OpenE2K/qemu-e2k
13
4
Fork 0
Code Issues 4 Pull Requests Projects 2 Releases Activity

target/arm: Convert multiply and multiply accumulate

Browse Source 
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20190904193059.26202-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Richard Henderson 2019-09-04 12:29:56 -07:00 committed by Peter Maydell
parent 581c6ebd17
commit bd92fe353b
3 changed files with 177 additions and 107 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
target/arm/a32.decode
View File

@ -25,6 +25,8 @@
&s_rrr_shi s rd rn rm shim shty &s_rrr_shi s rd rn rm shim shty
&s_rrr_shr s rn rd rm rs shty &s_rrr_shr s rn rd rm rs shty
&s_rri_rot s rn rd imm rot &s_rri_rot s rn rd imm rot
&s_rrrr s rd rn rm ra
&rrrr rd rn rm ra
# Data-processing (register) # Data-processing (register)
@ -105,3 +107,18 @@ ORR_rri .... 001 1100 . .... .... ............ @s_rri_rot
MOV_rxi .... 001 1101 . 0000 .... ............ @s_rxi_rot MOV_rxi .... 001 1101 . 0000 .... ............ @s_rxi_rot
BIC_rri .... 001 1110 . .... .... ............ @s_rri_rot BIC_rri .... 001 1110 . .... .... ............ @s_rri_rot
MVN_rxi .... 001 1111 . 0000 .... ............ @s_rxi_rot MVN_rxi .... 001 1111 . 0000 .... ............ @s_rxi_rot
# Multiply and multiply accumulate
@s_rdamn ---- .... ... s:1 rd:4 ra:4 rm:4 .... rn:4 &s_rrrr
@s_rd0mn ---- .... ... s:1 rd:4 .... rm:4 .... rn:4 &s_rrrr ra=0
@rdamn ---- .... ... . rd:4 ra:4 rm:4 .... rn:4 &rrrr
MUL .... 0000 000 . .... 0000 .... 1001 .... @s_rd0mn
MLA .... 0000 001 . .... .... .... 1001 .... @s_rdamn
UMAAL .... 0000 010 0 .... .... .... 1001 .... @rdamn
MLS .... 0000 011 0 .... .... .... 1001 .... @rdamn
UMULL .... 0000 100 . .... .... .... 1001 .... @s_rdamn
UMLAL .... 0000 101 . .... .... .... 1001 .... @s_rdamn
SMULL .... 0000 110 . .... .... .... 1001 .... @s_rdamn
SMLAL .... 0000 111 . .... .... .... 1001 .... @s_rdamn

19
target/arm/t32.decode
View File

@ -22,6 +22,8 @@
&s_rrr_shi !extern s rd rn rm shim shty &s_rrr_shi !extern s rd rn rm shim shty
&s_rrr_shr !extern s rn rd rm rs shty &s_rrr_shr !extern s rn rd rm rs shty
&s_rri_rot !extern s rn rd imm rot &s_rri_rot !extern s rn rd imm rot
&s_rrrr !extern s rd rn rm ra
&rrrr !extern rd rn rm ra
# Data-processing (register) # Data-processing (register)
@ -109,3 +111,20 @@ SBC_rri 1111 0.0 1011 . .... 0 ... .... ........ @s_rri_rot
SUB_rri 1111 0.0 1101 . .... 0 ... .... ........ @s_rri_rot SUB_rri 1111 0.0 1101 . .... 0 ... .... ........ @s_rri_rot
} }
RSB_rri 1111 0.0 1110 . .... 0 ... .... ........ @s_rri_rot RSB_rri 1111 0.0 1110 . .... 0 ... .... ........ @s_rri_rot
# Multiply and multiply accumulate
@s0_rnadm .... .... .... rn:4 ra:4 rd:4 .... rm:4 &s_rrrr s=0
@s0_rn0dm .... .... .... rn:4 .... rd:4 .... rm:4 &s_rrrr ra=0 s=0
@rnadm .... .... .... rn:4 ra:4 rd:4 .... rm:4 &rrrr
{
MUL 1111 1011 0000 .... 1111 .... 0000 .... @s0_rn0dm
MLA 1111 1011 0000 .... .... .... 0000 .... @s0_rnadm
}
MLS 1111 1011 0000 .... .... .... 0001 .... @rnadm
SMULL 1111 1011 1000 .... .... .... 0000 .... @s0_rnadm
UMULL 1111 1011 1010 .... .... .... 0000 .... @s0_rnadm
SMLAL 1111 1011 1100 .... .... .... 0000 .... @s0_rnadm
UMLAL 1111 1011 1110 .... .... .... 0000 .... @s0_rnadm
UMAAL 1111 1011 1110 .... .... .... 0110 .... @rnadm

226
target/arm/translate.c
View File

@ -8019,6 +8019,125 @@ static bool trans_ORN_rri(DisasContext *s, arg_s_rri_rot *a)
#undef DO_ANY2 #undef DO_ANY2
#undef DO_CMP2 #undef DO_CMP2
/*
* Multiply and multiply accumulate
*/
static bool op_mla(DisasContext *s, arg_s_rrrr *a, bool add)
{
TCGv_i32 t1, t2;
t1 = load_reg(s, a->rn);
t2 = load_reg(s, a->rm);
tcg_gen_mul_i32(t1, t1, t2);
tcg_temp_free_i32(t2);
if (add) {
t2 = load_reg(s, a->ra);
tcg_gen_add_i32(t1, t1, t2);
tcg_temp_free_i32(t2);
}
if (a->s) {
gen_logic_CC(t1);
}
store_reg(s, a->rd, t1);
return true;
}
static bool trans_MUL(DisasContext *s, arg_MUL *a)
{
return op_mla(s, a, false);
}
static bool trans_MLA(DisasContext *s, arg_MLA *a)
{
return op_mla(s, a, true);
}
static bool trans_MLS(DisasContext *s, arg_MLS *a)
{
TCGv_i32 t1, t2;
if (!ENABLE_ARCH_6T2) {
return false;
}
t1 = load_reg(s, a->rn);
t2 = load_reg(s, a->rm);
tcg_gen_mul_i32(t1, t1, t2);
tcg_temp_free_i32(t2);
t2 = load_reg(s, a->ra);
tcg_gen_sub_i32(t1, t2, t1);
tcg_temp_free_i32(t2);
store_reg(s, a->rd, t1);
return true;
}
static bool op_mlal(DisasContext *s, arg_s_rrrr *a, bool uns, bool add)
{
TCGv_i32 t0, t1, t2, t3;
t0 = load_reg(s, a->rm);
t1 = load_reg(s, a->rn);
if (uns) {
tcg_gen_mulu2_i32(t0, t1, t0, t1);
} else {
tcg_gen_muls2_i32(t0, t1, t0, t1);
}
if (add) {
t2 = load_reg(s, a->ra);
t3 = load_reg(s, a->rd);
tcg_gen_add2_i32(t0, t1, t0, t1, t2, t3);
tcg_temp_free_i32(t2);
tcg_temp_free_i32(t3);
}
if (a->s) {
gen_logicq_cc(t0, t1);
}
store_reg(s, a->ra, t0);
store_reg(s, a->rd, t1);
return true;
}
static bool trans_UMULL(DisasContext *s, arg_UMULL *a)
{
return op_mlal(s, a, true, false);
}
static bool trans_SMULL(DisasContext *s, arg_SMULL *a)
{
return op_mlal(s, a, false, false);
}
static bool trans_UMLAL(DisasContext *s, arg_UMLAL *a)
{
return op_mlal(s, a, true, true);
}
static bool trans_SMLAL(DisasContext *s, arg_SMLAL *a)
{
return op_mlal(s, a, false, true);
}
static bool trans_UMAAL(DisasContext *s, arg_UMAAL *a)
{
TCGv_i32 t0, t1;
TCGv_i64 t64;
if (s->thumb
? !arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)
: !ENABLE_ARCH_6) {
return false;
}
t0 = load_reg(s, a->rm);
t1 = load_reg(s, a->rn);
t64 = gen_mulu_i64_i32(t0, t1);
gen_addq_lo(s, t64, a->ra);
gen_addq_lo(s, t64, a->rd);
gen_storeq_reg(s, a->ra, a->rd, t64);
tcg_temp_free_i64(t64);
return true;
}
/* /*
* Legacy decoder. * Legacy decoder.
*/ */
@ -8565,71 +8684,9 @@ static void disas_arm_insn(DisasContext *s, unsigned int insn)
sh = (insn >> 5) & 3; sh = (insn >> 5) & 3;
if (sh == 0) { if (sh == 0) {
if (op1 == 0x0) { if (op1 == 0x0) {
rd = (insn >> 16) & 0xf; /* Multiply and multiply accumulate. */
rn = (insn >> 12) & 0xf; /* All done in decodetree. Reach here for illegal ops. */
rs = (insn >> 8) & 0xf;
rm = (insn) & 0xf;
op1 = (insn >> 20) & 0xf;
switch (op1) {
case 0: case 1: case 2: case 3: case 6:
/* 32 bit mul */
tmp = load_reg(s, rs);
tmp2 = load_reg(s, rm);
tcg_gen_mul_i32(tmp, tmp, tmp2);
tcg_temp_free_i32(tmp2);
if (insn & (1 << 22)) {
/* Subtract (mls) */
ARCH(6T2);
tmp2 = load_reg(s, rn);
tcg_gen_sub_i32(tmp, tmp2, tmp);
tcg_temp_free_i32(tmp2);
} else if (insn & (1 << 21)) {
/* Add */
tmp2 = load_reg(s, rn);
tcg_gen_add_i32(tmp, tmp, tmp2);
tcg_temp_free_i32(tmp2);
}
if (insn & (1 << 20))
gen_logic_CC(tmp);
store_reg(s, rd, tmp);
break;
case 4:
/* 64 bit mul double accumulate (UMAAL) */
ARCH(6);
tmp = load_reg(s, rs);
tmp2 = load_reg(s, rm);
tmp64 = gen_mulu_i64_i32(tmp, tmp2);
gen_addq_lo(s, tmp64, rn);
gen_addq_lo(s, tmp64, rd);
gen_storeq_reg(s, rn, rd, tmp64);
tcg_temp_free_i64(tmp64);
break;
case 8: case 9: case 10: case 11:
case 12: case 13: case 14: case 15:
/* 64 bit mul: UMULL, UMLAL, SMULL, SMLAL. */
tmp = load_reg(s, rs);
tmp2 = load_reg(s, rm);
if (insn & (1 << 22)) {
tcg_gen_muls2_i32(tmp, tmp2, tmp, tmp2);
} else {
tcg_gen_mulu2_i32(tmp, tmp2, tmp, tmp2);
}
if (insn & (1 << 21)) { /* mult accumulate */
TCGv_i32 al = load_reg(s, rn);
TCGv_i32 ah = load_reg(s, rd);
tcg_gen_add2_i32(tmp, tmp2, tmp, tmp2, al, ah);
tcg_temp_free_i32(al);
tcg_temp_free_i32(ah);
}
if (insn & (1 << 20)) {
gen_logicq_cc(tmp, tmp2);
}
store_reg(s, rn, tmp);
store_reg(s, rd, tmp2);
break;
default:
goto illegal_op; goto illegal_op;
}
} else { } else {
rn = (insn >> 16) & 0xf; rn = (insn >> 16) & 0xf;
rd = (insn >> 12) & 0xf; rd = (insn >> 12) & 0xf;
@ -10075,7 +10132,8 @@ static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
break; break;
case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */ case 4: case 5: /* 32-bit multiply. Sum of absolute differences. */
switch ((insn >> 20) & 7) { switch ((insn >> 20) & 7) {
case 0: /* 32 x 32 -> 32 */ case 0: /* 32 x 32 -> 32, in decodetree */
goto illegal_op;
case 7: /* Unsigned sum of absolute differences. */ case 7: /* Unsigned sum of absolute differences. */
break; break;
case 1: /* 16 x 16 -> 32 */ case 1: /* 16 x 16 -> 32 */
@ -10092,18 +10150,6 @@ static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
tmp = load_reg(s, rn); tmp = load_reg(s, rn);
tmp2 = load_reg(s, rm); tmp2 = load_reg(s, rm);
switch ((insn >> 20) & 7) { switch ((insn >> 20) & 7) {
case 0: /* 32 x 32 -> 32 */
tcg_gen_mul_i32(tmp, tmp, tmp2);
tcg_temp_free_i32(tmp2);
if (rs != 15) {
tmp2 = load_reg(s, rs);
if (op)
tcg_gen_sub_i32(tmp, tmp2, tmp);
else
tcg_gen_add_i32(tmp, tmp, tmp2);
tcg_temp_free_i32(tmp2);
}
break;
case 1: /* 16 x 16 -> 32 */ case 1: /* 16 x 16 -> 32 */
gen_mulxy(tmp, tmp2, op & 2, op & 1); gen_mulxy(tmp, tmp2, op & 2, op & 1);
tcg_temp_free_i32(tmp2); tcg_temp_free_i32(tmp2);
@ -10233,11 +10279,12 @@ static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
gen_storeq_reg(s, rs, rd, tmp64); gen_storeq_reg(s, rs, rd, tmp64);
tcg_temp_free_i64(tmp64); tcg_temp_free_i64(tmp64);
} else { } else {
if (op & 0x20) { if ((op & 0x20) || !(op & 8)) {
/* Unsigned 64-bit multiply */ /* Signed/unsigned 64-bit multiply, in decodetree */
tmp64 = gen_mulu_i64_i32(tmp, tmp2); tcg_temp_free_i32(tmp2);
} else { tcg_temp_free_i32(tmp);
if (op & 8) { goto illegal_op;
}
/* smlalxy */ /* smlalxy */
if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) { if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
tcg_temp_free_i32(tmp2); tcg_temp_free_i32(tmp2);
@ -10249,20 +10296,7 @@ static void disas_thumb2_insn(DisasContext *s, uint32_t insn)
tmp64 = tcg_temp_new_i64(); tmp64 = tcg_temp_new_i64();
tcg_gen_ext_i32_i64(tmp64, tmp); tcg_gen_ext_i32_i64(tmp64, tmp);
tcg_temp_free_i32(tmp); tcg_temp_free_i32(tmp);
} else { if (op & 0x40) {
/* Signed 64-bit multiply */
tmp64 = gen_muls_i64_i32(tmp, tmp2);
}
}
if (op & 4) {
/* umaal */
if (!arm_dc_feature(s, ARM_FEATURE_THUMB_DSP)) {
tcg_temp_free_i64(tmp64);
goto illegal_op;
}
gen_addq_lo(s, tmp64, rs);
gen_addq_lo(s, tmp64, rd);
} else if (op & 0x40) {
/* 64-bit accumulate. */ /* 64-bit accumulate. */
gen_addq(s, tmp64, rs, rd); gen_addq(s, tmp64, rs, rd);
} }