| Message ID | 20230925050545.30912-4-nbowler@draconx.ca |
|---|---|
| State | New |
| Headers | show |
| Series | SPARC VIS fixes | expand |
On 9/24/23 01:03, Nick Bowler wrote: > On a real UltraSparc II, the fmul8x16al instruction takes two single- > precision input operands and returns a double-precision result. For > the second operand, bits 15:0 are used, and bits 31:16 are ignored. > > However, the emulation is taking two double-precision input operands, > and furthermore it is using bits 31:16 of the second operand (ignoring > bits 15:0). These are unlikely to contain the correct values. > > Even still, the emulator overwrites the second input before all outputs > are calculated, so even if by chance the data loaded in happens to be > correct, the results are just garbage except in trivial cases. > > Signed-off-by: Nick Bowler<nbowler@draconx.ca> > --- > target/sparc/helper.h | 2 +- > target/sparc/translate.c | 2 +- > target/sparc/vis_helper.c | 11 ++++++----- > 3 files changed, 8 insertions(+), 7 deletions(-) Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
diff --git a/target/sparc/helper.h b/target/sparc/helper.h index 76e06b8ea5..25d6178ca5 100644 --- a/target/sparc/helper.h +++ b/target/sparc/helper.h @@ -127,7 +127,7 @@ DEF_HELPER_FLAGS_1(fqtox, TCG_CALL_NO_RWG, s64, env) DEF_HELPER_FLAGS_2(fpmerge, TCG_CALL_NO_RWG_SE, i64, i64, i64) DEF_HELPER_FLAGS_2(fmul8x16, TCG_CALL_NO_RWG_SE, i64, i32, i64) -DEF_HELPER_FLAGS_2(fmul8x16al, TCG_CALL_NO_RWG_SE, i64, i64, i64) +DEF_HELPER_FLAGS_2(fmul8x16al, TCG_CALL_NO_RWG_SE, i64, i32, i32) DEF_HELPER_FLAGS_2(fmul8x16au, TCG_CALL_NO_RWG_SE, i64, i32, i32) DEF_HELPER_FLAGS_2(fmul8sux16, TCG_CALL_NO_RWG_SE, i64, i64, i64) DEF_HELPER_FLAGS_2(fmul8ulx16, TCG_CALL_NO_RWG_SE, i64, i64, i64) diff --git a/target/sparc/translate.c b/target/sparc/translate.c index ca81b35a25..dddee9f974 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -4779,7 +4779,7 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn) break; case 0x035: /* VIS I fmul8x16al */ CHECK_FPU_FEATURE(dc, VIS1); - gen_ne_fop_DDD(dc, rd, rs1, rs2, gen_helper_fmul8x16al); + gen_ne_fop_DFF(dc, rd, rs1, rs2, gen_helper_fmul8x16al); break; case 0x036: /* VIS I fmul8sux16 */ CHECK_FPU_FEATURE(dc, VIS1); diff --git a/target/sparc/vis_helper.c b/target/sparc/vis_helper.c index 2fc783a054..386cfd0706 100644 --- a/target/sparc/vis_helper.c +++ b/target/sparc/vis_helper.c @@ -122,16 +122,17 @@ uint64_t helper_fmul8x16(uint32_t src1, uint64_t src2) return d.ll; } -uint64_t helper_fmul8x16al(uint64_t src1, uint64_t src2) +uint64_t helper_fmul8x16al(uint32_t src1, uint32_t src2) { - VIS64 s, d; + VIS32 s1, s2; + VIS64 d; uint32_t tmp; - s.ll = src1; - d.ll = src2; + s1.l = src1; + s2.l = src2; #define PMUL(r) \ - tmp = (int32_t)d.VIS_SW64(1) * (int32_t)s.VIS_B64(r); \ + tmp = (int32_t)s2.VIS_SW32(0) * (int32_t)s1.VIS_B64(r); \ if ((tmp & 0xff) > 0x7f) { \ tmp += 0x100; \ } \
On a real UltraSparc II, the fmul8x16al instruction takes two single- precision input operands and returns a double-precision result. For the second operand, bits 15:0 are used, and bits 31:16 are ignored. However, the emulation is taking two double-precision input operands, and furthermore it is using bits 31:16 of the second operand (ignoring bits 15:0). These are unlikely to contain the correct values. Even still, the emulator overwrites the second input before all outputs are calculated, so even if by chance the data loaded in happens to be correct, the results are just garbage except in trivial cases. Signed-off-by: Nick Bowler <nbowler@draconx.ca> --- target/sparc/helper.h | 2 +- target/sparc/translate.c | 2 +- target/sparc/vis_helper.c | 11 ++++++----- 3 files changed, 8 insertions(+), 7 deletions(-)