; | ||
534 | |||
535 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
536 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) << shift; | ||
537 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
538 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) << shift; | ||
539 | } | ||
540 | clear_high(d, oprsz, desc); | ||
541 | } | ||
542 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shl16i)(void *d, void *a, uint32_t desc) | ||
543 | int shift = simd_data(desc); | ||
544 | intptr_t i; | ||
545 | |||
546 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
547 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) << shift; | ||
548 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
549 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) << shift; | ||
550 | } | ||
551 | clear_high(d, oprsz, desc); | ||
552 | } | ||
553 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shl32i)(void *d, void *a, uint32_t desc) | ||
554 | int shift = simd_data(desc); | ||
555 | intptr_t i; | ||
556 | |||
557 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
558 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) << shift; | ||
559 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
560 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) << shift; | ||
561 | } | ||
562 | clear_high(d, oprsz, desc); | ||
563 | } | ||
564 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shl64i)(void *d, void *a, uint32_t desc) | ||
565 | int shift = simd_data(desc); | ||
566 | intptr_t i; | ||
567 | |||
568 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
569 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) << shift; | ||
570 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
571 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) << shift; | ||
572 | } | ||
573 | clear_high(d, oprsz, desc); | ||
574 | } | ||
575 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shr8i)(void *d, void *a, uint32_t desc) | ||
576 | int shift = simd_data(desc); | ||
577 | intptr_t i; | ||
578 | |||
579 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
580 | - *(vec8 *)(d + i) = *(vec8 *)(a + i) >> shift; | ||
581 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
582 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) >> shift; | ||
583 | } | ||
584 | clear_high(d, oprsz, desc); | ||
585 | } | ||
586 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shr16i)(void *d, void *a, uint32_t desc) | ||
587 | int shift = simd_data(desc); | ||
588 | intptr_t i; | ||
589 | |||
590 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
591 | - *(vec16 *)(d + i) = *(vec16 *)(a + i) >> shift; | ||
592 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
593 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) >> shift; | ||
594 | } | ||
595 | clear_high(d, oprsz, desc); | ||
596 | } | ||
597 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shr32i)(void *d, void *a, uint32_t desc) | ||
598 | int shift = simd_data(desc); | ||
599 | intptr_t i; | ||
600 | |||
601 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
602 | - *(vec32 *)(d + i) = *(vec32 *)(a + i) >> shift; | ||
603 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
604 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) >> shift; | ||
605 | } | ||
606 | clear_high(d, oprsz, desc); | ||
607 | } | ||
608 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_shr64i)(void *d, void *a, uint32_t desc) | ||
609 | int shift = simd_data(desc); | ||
610 | intptr_t i; | ||
611 | |||
612 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
613 | - *(vec64 *)(d + i) = *(vec64 *)(a + i) >> shift; | ||
614 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
615 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) >> shift; | ||
616 | } | ||
617 | clear_high(d, oprsz, desc); | ||
618 | } | ||
619 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar8i)(void *d, void *a, uint32_t desc) | ||
620 | int shift = simd_data(desc); | ||
621 | intptr_t i; | ||
622 | |||
623 | - for (i = 0; i < oprsz; i += sizeof(vec8)) { | ||
624 | - *(svec8 *)(d + i) = *(svec8 *)(a + i) >> shift; | ||
625 | + for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
626 | + *(int8_t *)(d + i) = *(int8_t *)(a + i) >> shift; | ||
627 | } | ||
628 | clear_high(d, oprsz, desc); | ||
629 | } | ||
630 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar16i)(void *d, void *a, uint32_t desc) | ||
631 | int shift = simd_data(desc); | ||
632 | intptr_t i; | ||
633 | |||
634 | - for (i = 0; i < oprsz; i += sizeof(vec16)) { | ||
635 | - *(svec16 *)(d + i) = *(svec16 *)(a + i) >> shift; | ||
636 | + for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
637 | + *(int16_t *)(d + i) = *(int16_t *)(a + i) >> shift; | ||
638 | } | ||
639 | clear_high(d, oprsz, desc); | ||
640 | } | ||
641 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar32i)(void *d, void *a, uint32_t desc) | ||
642 | int shift = simd_data(desc); | ||
643 | intptr_t i; | ||
644 | |||
645 | - for (i = 0; i < oprsz; i += sizeof(vec32)) { | ||
646 | - *(svec32 *)(d + i) = *(svec32 *)(a + i) >> shift; | ||
647 | + for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
648 | + *(int32_t *)(d + i) = *(int32_t *)(a + i) >> shift; | ||
649 | } | ||
650 | clear_high(d, oprsz, desc); | ||
651 | } | ||
652 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar64i)(void *d, void *a, uint32_t desc) | ||
653 | int shift = simd_data(desc); | ||
654 | intptr_t i; | ||
655 | |||
656 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
657 | - *(svec64 *)(d + i) = *(svec64 *)(a + i) >> shift; | ||
658 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
659 | + *(int64_t *)(d + i) = *(int64_t *)(a + i) >> shift; | ||
660 | } | ||
661 | clear_high(d, oprsz, desc); | ||
662 | } | ||
663 | @@ -XXX,XX +XXX,XX @@ void HELPER(NAME)(void *d, void *a, void *b, uint32_t desc) \ | ||
664 | } | ||
665 | |||
666 | #define DO_CMP2(SZ) \ | ||
667 | - DO_CMP1(gvec_eq##SZ, vec##SZ, ==) \ | ||
668 | - DO_CMP1(gvec_ne##SZ, vec##SZ, !=) \ | ||
669 | - DO_CMP1(gvec_lt##SZ, svec##SZ, <) \ | ||
670 | - DO_CMP1(gvec_le##SZ, svec##SZ, <=) \ | ||
671 | - DO_CMP1(gvec_ltu##SZ, vec##SZ, <) \ | ||
672 | - DO_CMP1(gvec_leu##SZ, vec##SZ, <=) | ||
673 | + DO_CMP1(gvec_eq##SZ, uint##SZ##_t, ==) \ | ||
674 | + DO_CMP1(gvec_ne##SZ, uint##SZ##_t, !=) \ | ||
675 | + DO_CMP1(gvec_lt##SZ, int##SZ##_t, <) \ | ||
676 | + DO_CMP1(gvec_le##SZ, int##SZ##_t, <=) \ | ||
677 | + DO_CMP1(gvec_ltu##SZ, uint##SZ##_t, <) \ | ||
678 | + DO_CMP1(gvec_leu##SZ, uint##SZ##_t, <=) | ||
679 | |||
680 | DO_CMP2(8) | ||
681 | DO_CMP2(16) | ||
682 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_bitsel)(void *d, void *a, void *b, void *c, uint32_t desc) | ||
683 | intptr_t oprsz = simd_oprsz(desc); | ||
684 | intptr_t i; | ||
685 | |||
686 | - for (i = 0; i < oprsz; i += sizeof(vec64)) { | ||
687 | - vec64 aa = *(vec64 *)(a + i); | ||
688 | - vec64 bb = *(vec64 *)(b + i); | ||
689 | - vec64 cc = *(vec64 *)(c + i); | ||
690 | - *(vec64 *)(d + i) = (bb & aa) | (cc & ~aa); | ||
691 | + for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
692 | + uint64_t aa = *(uint64_t *)(a + i); | ||
693 | + uint64_t bb = *(uint64_t *)(b + i); | ||
694 | + uint64_t cc = *(uint64_t *)(c + i); | ||
695 | + *(uint64_t *)(d + i) = (bb & aa) | (cc & ~aa); | ||
696 | } | ||
697 | clear_high(d, oprsz, desc); | ||
698 | } | ||
699 | -- | ||
700 | 2.20.1 | ||
701 | |||
702 | diff view generated by jsdifflib |
1 | Extend the vector generator infrastructure to handle | 1 | Partial cleanup from the CONFIG_VECTOR16 removal. |
---|---|---|---|
2 | 5 vector arguments. | 2 | Replace the DUP* expansions with the scalar argument. |
3 | 3 | ||
4 | Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> | 4 | Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> |
5 | Reviewed-by: Alex Bennée <alex.bennee@linaro.org> | ||
6 | Reviewed-by: Taylor Simpson <tsimpson@quicinc.com> | ||
7 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 5 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
8 | --- | 6 | --- |
9 | include/tcg/tcg-op-gvec.h | 7 +++++++ | 7 | accel/tcg/tcg-runtime-gvec.c | 50 +++++++++++------------------------- |
10 | tcg/tcg-op-gvec.c | 32 ++++++++++++++++++++++++++++++++ | 8 | 1 file changed, 15 insertions(+), 35 deletions(-) |
11 | 2 files changed, 39 insertions(+) | 9 | |
12 | 10 | diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c | |
13 | diff --git a/include/tcg/tcg-op-gvec.h b/include/tcg/tcg-op-gvec.h | ||
14 | index XXXXXXX..XXXXXXX 100644 | 11 | index XXXXXXX..XXXXXXX 100644 |
15 | --- a/include/tcg/tcg-op-gvec.h | 12 | --- a/accel/tcg/tcg-runtime-gvec.c |
16 | +++ b/include/tcg/tcg-op-gvec.h | 13 | +++ b/accel/tcg/tcg-runtime-gvec.c |
17 | @@ -XXX,XX +XXX,XX @@ void tcg_gen_gvec_4_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs, | 14 | @@ -XXX,XX +XXX,XX @@ |
18 | uint32_t maxsz, int32_t data, | 15 | #include "tcg/tcg-gvec-desc.h" |
19 | gen_helper_gvec_4_ptr *fn); | 16 | |
20 | 17 | ||
21 | +typedef void gen_helper_gvec_5_ptr(TCGv_ptr, TCGv_ptr, TCGv_ptr, TCGv_ptr, | 18 | -#define DUP16(X) X |
22 | + TCGv_ptr, TCGv_ptr, TCGv_i32); | 19 | -#define DUP8(X) X |
23 | +void tcg_gen_gvec_5_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs, | 20 | -#define DUP4(X) X |
24 | + uint32_t cofs, uint32_t eofs, TCGv_ptr ptr, | 21 | -#define DUP2(X) X |
25 | + uint32_t oprsz, uint32_t maxsz, int32_t data, | 22 | - |
26 | + gen_helper_gvec_5_ptr *fn); | 23 | static inline void clear_high(void *d, intptr_t oprsz, uint32_t desc) |
27 | + | 24 | { |
28 | /* Expand a gvec operation. Either inline or out-of-line depending on | 25 | intptr_t maxsz = simd_maxsz(desc); |
29 | the actual vector size and the operations supported by the host. */ | 26 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_add64)(void *d, void *a, void *b, uint32_t desc) |
30 | typedef struct { | 27 | void HELPER(gvec_adds8)(void *d, void *a, uint64_t b, uint32_t desc) |
31 | diff --git a/tcg/tcg-op-gvec.c b/tcg/tcg-op-gvec.c | 28 | { |
32 | index XXXXXXX..XXXXXXX 100644 | 29 | intptr_t oprsz = simd_oprsz(desc); |
33 | --- a/tcg/tcg-op-gvec.c | 30 | - uint8_t vecb = (uint8_t)DUP16(b); |
34 | +++ b/tcg/tcg-op-gvec.c | 31 | intptr_t i; |
35 | @@ -XXX,XX +XXX,XX @@ void tcg_gen_gvec_4_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs, | 32 | |
36 | tcg_temp_free_i32(desc); | 33 | for (i = 0; i < oprsz; i += sizeof(uint8_t)) { |
37 | } | 34 | - *(uint8_t *)(d + i) = *(uint8_t *)(a + i) + vecb; |
38 | 35 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) + (uint8_t)b; | |
39 | +/* Generate a call to a gvec-style helper with five vector operands | 36 | } |
40 | + and an extra pointer operand. */ | 37 | clear_high(d, oprsz, desc); |
41 | +void tcg_gen_gvec_5_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs, | 38 | } |
42 | + uint32_t cofs, uint32_t eofs, TCGv_ptr ptr, | 39 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds8)(void *d, void *a, uint64_t b, uint32_t desc) |
43 | + uint32_t oprsz, uint32_t maxsz, int32_t data, | 40 | void HELPER(gvec_adds16)(void *d, void *a, uint64_t b, uint32_t desc) |
44 | + gen_helper_gvec_5_ptr *fn) | 41 | { |
45 | +{ | 42 | intptr_t oprsz = simd_oprsz(desc); |
46 | + TCGv_ptr a0, a1, a2, a3, a4; | 43 | - uint16_t vecb = (uint16_t)DUP8(b); |
47 | + TCGv_i32 desc = tcg_const_i32(simd_desc(oprsz, maxsz, data)); | 44 | intptr_t i; |
48 | + | 45 | |
49 | + a0 = tcg_temp_new_ptr(); | 46 | for (i = 0; i < oprsz; i += sizeof(uint16_t)) { |
50 | + a1 = tcg_temp_new_ptr(); | 47 | - *(uint16_t *)(d + i) = *(uint16_t *)(a + i) + vecb; |
51 | + a2 = tcg_temp_new_ptr(); | 48 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) + (uint16_t)b; |
52 | + a3 = tcg_temp_new_ptr(); | 49 | } |
53 | + a4 = tcg_temp_new_ptr(); | 50 | clear_high(d, oprsz, desc); |
54 | + | 51 | } |
55 | + tcg_gen_addi_ptr(a0, cpu_env, dofs); | 52 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds16)(void *d, void *a, uint64_t b, uint32_t desc) |
56 | + tcg_gen_addi_ptr(a1, cpu_env, aofs); | 53 | void HELPER(gvec_adds32)(void *d, void *a, uint64_t b, uint32_t desc) |
57 | + tcg_gen_addi_ptr(a2, cpu_env, bofs); | 54 | { |
58 | + tcg_gen_addi_ptr(a3, cpu_env, cofs); | 55 | intptr_t oprsz = simd_oprsz(desc); |
59 | + tcg_gen_addi_ptr(a4, cpu_env, eofs); | 56 | - uint32_t vecb = (uint32_t)DUP4(b); |
60 | + | 57 | intptr_t i; |
61 | + fn(a0, a1, a2, a3, a4, ptr, desc); | 58 | |
62 | + | 59 | for (i = 0; i < oprsz; i += sizeof(uint32_t)) { |
63 | + tcg_temp_free_ptr(a0); | 60 | - *(uint32_t *)(d + i) = *(uint32_t *)(a + i) + vecb; |
64 | + tcg_temp_free_ptr(a1); | 61 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) + (uint32_t)b; |
65 | + tcg_temp_free_ptr(a2); | 62 | } |
66 | + tcg_temp_free_ptr(a3); | 63 | clear_high(d, oprsz, desc); |
67 | + tcg_temp_free_ptr(a4); | 64 | } |
68 | + tcg_temp_free_i32(desc); | 65 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_adds32)(void *d, void *a, uint64_t b, uint32_t desc) |
69 | +} | 66 | void HELPER(gvec_adds64)(void *d, void *a, uint64_t b, uint32_t desc) |
70 | + | 67 | { |
71 | /* Return true if we want to implement something of OPRSZ bytes | 68 | intptr_t oprsz = simd_oprsz(desc); |
72 | in units of LNSZ. This limits the expansion of inline code. */ | 69 | - uint64_t vecb = (uint64_t)DUP2(b); |
73 | static inline bool check_size_impl(uint32_t oprsz, uint32_t lnsz) | 70 | intptr_t i; |
71 | |||
72 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
73 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) + vecb; | ||
74 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) + b; | ||
75 | } | ||
76 | clear_high(d, oprsz, desc); | ||
77 | } | ||
78 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sub64)(void *d, void *a, void *b, uint32_t desc) | ||
79 | void HELPER(gvec_subs8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
80 | { | ||
81 | intptr_t oprsz = simd_oprsz(desc); | ||
82 | - uint8_t vecb = (uint8_t)DUP16(b); | ||
83 | intptr_t i; | ||
84 | |||
85 | for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
86 | - *(uint8_t *)(d + i) = *(uint8_t *)(a + i) - vecb; | ||
87 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) - (uint8_t)b; | ||
88 | } | ||
89 | clear_high(d, oprsz, desc); | ||
90 | } | ||
91 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
92 | void HELPER(gvec_subs16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
93 | { | ||
94 | intptr_t oprsz = simd_oprsz(desc); | ||
95 | - uint16_t vecb = (uint16_t)DUP8(b); | ||
96 | intptr_t i; | ||
97 | |||
98 | for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
99 | - *(uint16_t *)(d + i) = *(uint16_t *)(a + i) - vecb; | ||
100 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) - (uint16_t)b; | ||
101 | } | ||
102 | clear_high(d, oprsz, desc); | ||
103 | } | ||
104 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
105 | void HELPER(gvec_subs32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
106 | { | ||
107 | intptr_t oprsz = simd_oprsz(desc); | ||
108 | - uint32_t vecb = (uint32_t)DUP4(b); | ||
109 | intptr_t i; | ||
110 | |||
111 | for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
112 | - *(uint32_t *)(d + i) = *(uint32_t *)(a + i) - vecb; | ||
113 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) - (uint32_t)b; | ||
114 | } | ||
115 | clear_high(d, oprsz, desc); | ||
116 | } | ||
117 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_subs32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
118 | void HELPER(gvec_subs64)(void *d, void *a, uint64_t b, uint32_t desc) | ||
119 | { | ||
120 | intptr_t oprsz = simd_oprsz(desc); | ||
121 | - uint64_t vecb = (uint64_t)DUP2(b); | ||
122 | intptr_t i; | ||
123 | |||
124 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
125 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) - vecb; | ||
126 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) - b; | ||
127 | } | ||
128 | clear_high(d, oprsz, desc); | ||
129 | } | ||
130 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_mul64)(void *d, void *a, void *b, uint32_t desc) | ||
131 | void HELPER(gvec_muls8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
132 | { | ||
133 | intptr_t oprsz = simd_oprsz(desc); | ||
134 | - uint8_t vecb = (uint8_t)DUP16(b); | ||
135 | intptr_t i; | ||
136 | |||
137 | for (i = 0; i < oprsz; i += sizeof(uint8_t)) { | ||
138 | - *(uint8_t *)(d + i) = *(uint8_t *)(a + i) * vecb; | ||
139 | + *(uint8_t *)(d + i) = *(uint8_t *)(a + i) * (uint8_t)b; | ||
140 | } | ||
141 | clear_high(d, oprsz, desc); | ||
142 | } | ||
143 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls8)(void *d, void *a, uint64_t b, uint32_t desc) | ||
144 | void HELPER(gvec_muls16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
145 | { | ||
146 | intptr_t oprsz = simd_oprsz(desc); | ||
147 | - uint16_t vecb = (uint16_t)DUP8(b); | ||
148 | intptr_t i; | ||
149 | |||
150 | for (i = 0; i < oprsz; i += sizeof(uint16_t)) { | ||
151 | - *(uint16_t *)(d + i) = *(uint16_t *)(a + i) * vecb; | ||
152 | + *(uint16_t *)(d + i) = *(uint16_t *)(a + i) * (uint16_t)b; | ||
153 | } | ||
154 | clear_high(d, oprsz, desc); | ||
155 | } | ||
156 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls16)(void *d, void *a, uint64_t b, uint32_t desc) | ||
157 | void HELPER(gvec_muls32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
158 | { | ||
159 | intptr_t oprsz = simd_oprsz(desc); | ||
160 | - uint32_t vecb = (uint32_t)DUP4(b); | ||
161 | intptr_t i; | ||
162 | |||
163 | for (i = 0; i < oprsz; i += sizeof(uint32_t)) { | ||
164 | - *(uint32_t *)(d + i) = *(uint32_t *)(a + i) * vecb; | ||
165 | + *(uint32_t *)(d + i) = *(uint32_t *)(a + i) * (uint32_t)b; | ||
166 | } | ||
167 | clear_high(d, oprsz, desc); | ||
168 | } | ||
169 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_muls32)(void *d, void *a, uint64_t b, uint32_t desc) | ||
170 | void HELPER(gvec_muls64)(void *d, void *a, uint64_t b, uint32_t desc) | ||
171 | { | ||
172 | intptr_t oprsz = simd_oprsz(desc); | ||
173 | - uint64_t vecb = (uint64_t)DUP2(b); | ||
174 | intptr_t i; | ||
175 | |||
176 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
177 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) * vecb; | ||
178 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) * b; | ||
179 | } | ||
180 | clear_high(d, oprsz, desc); | ||
181 | } | ||
182 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_eqv)(void *d, void *a, void *b, uint32_t desc) | ||
183 | void HELPER(gvec_ands)(void *d, void *a, uint64_t b, uint32_t desc) | ||
184 | { | ||
185 | intptr_t oprsz = simd_oprsz(desc); | ||
186 | - uint64_t vecb = (uint64_t)DUP2(b); | ||
187 | intptr_t i; | ||
188 | |||
189 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
190 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) & vecb; | ||
191 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) & b; | ||
192 | } | ||
193 | clear_high(d, oprsz, desc); | ||
194 | } | ||
195 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_ands)(void *d, void *a, uint64_t b, uint32_t desc) | ||
196 | void HELPER(gvec_xors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
197 | { | ||
198 | intptr_t oprsz = simd_oprsz(desc); | ||
199 | - uint64_t vecb = (uint64_t)DUP2(b); | ||
200 | intptr_t i; | ||
201 | |||
202 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
203 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) ^ vecb; | ||
204 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) ^ b; | ||
205 | } | ||
206 | clear_high(d, oprsz, desc); | ||
207 | } | ||
208 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_xors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
209 | void HELPER(gvec_ors)(void *d, void *a, uint64_t b, uint32_t desc) | ||
210 | { | ||
211 | intptr_t oprsz = simd_oprsz(desc); | ||
212 | - uint64_t vecb = (uint64_t)DUP2(b); | ||
213 | intptr_t i; | ||
214 | |||
215 | for (i = 0; i < oprsz; i += sizeof(uint64_t)) { | ||
216 | - *(uint64_t *)(d + i) = *(uint64_t *)(a + i) | vecb; | ||
217 | + *(uint64_t *)(d + i) = *(uint64_t *)(a + i) | b; | ||
218 | } | ||
219 | clear_high(d, oprsz, desc); | ||
220 | } | ||
74 | -- | 221 | -- |
75 | 2.20.1 | 222 | 2.20.1 |
76 | 223 | ||
77 | 224 | diff view generated by jsdifflib |
1 | From: Max Filippov <jcmvbkbc@gmail.com> | 1 | Partial cleanup from the CONFIG_VECTOR16 removal. |
---|---|---|---|
2 | Replace DO_CMP0 with its scalar expansion, a simple negation. | ||
2 | 3 | ||
3 | When a breakpoint is inserted at location for which there's currently no | ||
4 | virtual to physical translation no action is taken on CPU TB cache. If a | ||
5 | TB for that virtual address already exists but is not visible ATM the | ||
6 | breakpoint won't be hit next time an instruction at that address will be | ||
7 | executed. | ||
8 | |||
9 | Flush entire CPU TB cache in breakpoint_invalidate to force | ||
10 | re-translation of all TBs for the breakpoint address. | ||
11 | |||
12 | This change fixes the following scenario: | ||
13 | - linux user application is running | ||
14 | - a breakpoint is inserted from QEMU gdbstub for a user address that is | ||
15 | not currently present in the target CPU TLB | ||
16 | - an instruction at that address is executed, but the external debugger | ||
17 | doesn't get control. | ||
18 | |||
19 | Reviewed-by: Richard Henderson <richard.henderson@linaro.org> | ||
20 | Signed-off-by: Max Filippov <jcmvbkbc@gmail.com> | ||
21 | Message-Id: <20191127220602.10827-2-jcmvbkbc@gmail.com> | ||
22 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> | 4 | Signed-off-by: Richard Henderson <richard.henderson@linaro.org> |
23 | --- | 5 | --- |
24 | exec.c | 15 +++++++-------- | 6 | accel/tcg/tcg-runtime-gvec.c | 5 +---- |
25 | 1 file changed, 7 insertions(+), 8 deletions(-) | 7 | 1 file changed, 1 insertion(+), 4 deletions(-) |
26 | 8 | ||
27 | diff --git a/exec.c b/exec.c | 9 | diff --git a/accel/tcg/tcg-runtime-gvec.c b/accel/tcg/tcg-runtime-gvec.c |
28 | index XXXXXXX..XXXXXXX 100644 | 10 | index XXXXXXX..XXXXXXX 100644 |
29 | --- a/exec.c | 11 | --- a/accel/tcg/tcg-runtime-gvec.c |
30 | +++ b/exec.c | 12 | +++ b/accel/tcg/tcg-runtime-gvec.c |
31 | @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr, MemTxAttrs attrs) | 13 | @@ -XXX,XX +XXX,XX @@ void HELPER(gvec_sar64v)(void *d, void *a, void *b, uint32_t desc) |
32 | 14 | clear_high(d, oprsz, desc); | |
33 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | ||
34 | { | ||
35 | - MemTxAttrs attrs; | ||
36 | - hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs); | ||
37 | - int asidx = cpu_asidx_from_attrs(cpu, attrs); | ||
38 | - if (phys != -1) { | ||
39 | - /* Locks grabbed by tb_invalidate_phys_addr */ | ||
40 | - tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as, | ||
41 | - phys | (pc & ~TARGET_PAGE_MASK), attrs); | ||
42 | - } | ||
43 | + /* | ||
44 | + * There may not be a virtual to physical translation for the pc | ||
45 | + * right now, but there may exist cached TB for this pc. | ||
46 | + * Flush the whole TB cache to force re-translation of such TBs. | ||
47 | + * This is heavyweight, but we're debugging anyway. | ||
48 | + */ | ||
49 | + tb_flush(cpu); | ||
50 | } | 15 | } |
51 | #endif | 16 | |
17 | -#define DO_CMP0(X) -(X) | ||
18 | - | ||
19 | #define DO_CMP1(NAME, TYPE, OP) \ | ||
20 | void HELPER(NAME)(void *d, void *a, void *b, uint32_t desc) \ | ||
21 | { \ | ||
22 | intptr_t oprsz = simd_oprsz(desc); \ | ||
23 | intptr_t i; \ | ||
24 | for (i = 0; i < oprsz; i += sizeof(TYPE)) { \ | ||
25 | - *(TYPE *)(d + i) = DO_CMP0(*(TYPE *)(a + i) OP *(TYPE *)(b + i)); \ | ||
26 | + *(TYPE *)(d + i) = -(*(TYPE *)(a + i) OP *(TYPE *)(b + i)); \ | ||
27 | } \ | ||
28 | clear_high(d, oprsz, desc); \ | ||
29 | } | ||
30 | @@ -XXX,XX +XXX,XX @@ DO_CMP2(16) | ||
31 | DO_CMP2(32) | ||
32 | DO_CMP2(64) | ||
33 | |||
34 | -#undef DO_CMP0 | ||
35 | #undef DO_CMP1 | ||
36 | #undef DO_CMP2 | ||
52 | 37 | ||
53 | -- | 38 | -- |
54 | 2.20.1 | 39 | 2.20.1 |
55 | 40 | ||
56 | 41 | diff view generated by jsdifflib |