riscv
/
riscv-isa-sim
mirror of https://github.com/riscv-software-src/riscv-isa-sim.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
Browse Source

Don't update any arch state on vl=0 dot products 

Fixes #2334

Signed-off-by: Andrew Waterman <andrew@sifive.com>
pull/2339/head
Andrew Waterman 4 weeks ago
parent
b0093c40c3
commit
e9dfa38bb5
Failed to extract signature
1 changed files with 25 additions and 21 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 46
      riscv/v_ext_macros.h

46
riscv/v_ext_macros.h
View File

@ -2245,16 +2245,18 @@ c_t generic_dot_product(const std::vector<a_t>& a, const std::vector<b_t>& b, c_
} }
#define ZVLDOT_LOOP(a_t, b_t, c_t, dot) \ #define ZVLDOT_LOOP(a_t, b_t, c_t, dot) \
std::vector<a_t> a(P.VU.vlmax, a_t()); \ if (auto vl = P.VU.vl->read()) { \
std::vector<b_t> b(P.VU.vlmax, b_t()); \ std::vector<a_t> a(P.VU.vlmax, a_t()); \
for (reg_t i = 0, vl = P.VU.vl->read(); i < vl; i++) { \ std::vector<b_t> b(P.VU.vlmax, b_t()); \
VI_LOOP_ELEMENT_SKIP(); \ for (reg_t i = 0; i < vl; i++) { \
a[i] = P.VU.elt<a_t>(insn.rs1(), i); \ VI_LOOP_ELEMENT_SKIP(); \
b[i] = P.VU.elt<b_t>(insn.rs2(), i); \ a[i] = P.VU.elt<a_t>(insn.rs1(), i); \
} \ b[i] = P.VU.elt<b_t>(insn.rs2(), i); \
auto& acc = P.VU.elt<c_t>(insn.rd(), 0, true); \ } \
acc = dot(a, b, acc); \ auto& acc = P.VU.elt<c_t>(insn.rd(), 0, true); \
set_fp_exceptions; acc = dot(a, b, acc); \
set_fp_exceptions; \
}
#define ZVLDOT_GENERIC_LOOP(a_t, b_t, c_t, macc) \ #define ZVLDOT_GENERIC_LOOP(a_t, b_t, c_t, macc) \
auto dot = std::bind(generic_dot_product<a_t, b_t, c_t>, std::placeholders::_1, std::placeholders::_2, std::placeholders::_3, macc); \ auto dot = std::bind(generic_dot_product<a_t, b_t, c_t>, std::placeholders::_1, std::placeholders::_2, std::placeholders::_3, macc); \
@ -2265,18 +2267,20 @@ c_t generic_dot_product(const std::vector<a_t>& a, const std::vector<b_t>& b, c_
ZVLDOT_GENERIC_LOOP(a_t, b_t, c_t, macc) ZVLDOT_GENERIC_LOOP(a_t, b_t, c_t, macc)
#define ZVBDOT_LOOP(a_t, b_t, c_t, dot) \ #define ZVBDOT_LOOP(a_t, b_t, c_t, dot) \
for (reg_t idx = 0; idx < 8; idx++) { \ if (auto vl = P.VU.vl->read()) { \
reg_t i = ci + idx; \ for (reg_t idx = 0; idx < 8; idx++) { \
VI_LOOP_ELEMENT_SKIP(); \ reg_t i = ci + idx; \
std::vector<a_t> a(P.VU.vlmax, a_t()); \ VI_LOOP_ELEMENT_SKIP(); \
std::vector<b_t> b(P.VU.vlmax, b_t()); \ std::vector<a_t> a(P.VU.vlmax, a_t()); \
for (reg_t k = 0, vl = P.VU.vl->read(); k < vl; k++) { \ std::vector<b_t> b(P.VU.vlmax, b_t()); \
a[k] = P.VU.elt<a_t>(insn.rs1(), k); \ for (reg_t k = 0; k < vl; k++) { \
b[k] = P.VU.elt<b_t>(vs2 + idx, k); \ a[k] = P.VU.elt<a_t>(insn.rs1(), k); \
b[k] = P.VU.elt<b_t>(vs2 + idx, k); \
} \
auto& acc = P.VU.elt<c_t>(insn.rd(), i, true); \
acc = dot(a, b, acc); \
set_fp_exceptions; \
} \ } \
auto& acc = P.VU.elt<c_t>(insn.rd(), i, true); \
acc = dot(a, b, acc); \
set_fp_exceptions; \
} }
#define ZVBDOT_GENERIC_LOOP(a_t, b_t, c_t, macc) \ #define ZVBDOT_GENERIC_LOOP(a_t, b_t, c_t, macc) \

Write Preview
Loading…
Cancel
Save