target/arm: Drop oddstatus from is_ebf and bfdotadd_ebf
This argument is no longer used. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-id: 20260609192110.752384-3-richard.henderson@linaro.org Message-Id: <20260517002550.321291-10-richard.henderson@linaro.org> Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Richard Henderson committed
Jun 9, 2026 at 12:20 UTC
f1fb6fac581d1d079ad96ea38ab714024d98fdf4
3 files changed
+29
-39
target/arm/tcg/sme_helper.c
+3
-3
@@ -1429,9 +1429,9 @@ static void do_bfmopa_w(void *vza, void *vzn, void *vzm,
1429
uint32_t desc, uint32_t negx, bool ah_neg)
1430
{
1431
intptr_t row, col, oprsz = simd_maxsz(desc);
1432
- float_status fpst, fpst_odd;
1432
+ float_status fpst;
1433
1434
- if (is_ebf(env, &fpst, &fpst_odd)) {
1434
+ if (is_ebf(env, &fpst)) {
1435
for (row = 0; row < oprsz; ) {
1436
uint16_t prow = pn[H2(row >> 4)];
1437
do {
@@ -1452,7 +1452,7 @@ static void do_bfmopa_w(void *vza, void *vzn, void *vzm,
1452
uint32_t m = *(uint32_t *)(vzm + H1_4(col));
1453
1454
m = f16mop_adj_pair(m, pcol, 0);
1455
- *a = bfdotadd_ebf(*a, n, m, &fpst, &fpst_odd);
1455
+ *a = bfdotadd_ebf(*a, n, m, &fpst);
1456
}
1457
col += 4;
1458
pcol >>= 4;
target/arm/tcg/vec_helper.c
+22
-28
@@ -2845,7 +2845,7 @@ DO_MMLA_B(gvec_usmmla_b, do_usmmla_b)
2845
* BFloat16 Dot Product
2846
*/
2847
2848
-bool is_ebf(CPUARMState *env, float_status *statusp, float_status *oddstatusp)
2848
+bool is_ebf(CPUARMState *env, float_status *statusp)
2849
{
2850
/*
2851
* For BFDOT, BFMMLA, etc, the behaviour depends on FPCR.EBF.
@@ -2865,11 +2865,7 @@ bool is_ebf(CPUARMState *env, float_status *statusp, float_status *oddstatusp)
2865
*statusp = env->vfp.fp_status[is_a64(env) ? FPST_A64 : FPST_A32];
2866
set_default_nan_mode(true, statusp);
2867
2868
- if (ebf) {
2869
- /* EBF=1 needs to do a step with round-to-odd semantics */
2870
- *oddstatusp = *statusp;
2871
- set_float_rounding_mode(float_round_to_odd, oddstatusp);
2872
- } else {
2868
+ if (!ebf) {
2869
set_flush_to_zero(true, statusp);
2870
set_flush_inputs_to_zero(true, statusp);
2871
set_float_rounding_mode(float_round_to_odd_inf, statusp);
@@ -2893,8 +2889,7 @@ float32 bfdotadd(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst)
2889
return t1;
2890
}
2891
2896
-float32 bfdotadd_ebf(float32 sum, uint32_t e1, uint32_t e2,
2897
- float_status *fpst, float_status *fpst_odd)
2892
+float32 bfdotadd_ebf(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst)
2893
{
2894
/* Unpack two BFloat16 into two Float32, trivially. */
2895
float32 s1r = e1 << 16;
@@ -2964,11 +2959,11 @@ void HELPER(gvec_bfdot)(void *vd, void *vn, void *vm, void *va,
2959
intptr_t i, opr_sz = simd_oprsz(desc);
2960
float32 *d = vd, *a = va;
2961
uint32_t *n = vn, *m = vm;
2967
- float_status fpst, fpst_odd;
2962
+ float_status fpst;
2963
2969
- if (is_ebf(env, &fpst, &fpst_odd)) {
2964
+ if (is_ebf(env, &fpst)) {
2965
for (i = 0; i < opr_sz / 4; ++i) {
2971
- d[i] = bfdotadd_ebf(a[i], n[i], m[i], &fpst, &fpst_odd);
2966
+ d[i] = bfdotadd_ebf(a[i], n[i], m[i], &fpst);
2967
}
2968
} else {
2969
for (i = 0; i < opr_sz / 4; ++i) {
@@ -2987,14 +2982,14 @@ void HELPER(gvec_bfdot_idx)(void *vd, void *vn, void *vm,
2982
intptr_t eltspersegment = MIN(16 / 4, elements);
2983
float32 *d = vd, *a = va;
2984
uint32_t *n = vn, *m = vm;
2990
- float_status fpst, fpst_odd;
2985
+ float_status fpst;
2986
2992
- if (is_ebf(env, &fpst, &fpst_odd)) {
2987
+ if (is_ebf(env, &fpst)) {
2988
for (i = 0; i < elements; i += eltspersegment) {
2989
uint32_t m_idx = m[i + H4(index)];
2990
2991
for (j = i; j < i + eltspersegment; j++) {
2997
- d[j] = bfdotadd_ebf(a[j], n[j], m_idx, &fpst, &fpst_odd);
2992
+ d[j] = bfdotadd_ebf(a[j], n[j], m_idx, &fpst);
2993
}
2994
}
2995
} else {
@@ -3021,17 +3016,16 @@ void HELPER(sme2_bfvdot_idx)(void *vd, void *vn, void *vm,
3016
uint16_t *n0 = vn;
3017
uint16_t *n1 = vn + sizeof(ARMVectorReg);
3018
uint32_t *m = vm;
3024
- float_status fpst, fpst_odd;
3019
+ float_status fpst;
3020
3026
- if (is_ebf(env, &fpst, &fpst_odd)) {
3021
+ if (is_ebf(env, &fpst)) {
3022
for (i = 0; i < elements; i += eltspersegment) {
3023
uint32_t m_idx = m[i + H4(idx)];
3024
3025
for (j = 0; j < eltspersegment; j++) {
3026
uint32_t nn = (n0[H2(2 * (i + j) + sel)])
3027
| (n1[H2(2 * (i + j) + sel)] << 16);
3033
- d[i + H4(j)] = bfdotadd_ebf(a[i + H4(j)], nn, m_idx,
3034
- &fpst, &fpst_odd);
3028
+ d[i + H4(j)] = bfdotadd_ebf(a[i + H4(j)], nn, m_idx, &fpst);
3029
}
3030
}
3031
} else {
@@ -3054,9 +3048,9 @@ void HELPER(gvec_bfmmla)(void *vd, void *vn, void *vm, void *va,
3048
intptr_t s, opr_sz = simd_oprsz(desc);
3049
float32 *d = vd, *a = va;
3050
uint32_t *n = vn, *m = vm;
3057
- float_status fpst, fpst_odd;
3051
+ float_status fpst;
3052
3059
- if (is_ebf(env, &fpst, &fpst_odd)) {
3053
+ if (is_ebf(env, &fpst)) {
3054
for (s = 0; s < opr_sz / 4; s += 4) {
3055
float32 sum00, sum01, sum10, sum11;
3056
@@ -3068,20 +3062,20 @@ void HELPER(gvec_bfmmla)(void *vd, void *vn, void *vm, void *va,
3062
* i j i k j k
3063
*/
3064
sum00 = a[s + H4(0 + 0)];
3071
- sum00 = bfdotadd_ebf(sum00, n[s + H4(0 + 0)], m[s + H4(0 + 0)], &fpst, &fpst_odd);
3072
- sum00 = bfdotadd_ebf(sum00, n[s + H4(0 + 1)], m[s + H4(0 + 1)], &fpst, &fpst_odd);
3065
+ sum00 = bfdotadd_ebf(sum00, n[s + H4(0 + 0)], m[s + H4(0 + 0)], &fpst);
3066
+ sum00 = bfdotadd_ebf(sum00, n[s + H4(0 + 1)], m[s + H4(0 + 1)], &fpst);
3067
3068
sum01 = a[s + H4(0 + 1)];
3075
- sum01 = bfdotadd_ebf(sum01, n[s + H4(0 + 0)], m[s + H4(2 + 0)], &fpst, &fpst_odd);
3076
- sum01 = bfdotadd_ebf(sum01, n[s + H4(0 + 1)], m[s + H4(2 + 1)], &fpst, &fpst_odd);
3069
+ sum01 = bfdotadd_ebf(sum01, n[s + H4(0 + 0)], m[s + H4(2 + 0)], &fpst);
3070
+ sum01 = bfdotadd_ebf(sum01, n[s + H4(0 + 1)], m[s + H4(2 + 1)], &fpst);
3071
3072
sum10 = a[s + H4(2 + 0)];
3079
- sum10 = bfdotadd_ebf(sum10, n[s + H4(2 + 0)], m[s + H4(0 + 0)], &fpst, &fpst_odd);
3080
- sum10 = bfdotadd_ebf(sum10, n[s + H4(2 + 1)], m[s + H4(0 + 1)], &fpst, &fpst_odd);
3073
+ sum10 = bfdotadd_ebf(sum10, n[s + H4(2 + 0)], m[s + H4(0 + 0)], &fpst);
3074
+ sum10 = bfdotadd_ebf(sum10, n[s + H4(2 + 1)], m[s + H4(0 + 1)], &fpst);
3075
3076
sum11 = a[s + H4(2 + 1)];
3083
- sum11 = bfdotadd_ebf(sum11, n[s + H4(2 + 0)], m[s + H4(2 + 0)], &fpst, &fpst_odd);
3084
- sum11 = bfdotadd_ebf(sum11, n[s + H4(2 + 1)], m[s + H4(2 + 1)], &fpst, &fpst_odd);
3077
+ sum11 = bfdotadd_ebf(sum11, n[s + H4(2 + 0)], m[s + H4(2 + 0)], &fpst);
3078
+ sum11 = bfdotadd_ebf(sum11, n[s + H4(2 + 1)], m[s + H4(2 + 1)], &fpst);
3079
3080
d[s + H4(0 + 0)] = sum00;
3081
d[s + H4(0 + 1)] = sum01;
target/arm/tcg/vec_internal.h
+4
-8
@@ -271,7 +271,6 @@ float32 bfdotadd(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst);
271
* @sum: addend
272
* @e1, @e2: multiplicand vectors
273
* @fpst: floating-point status to use
274
- * @fpst_odd: floating-point status to use for round-to-odd operations
274
*
275
* BFloat16 2-way dot product of @e1 & @e2, accumulating with @sum.
276
* The @e1 and @e2 operands correspond to the 32-bit source vector
@@ -280,23 +279,20 @@ float32 bfdotadd(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst);
279
* Corresponds to the ARM pseudocode function BFDotAdd, specialized
280
* for the FPCR.EBF == 1 case.
281
*/
283
-float32 bfdotadd_ebf(float32 sum, uint32_t e1, uint32_t e2,
284
- float_status *fpst, float_status *fpst_odd);
282
+float32 bfdotadd_ebf(float32 sum, uint32_t e1, uint32_t e2, float_status *fpst);
283
284
/**
285
* is_ebf:
286
* @env: CPU state
287
* @statusp: pointer to floating point status to fill in
290
- * @oddstatusp: pointer to floating point status to fill in for round-to-odd
288
*
289
* Determine whether a BFDotAdd operation should use FPCR.EBF = 0
293
- * or FPCR.EBF = 1 semantics. On return, has initialized *statusp
294
- * and *oddstatusp to suitable float_status arguments to use with either
295
- * bfdotadd() or bfdotadd_ebf().
290
+ * or FPCR.EBF = 1 semantics. On return, has initialized *statusp as suitable
291
+ * for float_status arguments to either bfdotadd() or bfdotadd_ebf().
292
* Returns true for EBF = 1, false for EBF = 0. (The caller should use this
293
* to decide whether to call bfdotadd() or bfdotadd_ebf().)
294
*/
299
-bool is_ebf(CPUARMState *env, float_status *statusp, float_status *oddstatusp);
295
+bool is_ebf(CPUARMState *env, float_status *statusp);
296
297
/*
298
* Negate as for FPCR.AH=1 -- do not negate NaNs.