target/hexagon: use TCG_COND_TSTEQ/TSTNE for predicate branches
Replace TCG_COND_EQ/NE comparisons against 0 with TCG_COND_TSTEQ/TSTNE comparisons against 1 for all predicate-conditional branches. This tests bit 0 of the predicate register directly, eliminating redundant andi operations that previously extracted the LSB before the comparison. For predicate-conditional jumps (jumpt, jumpf, jumptnew, etc.) and jump-register variants (jumprt, jumprf, etc.), pass the raw predicate value directly instead of going through fLSBOLD/fLSBNEW extraction. For callers that produce a 0/1 result via setcond (compare-and-jump, jumprz, etc.), the TSTEQ/TSTNE test on bit 0 is equivalent to the previous EQ/NE test against 0. Reviewed-by: Taylor Simpson <ltaylorsimpson@gmail.com> Signed-off-by: Brian Cain <brian.cain@oss.qualcomm.com>
Brian Cain committed
Mar 2, 2026 at 21:29 UTC
dfa4c8810f0a3bcc47a54b0a8792b14208de1be6
3 files changed
+50
-78
target/hexagon/gen_tcg.h
+37
-57
@@ -696,13 +696,13 @@
696
gen_callr(ctx, RsV)
697
698
#define fGEN_TCG_J2_callt(SHORTCODE) \
699
- gen_cond_call(ctx, PuV, TCG_COND_EQ, riV)
699
+ gen_cond_call(ctx, PuV, TCG_COND_TSTEQ, riV)
700
#define fGEN_TCG_J2_callf(SHORTCODE) \
701
- gen_cond_call(ctx, PuV, TCG_COND_NE, riV)
701
+ gen_cond_call(ctx, PuV, TCG_COND_TSTNE, riV)
702
#define fGEN_TCG_J2_callrt(SHORTCODE) \
703
- gen_cond_callr(ctx, TCG_COND_EQ, PuV, RsV)
703
+ gen_cond_callr(ctx, TCG_COND_TSTEQ, PuV, RsV)
704
#define fGEN_TCG_J2_callrf(SHORTCODE) \
705
- gen_cond_callr(ctx, TCG_COND_NE, PuV, RsV)
705
+ gen_cond_callr(ctx, TCG_COND_TSTNE, PuV, RsV)
706
707
#define fGEN_TCG_J2_loop0r(SHORTCODE) \
708
gen_loop0r(ctx, RsV, riV)
@@ -893,21 +893,21 @@
893
gen_cmpnd_cmp_n1_jmp_f(ctx, 1, TCG_COND_GT, RsV, riV)
894
895
#define fGEN_TCG_J4_tstbit0_tp0_jump_nt(SHORTCODE) \
896
- gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_EQ, riV)
896
+ gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_TSTEQ, riV)
897
#define fGEN_TCG_J4_tstbit0_tp0_jump_t(SHORTCODE) \
898
- gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_EQ, riV)
898
+ gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_TSTEQ, riV)
899
#define fGEN_TCG_J4_tstbit0_fp0_jump_nt(SHORTCODE) \
900
- gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_NE, riV)
900
+ gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_TSTNE, riV)
901
#define fGEN_TCG_J4_tstbit0_fp0_jump_t(SHORTCODE) \
902
- gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_NE, riV)
902
+ gen_cmpnd_tstbit0_jmp(ctx, 0, RsV, TCG_COND_TSTNE, riV)
903
#define fGEN_TCG_J4_tstbit0_tp1_jump_nt(SHORTCODE) \
904
- gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_EQ, riV)
904
+ gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_TSTEQ, riV)
905
#define fGEN_TCG_J4_tstbit0_tp1_jump_t(SHORTCODE) \
906
- gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_EQ, riV)
906
+ gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_TSTEQ, riV)
907
#define fGEN_TCG_J4_tstbit0_fp1_jump_nt(SHORTCODE) \
908
- gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_NE, riV)
908
+ gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_TSTNE, riV)
909
#define fGEN_TCG_J4_tstbit0_fp1_jump_t(SHORTCODE) \
910
- gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_NE, riV)
910
+ gen_cmpnd_tstbit0_jmp(ctx, 1, RsV, TCG_COND_TSTNE, riV)
911
912
/* p0 = cmp.eq(r0, #7) */
913
#define fGEN_TCG_SA1_cmpeqi(SHORTCODE) \
@@ -933,31 +933,24 @@
933
do { \
934
TCGv LSB = tcg_temp_new(); \
935
COND; \
936
- gen_cond_jump(ctx, TCG_COND_EQ, LSB, riV); \
936
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, LSB, riV); \
937
} while (0)
938
-#define fGEN_TCG_cond_jumpf(COND) \
939
- do { \
940
- TCGv LSB = tcg_temp_new(); \
941
- COND; \
942
- gen_cond_jump(ctx, TCG_COND_NE, LSB, riV); \
943
- } while (0)
944
-
938
#define fGEN_TCG_J2_jumpt(SHORTCODE) \
946
- fGEN_TCG_cond_jumpt(fLSBOLD(PuV))
939
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, PuV, riV)
940
#define fGEN_TCG_J2_jumptpt(SHORTCODE) \
948
- fGEN_TCG_cond_jumpt(fLSBOLD(PuV))
941
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, PuV, riV)
942
#define fGEN_TCG_J2_jumpf(SHORTCODE) \
950
- fGEN_TCG_cond_jumpf(fLSBOLD(PuV))
943
+ gen_cond_jump(ctx, TCG_COND_TSTNE, PuV, riV)
944
#define fGEN_TCG_J2_jumpfpt(SHORTCODE) \
952
- fGEN_TCG_cond_jumpf(fLSBOLD(PuV))
945
+ gen_cond_jump(ctx, TCG_COND_TSTNE, PuV, riV)
946
#define fGEN_TCG_J2_jumptnew(SHORTCODE) \
954
- fGEN_TCG_cond_jumpt(fLSBNEW(PuN))
947
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, PuN, riV)
948
#define fGEN_TCG_J2_jumptnewpt(SHORTCODE) \
956
- fGEN_TCG_cond_jumpt(fLSBNEW(PuN))
949
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, PuN, riV)
950
#define fGEN_TCG_J2_jumpfnewpt(SHORTCODE) \
958
- fGEN_TCG_cond_jumpf(fLSBNEW(PuN))
951
+ gen_cond_jump(ctx, TCG_COND_TSTNE, PuN, riV)
952
#define fGEN_TCG_J2_jumpfnew(SHORTCODE) \
960
- fGEN_TCG_cond_jumpf(fLSBNEW(PuN))
953
+ gen_cond_jump(ctx, TCG_COND_TSTNE, PuN, riV)
954
#define fGEN_TCG_J2_jumprz(SHORTCODE) \
955
fGEN_TCG_cond_jumpt(tcg_gen_setcondi_tl(TCG_COND_NE, LSB, RsV, 0))
956
#define fGEN_TCG_J2_jumprzpt(SHORTCODE) \
@@ -975,35 +968,22 @@
968
#define fGEN_TCG_J2_jumprltezpt(SHORTCODE) \
969
fGEN_TCG_cond_jumpt(tcg_gen_setcondi_tl(TCG_COND_LE, LSB, RsV, 0))
970
978
-#define fGEN_TCG_cond_jumprt(COND) \
979
- do { \
980
- TCGv LSB = tcg_temp_new(); \
981
- COND; \
982
- gen_cond_jumpr(ctx, RsV, TCG_COND_EQ, LSB); \
983
- } while (0)
984
-#define fGEN_TCG_cond_jumprf(COND) \
985
- do { \
986
- TCGv LSB = tcg_temp_new(); \
987
- COND; \
988
- gen_cond_jumpr(ctx, RsV, TCG_COND_NE, LSB); \
989
- } while (0)
990
-
971
#define fGEN_TCG_J2_jumprt(SHORTCODE) \
992
- fGEN_TCG_cond_jumprt(fLSBOLD(PuV))
972
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTEQ, PuV)
973
#define fGEN_TCG_J2_jumprtpt(SHORTCODE) \
994
- fGEN_TCG_cond_jumprt(fLSBOLD(PuV))
974
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTEQ, PuV)
975
#define fGEN_TCG_J2_jumprf(SHORTCODE) \
996
- fGEN_TCG_cond_jumprf(fLSBOLD(PuV))
976
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTNE, PuV)
977
#define fGEN_TCG_J2_jumprfpt(SHORTCODE) \
998
- fGEN_TCG_cond_jumprf(fLSBOLD(PuV))
978
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTNE, PuV)
979
#define fGEN_TCG_J2_jumprtnew(SHORTCODE) \
1000
- fGEN_TCG_cond_jumprt(fLSBNEW(PuN))
980
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTEQ, PuN)
981
#define fGEN_TCG_J2_jumprtnewpt(SHORTCODE) \
1002
- fGEN_TCG_cond_jumprt(fLSBNEW(PuN))
982
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTEQ, PuN)
983
#define fGEN_TCG_J2_jumprfnew(SHORTCODE) \
1004
- fGEN_TCG_cond_jumprf(fLSBNEW(PuN))
984
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTNE, PuN)
985
#define fGEN_TCG_J2_jumprfnewpt(SHORTCODE) \
1006
- fGEN_TCG_cond_jumprf(fLSBNEW(PuN))
986
+ gen_cond_jumpr(ctx, RsV, TCG_COND_TSTNE, PuN)
987
988
/*
989
* New value compare & jump instructions
@@ -1101,13 +1081,13 @@
1081
gen_cmpi_jumpnv(ctx, TCG_COND_LE, NsN, -1, riV)
1082
1083
#define fGEN_TCG_J4_tstbit0_t_jumpnv_t(SHORTCODE) \
1104
- gen_testbit0_jumpnv(ctx, NsN, TCG_COND_EQ, riV)
1084
+ gen_testbit0_jumpnv(ctx, NsN, TCG_COND_TSTEQ, riV)
1085
#define fGEN_TCG_J4_tstbit0_t_jumpnv_nt(SHORTCODE) \
1106
- gen_testbit0_jumpnv(ctx, NsN, TCG_COND_EQ, riV)
1086
+ gen_testbit0_jumpnv(ctx, NsN, TCG_COND_TSTEQ, riV)
1087
#define fGEN_TCG_J4_tstbit0_f_jumpnv_t(SHORTCODE) \
1108
- gen_testbit0_jumpnv(ctx, NsN, TCG_COND_NE, riV)
1088
+ gen_testbit0_jumpnv(ctx, NsN, TCG_COND_TSTNE, riV)
1089
#define fGEN_TCG_J4_tstbit0_f_jumpnv_nt(SHORTCODE) \
1110
- gen_testbit0_jumpnv(ctx, NsN, TCG_COND_NE, riV)
1090
+ gen_testbit0_jumpnv(ctx, NsN, TCG_COND_TSTNE, riV)
1091
1092
/* r0 = r1 ; jump address */
1093
#define fGEN_TCG_J4_jumpsetr(SHORTCODE) \
@@ -1150,14 +1130,14 @@
1130
gen_jumpr(ctx, hex_gpr[HEX_REG_LR])
1131
1132
#define fGEN_TCG_SL2_jumpr31_t(SHORTCODE) \
1153
- gen_cond_jumpr31(ctx, TCG_COND_EQ, hex_pred[0])
1133
+ gen_cond_jumpr31(ctx, TCG_COND_TSTEQ, hex_pred[0])
1134
#define fGEN_TCG_SL2_jumpr31_f(SHORTCODE) \
1155
- gen_cond_jumpr31(ctx, TCG_COND_NE, hex_pred[0])
1135
+ gen_cond_jumpr31(ctx, TCG_COND_TSTNE, hex_pred[0])
1136
1137
#define fGEN_TCG_SL2_jumpr31_tnew(SHORTCODE) \
1158
- gen_cond_jumpr31(ctx, TCG_COND_EQ, ctx->new_pred_value[0])
1138
+ gen_cond_jumpr31(ctx, TCG_COND_TSTEQ, ctx->new_pred_value[0])
1139
#define fGEN_TCG_SL2_jumpr31_fnew(SHORTCODE) \
1160
- gen_cond_jumpr31(ctx, TCG_COND_NE, ctx->new_pred_value[0])
1140
+ gen_cond_jumpr31(ctx, TCG_COND_TSTNE, ctx->new_pred_value[0])
1141
1142
/* Count trailing zeros/ones */
1143
#define fGEN_TCG_S2_ct0(SHORTCODE) \
target/hexagon/genptr.c
+12
-20
@@ -455,7 +455,7 @@ static void gen_write_new_pc_addr(DisasContext *ctx, TCGv addr,
455
TCGLabel *pred_false = NULL;
456
if (cond != TCG_COND_ALWAYS) {
457
pred_false = gen_new_label();
458
- tcg_gen_brcondi_tl(cond, pred, 0, pred_false);
458
+ tcg_gen_brcondi_tl(cond, pred, 1, pred_false);
459
}
460
461
if (ctx->pkt->pkt_has_multi_cof) {
@@ -579,9 +579,7 @@ static void gen_cond_jumpr(DisasContext *ctx, TCGv dst_pc,
579
580
static void gen_cond_jumpr31(DisasContext *ctx, TCGCond cond, TCGv pred)
581
{
582
- TCGv LSB = tcg_temp_new();
583
- tcg_gen_andi_tl(LSB, pred, 1);
584
- gen_cond_jumpr(ctx, hex_gpr[HEX_REG_LR], cond, LSB);
582
+ gen_cond_jumpr(ctx, hex_gpr[HEX_REG_LR], cond, pred);
583
}
584
585
static void gen_cond_jump(DisasContext *ctx, TCGCond cond, TCGv pred,
@@ -609,14 +607,14 @@ static void gen_cmpnd_cmp_jmp_t(DisasContext *ctx,
607
int pnum, TCGCond cond, TCGv arg1, TCGv arg2,
608
int pc_off)
609
{
612
- gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, arg2, TCG_COND_EQ, pc_off);
610
+ gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, arg2, TCG_COND_TSTEQ, pc_off);
611
}
612
613
static void gen_cmpnd_cmp_jmp_f(DisasContext *ctx,
614
int pnum, TCGCond cond, TCGv arg1, TCGv arg2,
615
int pc_off)
616
{
619
- gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, arg2, TCG_COND_NE, pc_off);
617
+ gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, arg2, TCG_COND_TSTNE, pc_off);
618
}
619
620
static void gen_cmpnd_cmpi_jmp_t(DisasContext *ctx,
@@ -624,7 +622,7 @@ static void gen_cmpnd_cmpi_jmp_t(DisasContext *ctx,
622
int pc_off)
623
{
624
TCGv tmp = tcg_constant_tl(arg2);
627
- gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, tmp, TCG_COND_EQ, pc_off);
625
+ gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, tmp, TCG_COND_TSTEQ, pc_off);
626
}
627
628
static void gen_cmpnd_cmpi_jmp_f(DisasContext *ctx,
@@ -632,7 +630,7 @@ static void gen_cmpnd_cmpi_jmp_f(DisasContext *ctx,
630
int pc_off)
631
{
632
TCGv tmp = tcg_constant_tl(arg2);
635
- gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, tmp, TCG_COND_NE, pc_off);
633
+ gen_cmpnd_cmp_jmp(ctx, pnum, cond, arg1, tmp, TCG_COND_TSTNE, pc_off);
634
}
635
636
static void gen_cmpnd_cmp_n1_jmp_t(DisasContext *ctx, int pnum, TCGCond cond,
@@ -665,9 +663,7 @@ static void gen_cmpnd_tstbit0_jmp(DisasContext *ctx,
663
static void gen_testbit0_jumpnv(DisasContext *ctx,
664
TCGv arg, TCGCond cond, int pc_off)
665
{
668
- TCGv pred = tcg_temp_new();
669
- tcg_gen_andi_tl(pred, arg, 1);
670
- gen_cond_jump(ctx, cond, pred, pc_off);
666
+ gen_cond_jump(ctx, cond, arg, pc_off);
667
}
668
669
static void gen_jump(DisasContext *ctx, int pc_off)
@@ -698,11 +694,9 @@ static void gen_cond_call(DisasContext *ctx, TCGv pred,
694
TCGCond cond, int pc_off)
695
{
696
TCGv lr = get_result_gpr(ctx, HEX_REG_LR);
701
- TCGv lsb = tcg_temp_new();
697
TCGLabel *skip = gen_new_label();
703
- tcg_gen_andi_tl(lsb, pred, 1);
704
- gen_write_new_pc_pcrel(ctx, pc_off, cond, lsb);
705
- tcg_gen_brcondi_tl(cond, lsb, 0, skip);
698
+ gen_write_new_pc_pcrel(ctx, pc_off, cond, pred);
699
+ tcg_gen_brcondi_tl(cond, pred, 1, skip);
700
tcg_gen_movi_tl(lr, ctx->next_PC);
701
gen_set_label(skip);
702
}
@@ -710,10 +704,8 @@ static void gen_cond_call(DisasContext *ctx, TCGv pred,
704
static void gen_cond_callr(DisasContext *ctx,
705
TCGCond cond, TCGv pred, TCGv new_pc)
706
{
713
- TCGv lsb = tcg_temp_new();
707
TCGLabel *skip = gen_new_label();
715
- tcg_gen_andi_tl(lsb, pred, 1);
716
- tcg_gen_brcondi_tl(cond, lsb, 0, skip);
708
+ tcg_gen_brcondi_tl(cond, pred, 1, skip);
709
gen_callr(ctx, new_pc);
710
gen_set_label(skip);
711
}
@@ -955,7 +947,7 @@ static void gen_cmp_jumpnv(DisasContext *ctx,
947
{
948
TCGv pred = tcg_temp_new();
949
tcg_gen_setcond_tl(cond, pred, val, src);
958
- gen_cond_jump(ctx, TCG_COND_EQ, pred, pc_off);
950
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, pred, pc_off);
951
}
952
953
static void gen_cmpi_jumpnv(DisasContext *ctx,
@@ -963,7 +955,7 @@ static void gen_cmpi_jumpnv(DisasContext *ctx,
955
{
956
TCGv pred = tcg_temp_new();
957
tcg_gen_setcondi_tl(cond, pred, val, src);
966
- gen_cond_jump(ctx, TCG_COND_EQ, pred, pc_off);
958
+ gen_cond_jump(ctx, TCG_COND_TSTEQ, pred, pc_off);
959
}
960
961
/* Shift left with saturation */
target/hexagon/translate.c
+1
-1
@@ -163,7 +163,7 @@ static void gen_end_tb(DisasContext *ctx)
163
if (ctx->branch_cond != TCG_COND_NEVER) {
164
if (ctx->branch_cond != TCG_COND_ALWAYS) {
165
TCGLabel *skip = gen_new_label();
166
- tcg_gen_brcondi_tl(ctx->branch_cond, ctx->branch_taken, 0, skip);
166
+ tcg_gen_brcondi_tl(ctx->branch_cond, ctx->branch_taken, 1, skip);
167
gen_goto_tb(ctx, 0, ctx->branch_dest, true);
168
gen_set_label(skip);
169
gen_goto_tb(ctx, 1, ctx->next_PC, false);