<font size=2 face="sans-serif">Hi, </font><br><br><font size=2 face="sans-serif">I've been looking at the Machine Scheduler
on Power PC. I am looking only at the pre-RA machine scheduler and
I am running it in the default bi-directional mode (so, both top down and
bottom up queues are considered). I've come across an example where the
scheduler picks a poor ordering for the instructions which results in very
high register pressure which results in spills. The problem comes from
the fact that the Machine Scheduler uses a maximum latency limit when it
considers instructions to schedule. A high latency instruction will not
be scheduled before all of the available lower latency instructions are
scheduled. This happens regardless of register pressure since the higher
latency instruction is not even added to the "Available" queue
that is used when the heuristics pick an instruction to schedule next.
</font><br><br><font size=2 face="sans-serif">My question is: Why do we have that
latency limit in the first place? If an instruction can be scheduled (ie
all the instructions it depends on are already scheduled) shouldn't it
be at least considered?</font><br><br><br><br><font size=2 face="sans-serif">The example is listed below:</font><br><br><font size=2 face="sans-serif">test.c</font><br><font size=2 face="sans-serif">--</font><br><font size=2 face="sans-serif">long A[100];</font><br><font size=2 face="sans-serif">long func(long* num, long* den) {</font><br><font size=2 face="sans-serif">// This loop is unrolled</font><br><font size=2 face="sans-serif">for (int i=0; i<6; i++) {</font><br><font size=2 face="sans-serif"> A[i] = num[i] / den[i];</font><br><font size=2 face="sans-serif">}</font><br><font size=2 face="sans-serif">return 0;</font><br><font size=2 face="sans-serif">}</font><br><font size=2 face="sans-serif">--</font><br><br><font size=2 face="sans-serif">Compile commands</font><br><font size=2 face="sans-serif">--</font><br><font size=2 face="sans-serif">clang -c -m64 -O3 -target powerpc64le-unknown-linux-gnu
-mcpu=pwr9 -fexperimental-new-pass-manager test.c -S -emit-llvm</font><br><font size=2 face="sans-serif">llc test.ll -O3 -ppc-asm-full-reg-names
-debug-only=machine-scheduler -o test-p9.s > listing.out 2>&1</font><br><font size=2 face="sans-serif">--</font><br><br><font size=2 face="sans-serif">Looking at the listing.out file I've
noticed that all of the loads are grouped together at the start of the
function. Those loads use 12 registers before any of the divides are scheduled.
As a result, we end up with significantly higher register pressure after
all the loads.</font><br><font size=2 face="sans-serif">--</font><br><font size=2 face="sans-serif">0B BB#0: derived
from LLVM BB %entry</font><br><font size=2 face="sans-serif">
Live Ins: %X3 %X4</font><br><font size=2 face="sans-serif">16B
%vreg1<def> = COPY %X4; G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">32B
%vreg0<def> = COPY %X3; G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">48B
%vreg2<def> = LD 0, %vreg0; mem:LD8[%num](tbaa=!4) G8RC:%vreg2
G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">64B
%vreg3<def> = LD 0, %vreg1; mem:LD8[%den](tbaa=!4) G8RC:%vreg3
G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">144B
%vreg7<def> = LD 8, %vreg0; mem:LD8[%arrayidx.1](tbaa=!4) G8RC:%vreg7
G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">160B
%vreg8<def> = LD 8, %vreg1; mem:LD8[%arrayidx2.1](tbaa=!4)
G8RC:%vreg8 G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">208B
%vreg10<def> = LD 16, %vreg0; mem:LD8[%arrayidx.2](tbaa=!4)
G8RC:%vreg10 G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">224B
%vreg11<def> = LD 16, %vreg1; mem:LD8[%arrayidx2.2](tbaa=!4)
G8RC:%vreg11 G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">272B
%vreg13<def> = LD 24, %vreg0; mem:LD8[%arrayidx.3](tbaa=!4)
G8RC:%vreg13 G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">288B
%vreg14<def> = LD 24, %vreg1; mem:LD8[%arrayidx2.3](tbaa=!4)
G8RC:%vreg14 G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">336B
%vreg16<def> = LD 32, %vreg0; mem:LD8[%arrayidx.4](tbaa=!4)
G8RC:%vreg16 G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">352B
%vreg17<def> = LD 32, %vreg1; mem:LD8[%arrayidx2.4](tbaa=!4)
G8RC:%vreg17 G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">400B
%vreg19<def> = LD 40, %vreg0; mem:LD8[%arrayidx.5](tbaa=!4)
G8RC:%vreg19 G8RC_and_G8RC_NOX0:%vreg0</font><br><font size=2 face="sans-serif">416B
%vreg20<def> = LD 40, %vreg1; mem:LD8[%arrayidx2.5](tbaa=!4)
G8RC:%vreg20 G8RC_and_G8RC_NOX0:%vreg1</font><br><font size=2 face="sans-serif">424B
%vreg4<def> = DIVD %vreg2, %vreg3; G8RC:%vreg4,%vreg2,%vreg3</font><br><font size=2 face="sans-serif">432B
%vreg9<def> = DIVD %vreg7, %vreg8; G8RC:%vreg9,%vreg7,%vreg8</font><br><font size=2 face="sans-serif">440B
%vreg12<def> = DIVD %vreg10, %vreg11; G8RC:%vreg12,%vreg10,%vreg11</font><br><font size=2 face="sans-serif">448B
%vreg15<def> = DIVD %vreg13, %vreg14; G8RC:%vreg15,%vreg13,%vreg14</font><br><font size=2 face="sans-serif">456B
%vreg18<def> = DIVD %vreg16, %vreg17; G8RC:%vreg18,%vreg16,%vreg17</font><br><font size=2 face="sans-serif">464B
%vreg21<def> = DIVD %vreg19, %vreg20; G8RC:%vreg21,%vreg19,%vreg20</font><br><font size=2 face="sans-serif">472B
%vreg5<def> = ADDIStocHA %X2, <ga:@A>; G8RC_and_G8RC_NOX0:%vreg5</font><br><font size=2 face="sans-serif">480B
%vreg6<def> = LDtocL <ga:@A>, %vreg5, %X2<imp-use>;
mem:LD8[GOT] G8RC_and_G8RC_NOX0:%vreg6,%vreg5</font><br><font size=2 face="sans-serif">504B
%X3<def> = LI8 0</font><br><font size=2 face="sans-serif">512B
STD %vreg4, 0, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 0)](tbaa=!4) G8RC:%vreg4 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">520B
STD %vreg9, 8, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 1)](tbaa=!4) G8RC:%vreg9 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">528B
STD %vreg12, 16, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 2)](tbaa=!4) G8RC:%vreg12 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">536B
STD %vreg15, 24, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 3)](tbaa=!4) G8RC:%vreg15 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">544B
STD %vreg18, 32, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 4)](tbaa=!4) G8RC:%vreg18 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">552B
STD %vreg21, 40, %vreg6; mem:ST8[getelementptr inbounds ([100 x i64],
[100 x i64]* @A, i64 0, i64 5)](tbaa=!4) G8RC:%vreg21 G8RC_and_G8RC_NOX0:%vreg6</font><br><font size=2 face="sans-serif">560B
BLR8 %LR8<imp-use>, %RM<imp-use>, %X3<imp-use></font><br><font size=2 face="sans-serif">--</font><br><br><font size=2 face="sans-serif">Due to all of the register pressure
built up by those loads we are forced to spill. Here is the final assembly.</font><br><font size=2 face="sans-serif">--</font><br><font size=2 face="sans-serif"># BB#0:
# %entry</font><br><font size=2 face="sans-serif"> std r30,
-16(r1) # 8-byte
Folded Spill</font><br><font size=2 face="sans-serif"> ld r5, 0(r3)</font><br><font size=2 face="sans-serif"> ld r6, 0(r4)</font><br><font size=2 face="sans-serif"> ld r7, 8(r3)</font><br><font size=2 face="sans-serif"> ld r8, 8(r4)</font><br><font size=2 face="sans-serif"> ld r9, 16(r3)</font><br><font size=2 face="sans-serif"> ld r10,
16(r4)</font><br><font size=2 face="sans-serif"> ld r11,
24(r3)</font><br><font size=2 face="sans-serif"> ld r0, 32(r3)</font><br><font size=2 face="sans-serif"> ld r12,
24(r4)</font><br><font size=2 face="sans-serif"> ld r30,
32(r4)</font><br><font size=2 face="sans-serif"> ld r3, 40(r3)</font><br><font size=2 face="sans-serif"> ld r4, 40(r4)</font><br><font size=2 face="sans-serif"> divd r5,
r5, r6</font><br><font size=2 face="sans-serif"> divd r6,
r7, r8</font><br><font size=2 face="sans-serif"> divd r7,
r9, r10</font><br><font size=2 face="sans-serif"> divd r9,
r0, r30</font><br><font size=2 face="sans-serif"> divd r4,
r3, r4</font><br><font size=2 face="sans-serif"> divd r8,
r11, r12</font><br><font size=2 face="sans-serif"> addis r3,
r2, .LC0@toc@ha</font><br><font size=2 face="sans-serif"> ld r30,
-16(r1) # 8-byte
Folded Reload</font><br><font size=2 face="sans-serif"> ld r10,
.LC0@toc@l(r3)</font><br><font size=2 face="sans-serif"> li r3, 0</font><br><font size=2 face="sans-serif"> std r5,
0(r10)</font><br><font size=2 face="sans-serif"> std r6,
8(r10)</font><br><font size=2 face="sans-serif"> std r7,
16(r10)</font><br><font size=2 face="sans-serif"> std r9,
32(r10)</font><br><font size=2 face="sans-serif"> std r8,
24(r10)</font><br><font size=2 face="sans-serif"> std r4,
40(r10)</font><br><font size=2 face="sans-serif"> blr</font><br><font size=2 face="sans-serif">--</font><br><br><font size=2 face="sans-serif">Thank you, </font><br><font size=2 face="sans-serif">Stefan Pintilie</font><BR>