Microprocessor that builds inconsistent loop that iteration count unrolled loop multi-fetch block macro-op cache entries
A microprocessor includes a prediction unit (PRU) that predicts a sequence of fetch blocks (FBlks) in a program instruction stream, a macro-op (MOP) cache (MOC) that comprises MOC entries (MEs), a fusion engine. An ME holds MOPs into which architectural instructions of one or more FBlks are decoded. The PRU detects a loop body ME within the program instruction stream, accumulates loop iteration count information about a series of instances of a loop on the loop body ME in the program instruction stream, updates a consistency counter of the loop body ME while accumulating the loop iteration count information, and in response to detecting that the consistency counter has reached a threshold, instructs the fusion engine to use F copies of the MOPs of the loop body ME to build in the MOC an unrolled loop multi-FBlk ME; F is a loop unroll factor that is at least two.
1 . A microprocessor, comprising:
a prediction unit (PRU) that predicts a sequence of fetch blocks (FBlks) in a program instruction stream, wherein a FBlk comprises a sequential run of architectural instructions;
a macro-op (MOP) cache (MOC) that comprises MOC entries (MEs), wherein a ME holds MOPs into which architectural instructions of one or more FBlks are decoded; and
a fusion engine;
wherein the PRU is configured to:
detect a loop body ME within the program instruction stream;
accumulate loop iteration count information about a series of instances of a loop on the loop body ME in the program instruction stream;
update a consistency counter of the loop body ME while accumulating the loop iteration count information;
in response to detecting that the consistency counter has reached a threshold, instruct the fusion engine to use F copies of the MOPs of the loop body ME to build in the MOC an unrolled loop multi-FBlk ME (MF-ME) (ULP-MF-ME), wherein F is a loop unroll factor that is at least two.
2 . The microprocessor of claim 1 ,
wherein the loop iteration count information comprises a minimum loop iteration count observed among the series of instances.
3 . The microprocessor of claim 2 ,
wherein the PRU is configured to select F based on the minimum loop iteration count.
4 . The microprocessor of claim 1 ,
wherein the loop body ME comprises a minimum loop iteration count that is initialized to a maximum value; and
wherein for each current instance of the series of instances of the loop on the loop body ME, wherein to update the consistency counter, the PRU:
when the loop iteration count of the current instance of the loop on the loop body ME is less than the minimum loop iteration count of the loop body ME:
resets the consistency counter; and
updates the minimum loop iteration count to the loop iteration count of the current instance; and
otherwise:
increments the consistency counter.
5 . The microprocessor of claim 1 ,
wherein the loop iteration count information comprises a histogram of frequencies over a range of loop iteration counts observed in the series of instances.
6 . The microprocessor of claim 5 ,
wherein for each instance of the series of instances of the loop on the loop body ME, to update the consistency counter the PRU increments the consistency counter.
7 . The microprocessor of claim 5 ,
wherein the PRU is configured to select F based on the histogram.
8 . The microprocessor of claim 5 ,
wherein the PRU is configured to select a loop iteration count from the range of loop iteration counts of the histogram.
9 . The microprocessor of claim 8 ,
wherein to select a loop iteration count from the range of loop iteration counts of the histogram, the PRU:
selects as the loop iteration count a smallest of the loop iteration counts of the range having a frequency greater than a predetermined value.
10 . The microprocessor of claim 8 ,
wherein to selecting a loop iteration count from the range of loop iteration counts of the histogram, the PRU:
selects as the loop iteration count a loop iteration count from the range of loop iteration counts of the histogram based on variability of the frequencies of the histogram.
11 . The microprocessor of claim 5 ,
wherein the loop body ME comprises an unrolled loop iteration count;
wherein the PRU is configured to:
populate the unrolled loop iteration count based on F and the histogram when the ULP-MF-ME is built; and
in response to a hit in the MOC on the ULP-MF-ME, fetch from the MOC a number of copies of the MOPs of the ULP-MF-ME equal to the unrolled loop iteration count.
12 . The microprocessor of claim 11 ,
wherein the PRU is configured to, after fetching from the MOC a number of copies of the MOPs of the ULP-MF-ME equal to the unrolled loop iteration count:
in response to a multiple-hit in the MOC on the ULP-MF-ME and on the loop body ME immediately subsequent to the hit in the MOC on the ULP-MF-ME, fetch the MOPs of the loop body ME from the MOC until falling out of the loop.
13 . The microprocessor of claim 1 ,
wherein the PRU is configured to:
select a loop iteration count using the loop iteration count information; and
populate the unrolled loop iteration count with a floor function of a quotient of the selected loop iteration count and F.
14 . The microprocessor of claim 1 ,
wherein the PRU is configured to:
select a loop iteration count using the loop iteration count information; and
populate the unrolled loop iteration count with a ceiling function of a quotient of the selected loop iteration count and F.
15 . The microprocessor of claim 1 ,
wherein the PRU is configured to:
select one loop iteration count from the range of loop iteration counts of the histogram; and
populate the unrolled loop iteration count with a ceiling function of a quotient of the selected one loop iteration count and F.
16 . The microprocessor of claim 1 ,
wherein the PRU is configured to, prior to said detecting the loop body ME, instruct the fusion engine to build in the MOC a sequential multi-FBlk ME (MF-ME) (SEQ-MF-ME) using the MOPs of a sequence of MEs in the program instruction stream;
wherein the loop body ME is the SEQ-MF-ME.
17 . The microprocessor of claim 16 ,
wherein the PRU is configured to retain the loop body ME in the MOC co-resident with the ULP-MF-ME when allocating the ULP-MF-ME in the MOC.
18 . The microprocessor of claim 17 ,
wherein the PRU is configured to set an indicator in the ULP-MF-ME to distinguish the ULP-MF-ME from the loop body ME when allocating the ULP-MF-ME in the MOC.
19 . The microprocessor of claim 1 ,
wherein the threshold is software configurable.
20 . The microprocessor of claim 1 ,
wherein the threshold is dynamically varied based on recent characteristics of the program instruction stream.
21 . The microprocessor of claim 1 ,
wherein the F copies of the MOPs of the loop body ME have a total number of MOPs;
wherein the ULP-MF-ME has a second number of MOPs; and
wherein to use the F copies of the MOPs of the loop body ME to build in the MOC the ULP-MF-ME, the fusion engine:
fuses the F copies of the MOPs of the loop body ME into the MOPs of the ULP-MF-ME such that the second number is fewer than the total number.
22 . A method, comprising:
in a microprocessor comprising:
a prediction unit (PRU) that predicts a sequence of fetch blocks (FBlks) in a program instruction stream, wherein a FBlk comprises a sequential run of architectural instructions; and
a macro-op (MOP) cache (MOC) that comprises MOC entries (MEs), wherein a ME holds MOPs into which architectural instructions of one or more FBlks are decoded;
detecting a loop body ME within the program instruction stream;
accumulating loop iteration count information about a series of instances of a loop on the loop body ME in the program instruction stream;
updating a consistency counter of the loop body ME while said accumulating the loop iteration count information;
in response to detecting that the consistency counter has reached a threshold, using F copies of the MOPs of the loop body ME to build in the MOC an unrolled loop multi-FBlk ME (MF-ME) (ULP-MF-ME), wherein F is a loop unroll factor that is at least two.
23 . The method of claim 22 ,
wherein the loop iteration count information comprises a minimum loop iteration count observed among the series of instances.
24 . The method of claim 23 , further comprising:
selecting F based on the minimum loop iteration count.
25 . The method of claim 22 ,
wherein the loop body ME comprises a minimum loop iteration count that is initialized to a maximum value; and
wherein for each current instance of the series of instances of the loop on the loop body ME, said updating the consistency counter comprises:
when the loop iteration count of the current instance of the loop on the loop body ME is less than the minimum loop iteration count of the loop body ME:
resetting the consistency counter; and
updating the minimum loop iteration count to the loop iteration count of the current instance; and
otherwise:
incrementing the consistency counter.
26 . The method of claim 22 ,
wherein the loop iteration count information comprises a histogram of frequencies over a range of loop iteration counts observed in the series of instances.
27 . The method of claim 26 ,
wherein for each instance of the series of instances of the loop on the loop body ME, said updating the consistency counter comprises incrementing the consistency counter.
28 . The method of claim 26 , further comprising:
selecting F based on the histogram.
29 . The method of claim 26 , further comprising:
selecting a loop iteration count from the range of loop iteration counts of the histogram.
30 . The method of claim 29 ,
wherein said selecting a loop iteration count from the range of loop iteration counts of the histogram comprises:
selecting as the loop iteration count a smallest of the loop iteration counts of the range having a frequency greater than a predetermined value.
31 . The method of claim 29 ,
wherein said selecting a loop iteration count from the range of loop iteration counts of the histogram comprises:
selecting as the loop iteration count a loop iteration count from the range of loop iteration counts of the histogram based on variability of the frequencies of the histogram.
32 . The method of claim 26 , further comprising:
wherein the loop body ME comprises an unrolled loop iteration count;
populating the unrolled loop iteration count based on F and the histogram when the ULP-MF-ME is built; and
in response to a hit in the MOC on the ULP-MF-ME, fetching from the MOC a number of copies of the MOPs of the ULP-MF-ME equal to the unrolled loop iteration count.
33 . The method of claim 32 , further comprising:
after said fetching from the MOC a number of copies of the MOPs of the ULP-MF-ME equal to the unrolled loop iteration count:
in response to a multiple-hit in the MOC on the ULP-MF-ME and on the loop body ME immediately subsequent to the hit in the MOC on the ULP-MF-ME, fetching the MOPs of the loop body ME from the MOC until falling out of the loop.
34 . The method of claim 22 , further comprising:
selecting a loop iteration count using the loop iteration count information; and
populating the unrolled loop iteration count with a floor function of a quotient of the selected loop iteration count and F.
35 . The method of claim 22 , further comprising:
selecting a loop iteration count using the loop iteration count information; and
populating the unrolled loop iteration count with a ceiling function of a quotient of the selected loop iteration count and F.
36 . The method of claim 22 , further comprising:
selecting one loop iteration count from the range of loop iteration counts of the histogram; and
populating the unrolled loop iteration count with a ceiling function of a quotient of the selected one loop iteration count and F.
37 . The method of claim 22 , further comprising:
prior to said detecting the loop body ME, building in the MOC a sequential multi-FBlk ME (MF-ME) (SEQ-MF-ME) using the MOPs of a sequence of MEs in the program instruction stream;
wherein the loop body ME is the SEQ-MF-ME.
38 . The method of claim 37 , further comprising:
retaining the loop body ME in the MOC co-resident with the ULP-MF-ME when allocating the ULP-MF-ME in the MOC.
39 . The method of claim 38 , further comprising:
setting an indicator in the ULP-MF-ME to distinguish the ULP-MF-ME from the loop body ME when allocating the ULP-MF-ME in the MOC.
40 . The method of claim 22 ,
wherein the threshold is software configurable.
41 . The method of claim 22 , further comprising:
dynamically varying the threshold based on recent characteristics of the program instruction stream.
42 . The method of claim 22 ,
wherein the F copies of the MOPs of the loop body ME have a total number of MOPs;
wherein the ULP-MF-ME has a second number of MOPs; and
wherein said using the F copies of the MOPs of the loop body ME to build in the MOC the ULP-MF-ME comprises:
fusing the F copies of the MOPs of the loop body ME into the MOPs of the ULP-MF-ME such that the second number is fewer than the total number.
43 . A non-transitory computer-readable medium having instructions stored thereon that are capable of causing or configuring a microprocessor comprising:
a prediction unit (PRU) that predicts a sequence of fetch blocks (FBlks) in a program instruction stream, wherein a FBlk comprises a sequential run of architectural instructions;
a macro-op (MOP) cache (MOC) that comprises MOC entries (MEs), wherein a ME holds MOPs into which architectural instructions of one or more FBlks are decoded; and
a fusion engine;
wherein the PRU is configured to:
detect a loop body ME within the program instruction stream;
accumulate loop iteration count information about a series of instances of a loop on the loop body ME in the program instruction stream;
update a consistency counter of the loop body ME while accumulating the loop iteration count information;
in response to detecting that the consistency counter has reached a threshold, instruct the fusion engine to use F copies of the MOPs of the loop body ME to build in the MOC an unrolled loop multi-FBlk ME (MF-ME) (ULP-MF-ME), wherein F is a loop unroll factor that is at least two.