KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

broadcom/compiler: Merge instructions more efficiently 

Instructions are allowed to access up to two rf registers, or one rf
register and a small immediate. This change allows qpu_merge_inst to
take full advantage of this by allowint the merging of two instructions
if they have no more than two different rf registers between them,
or one rf register and one small immediate. qpu_merge_inst rewrites
the instructions as needed to pack everything into raddr_a and raddr_b
in the merged instruction.

shader-db stats:
total instructions in shared programs: 19938769 -> 18929664 (-5.06%)
instructions in affected programs: 17929438 -> 16920333 (-5.63%)
helped: 95008
HURT: 242
helped stats (abs) min: 1 max: 785 x̄: 10.62 x̃: 7
helped stats (rel) min: 0.30% max: 21.25% x̄: 5.37% x̃: 4.98%
HURT stats (abs)   min: 1 max: 2 x̄: 1.10 x̃: 1
HURT stats (rel)   min: 0.30% max: 3.12% x̄: 1.62% x̃: 1.54%
95% mean confidence interval for instructions value: -10.67 -10.52
95% mean confidence interval for instructions %-change: -5.37% -5.33%
Instructions are helped.

total max-temps in shared programs: 3122664 -> 3112446 (-0.33%)
max-temps in affected programs: 124881 -> 114663 (-8.18%)
helped: 5445
HURT: 0
helped stats (abs) min: 1 max: 15 x̄: 1.88 x̃: 1
helped stats (rel) min: 1.49% max: 40.54% x̄: 8.97% x̃: 6.67%
95% mean confidence interval for max-temps value: -1.91 -1.84
95% mean confidence interval for max-temps %-change: -9.12% -8.81%
Max-temps are helped.

total sfu-stalls in shared programs: 38028 -> 41231 (8.42%)
sfu-stalls in affected programs: 6053 -> 9256 (52.92%)
helped: 664
HURT: 3380
helped stats (abs) min: 1 max: 2 x̄: 1.04 x̃: 1
helped stats (rel) min: 9.09% max: 100.00% x̄: 70.81% x̃: 100.00%
HURT stats (abs)   min: 1 max: 4 x̄: 1.15 x̃: 1
HURT stats (rel)   min: 0.00% max: 300.00% x̄: 46.39% x̃: 25.00%
95% mean confidence interval for sfu-stalls value: 0.76 0.82
95% mean confidence interval for sfu-stalls %-change: 25.03% 29.26%
Sfu-stalls are HURT.

total inst-and-stalls in shared programs: 19976797 -> 18970895 (-5.04%)
inst-and-stalls in affected programs: 17963129 -> 16957227 (-5.60%)
helped: 95017
HURT: 245
helped stats (abs) min: 1 max: 785 x̄: 10.59 x̃: 7
helped stats (rel) min: 0.30% max: 21.25% x̄: 5.35% x̃: 4.95%
HURT stats (abs)   min: 1 max: 2 x̄: 1.09 x̃: 1
HURT stats (rel)   min: 0.30% max: 3.12% x̄: 1.61% x̃: 1.54%
95% mean confidence interval for inst-and-stalls value: -10.64 -10.48
95% mean confidence interval for inst-and-stalls %-change: -5.35% -5.31%
Inst-and-stalls are helped.

v2 (Iago):
 - moved early return for naddrs > 2 even earlier.
 - only update {add,mul}.b mux if instruction has more than one operand.
 - don't OR b->raddr_{a,b} if we are not merging add/mul instructions.
 - don't initialize packed to 0.
 - minor style fixes.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9026>
This commit is contained in:
Arcady Goldmints-Orlov 2021-01-31 14:53:55 -05:00 committed by Marge Bot
parent 6ccbb89207
commit 7f61ff7b4d
1 changed files with 115 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

129
src/broadcom/compiler/qpu_schedule.c
View File

@ -687,6 +687,112 @@ qpu_compatible_peripheral_access(const struct v3d_device_info *devinfo,
return false;
}
/* Compute a bitmask of which rf registers are used between
* the two instructions.
*/
static uint64_t
qpu_raddrs_used(const struct v3d_qpu_instr *a,
const struct v3d_qpu_instr *b)
{
assert(a->type == V3D_QPU_INSTR_TYPE_ALU);
assert(b->type == V3D_QPU_INSTR_TYPE_ALU);
uint64_t raddrs_used = 0;
if (v3d_qpu_uses_mux(a, V3D_QPU_MUX_A))
raddrs_used |= (1ll << a->raddr_a);
if (!a->sig.small_imm && v3d_qpu_uses_mux(a, V3D_QPU_MUX_B))
raddrs_used |= (1ll << a->raddr_b);
if (v3d_qpu_uses_mux(b, V3D_QPU_MUX_A))
raddrs_used |= (1ll << b->raddr_a);
if (!b->sig.small_imm && v3d_qpu_uses_mux(b, V3D_QPU_MUX_B))
raddrs_used |= (1ll << b->raddr_b);
return raddrs_used;
}
/* Take two instructions and attempt to merge their raddr fields
* into one merged instruction. Returns false if the two instructions
* access more than two different rf registers between them, or more
* than one rf register and one small immediate.
*/
static bool
qpu_merge_raddrs(struct v3d_qpu_instr *result,
const struct v3d_qpu_instr *add_instr,
const struct v3d_qpu_instr *mul_instr)
{
uint64_t raddrs_used = qpu_raddrs_used(add_instr, mul_instr);
int naddrs = util_bitcount64(raddrs_used);
if (naddrs > 2)
return false;
if ((add_instr->sig.small_imm || mul_instr->sig.small_imm)) {
if (naddrs > 1)
return false;
if (add_instr->sig.small_imm && mul_instr->sig.small_imm)
if (add_instr->raddr_b != mul_instr->raddr_b)
return false;
result->sig.small_imm = true;
result->raddr_b = add_instr->sig.small_imm ?
add_instr->raddr_b : mul_instr->raddr_b;
}
if (naddrs == 0)
return true;
int raddr_a = ffsll(raddrs_used) - 1;
raddrs_used &= ~(1ll << raddr_a);
result->raddr_a = raddr_a;
if (!result->sig.small_imm) {
if (v3d_qpu_uses_mux(add_instr, V3D_QPU_MUX_B) &&
raddr_a == add_instr->raddr_b) {
if (add_instr->alu.add.a == V3D_QPU_MUX_B)
result->alu.add.a = V3D_QPU_MUX_A;
if (add_instr->alu.add.b == V3D_QPU_MUX_B &&
v3d_qpu_add_op_num_src(add_instr->alu.add.op) > 1) {
result->alu.add.b = V3D_QPU_MUX_A;
}
}
if (v3d_qpu_uses_mux(mul_instr, V3D_QPU_MUX_B) &&
raddr_a == mul_instr->raddr_b) {
if (mul_instr->alu.mul.a == V3D_QPU_MUX_B)
result->alu.mul.a = V3D_QPU_MUX_A;
if (mul_instr->alu.mul.b == V3D_QPU_MUX_B &&
v3d_qpu_mul_op_num_src(mul_instr->alu.mul.op) > 1) {
result->alu.mul.b = V3D_QPU_MUX_A;
}
}
}
if (!raddrs_used)
return true;
int raddr_b = ffsll(raddrs_used) - 1;
result->raddr_b = raddr_b;
if (v3d_qpu_uses_mux(add_instr, V3D_QPU_MUX_A) &&
raddr_b == add_instr->raddr_a) {
if (add_instr->alu.add.a == V3D_QPU_MUX_A)
result->alu.add.a = V3D_QPU_MUX_B;
if (add_instr->alu.add.b == V3D_QPU_MUX_A &&
v3d_qpu_add_op_num_src(add_instr->alu.add.op) > 1) {
result->alu.add.b = V3D_QPU_MUX_B;
}
}
if (v3d_qpu_uses_mux(mul_instr, V3D_QPU_MUX_A) &&
raddr_b == mul_instr->raddr_a) {
if (mul_instr->alu.mul.a == V3D_QPU_MUX_A)
result->alu.mul.a = V3D_QPU_MUX_B;
if (mul_instr->alu.mul.b == V3D_QPU_MUX_A &&
v3d_qpu_add_op_num_src(mul_instr->alu.mul.op) > 1) {
result->alu.mul.b = V3D_QPU_MUX_B;
}
}
return true;
}
static bool
qpu_merge_inst(const struct v3d_device_info *devinfo,
struct v3d_qpu_instr *result,
@ -702,6 +808,7 @@ qpu_merge_inst(const struct v3d_device_info *devinfo,
return false;
struct v3d_qpu_instr merge = *a;
const struct v3d_qpu_instr *add_instr = NULL, *mul_instr = NULL;
if (b->alu.add.op != V3D_QPU_A_NOP) {
if (a->alu.add.op != V3D_QPU_A_NOP)
@ -711,6 +818,9 @@ qpu_merge_inst(const struct v3d_device_info *devinfo,
merge.flags.ac = b->flags.ac;
merge.flags.apf = b->flags.apf;
merge.flags.auf = b->flags.auf;
add_instr = b;
mul_instr = a;
}
if (b->alu.mul.op != V3D_QPU_M_NOP) {
@ -721,23 +831,14 @@ qpu_merge_inst(const struct v3d_device_info *devinfo,
merge.flags.mc = b->flags.mc;
merge.flags.mpf = b->flags.mpf;
merge.flags.muf = b->flags.muf;
mul_instr = b;
add_instr = a;
}
if (v3d_qpu_uses_mux(b, V3D_QPU_MUX_A)) {
if (v3d_qpu_uses_mux(a, V3D_QPU_MUX_A) &&
a->raddr_a != b->raddr_a) {
if (add_instr && mul_instr &&
!qpu_merge_raddrs(&merge, add_instr, mul_instr)) {
return false;
}
merge.raddr_a = b->raddr_a;
}
if (v3d_qpu_uses_mux(b, V3D_QPU_MUX_B)) {
if (v3d_qpu_uses_mux(a, V3D_QPU_MUX_B) &&
(a->raddr_b != b->raddr_b ||
a->sig.small_imm != b->sig.small_imm)) {
return false;
}
merge.raddr_b = b->raddr_b;
}
merge.sig.thrsw |= b->sig.thrsw;