OV - exec - Loop 4669

	Loop Id: 4669	Module: exec	Source: par_csr_matop.c:865-989 [...]	Coverage: 0.02%

0x55aab0 MOV	%RCX,-0x30(%RBP)

0x55aab4 MOV	%RAX,%RBX

0x55aab7 CMP	-0x48(%RBP),%R8

0x55aabb JGE	55adf5

0x55aac1 MOV	%R8,%RDX

0x55aac4 CMPQ	$0,-0x98(%RBP)

0x55aacc JE	55aae6

0x55aace MOV	%RBX,(%RDI,%RDX,8)

0x55aad2 MOVQ	$0,(%R12,%RBX,8)

0x55aada MOV	-0x50(%RBP),%RAX

0x55aade MOV	%RDX,(%RAX,%RBX,8)

0x55aae2 LEA	0x1(%RBX),%RAX

0x55aae6 CMPQ	$0,-0x90(%RBP)

0x55aaee JE	55ac70

0x55aaf4 MOV	-0x58(%RBP),%RCX

0x55aaf8 MOV	(%RCX,%RDX,8),%R9

0x55aafc LEA	0x1(%RDX),%R8

0x55ab00 CMP	0x8(%RCX,%RDX,8),%R9

0x55ab05 JGE	55ac74

0x55ab0b MOV	%R8,-0x38(%RBP)

0x55ab0f MOV	-0x30(%RBP),%RCX

0x55ab13 MOV	%RDX,-0xb0(%RBP)

0x55ab1a JMP	55ab3d

  (4673) 0x55ab20 MOV	-0x40(%RBP),%R9

  (4673) 0x55ab24 INC	%R9

  (4673) 0x55ab27 MOV	-0x58(%RBP),%RDX

  (4673) 0x55ab2b MOV	-0xb0(%RBP),%RSI

  (4673) 0x55ab32 CMP	0x8(%RDX,%RSI,8),%R9

  (4673) 0x55ab37 JGE	55ac90

  (4673) 0x55ab3d MOV	-0xa0(%RBP),%RDX

  (4673) 0x55ab44 MOV	(%RDX,%R9,8),%R8

  (4673) 0x55ab48 MOV	-0xa8(%RBP),%RDX

  (4673) 0x55ab4f MOV	%R9,-0x40(%RBP)

  (4673) 0x55ab53 MOVSD	(%RDX,%R9,8),%XMM0

  (4673) 0x55ab59 MOV	-0x78(%RBP),%RDX

  (4673) 0x55ab5d MOV	(%RDX,%R8,8),%R9

  (4673) 0x55ab61 MOV	0x8(%RDX,%R8,8),%R10

  (4673) 0x55ab66 JMP	55ab84

    (4675) 0x55ab70 ADDSD	(%R13,%RDX,8),%XMM1

    (4675) 0x55ab77 MOVSD	%XMM1,(%R13,%RDX,8)

    (4675) 0x55ab7e MOV	%R14,%R15

    (4675) 0x55ab81 INC	%R9

    (4675) 0x55ab84 CMP	%R10,%R9

    (4675) 0x55ab87 JGE	55abe0

    (4675) 0x55ab89 MOV	-0x100(%RBP),%RDX

    (4675) 0x55ab90 MOV	(%RDX,%R9,8),%R11

    (4675) 0x55ab94 LEA	(%R11,%R15,1),%RSI

    (4675) 0x55ab98 MOV	(%RDI,%RSI,8),%RDX

    (4675) 0x55ab9c MOV	-0xf8(%RBP),%R15

    (4675) 0x55aba3 MOVSD	(%R15,%R9,8),%XMM1

    (4675) 0x55aba9 MULSD	%XMM0,%XMM1

    (4675) 0x55abad CMP	-0x30(%RBP),%RDX

    (4675) 0x55abb1 JGE	55ab70

    (4675) 0x55abb3 MOV	%RCX,(%RDI,%RSI,8)

    (4675) 0x55abb7 MOVSD	%XMM1,(%R13,%RCX,8)

    (4675) 0x55abbe MOV	-0x88(%RBP),%RDX

    (4675) 0x55abc5 MOV	%R11,(%RDX,%RCX,8)

    (4675) 0x55abc9 INC	%RCX

    (4675) 0x55abcc MOV	-0x78(%RBP),%RDX

    (4675) 0x55abd0 MOV	0x8(%RDX,%R8,8),%R10

    (4675) 0x55abd5 JMP	55ab7e

  (4673) 0x55abe0 MOV	-0x70(%RBP),%RDX

  (4673) 0x55abe4 MOV	(%RDX,%R8,8),%R9

  (4673) 0x55abe8 MOV	0x8(%RDX,%R8,8),%R10

  (4673) 0x55abed CMP	%R10,%R9

  (4673) 0x55abf0 JL	55ac2a

  (4673) 0x55abf2 JMP	55ab20

    (4674) 0x55ac00 MOV	%RAX,(%RDI,%R11,8)

    (4674) 0x55ac04 MOVSD	%XMM1,(%R12,%RAX,8)

    (4674) 0x55ac0a MOV	-0x50(%RBP),%RDX

    (4674) 0x55ac0e MOV	%R11,(%RDX,%RAX,8)

    (4674) 0x55ac12 INC	%RAX

    (4674) 0x55ac15 MOV	-0x70(%RBP),%RDX

    (4674) 0x55ac19 MOV	0x8(%RDX,%R8,8),%R10

    (4674) 0x55ac1e INC	%R9

    (4674) 0x55ac21 CMP	%R10,%R9

    (4674) 0x55ac24 JGE	55ab20

    (4674) 0x55ac2a MOV	-0xf0(%RBP),%RDX

    (4674) 0x55ac31 MOV	(%RDX,%R9,8),%R11

    (4674) 0x55ac35 MOV	(%RDI,%R11,8),%RDX

    (4674) 0x55ac39 MOV	-0xe8(%RBP),%RSI

    (4674) 0x55ac40 MOVSD	(%RSI,%R9,8),%XMM1

    (4674) 0x55ac46 MULSD	%XMM0,%XMM1

    (4674) 0x55ac4a CMP	%RBX,%RDX

    (4674) 0x55ac4d JL	55ac00

    (4674) 0x55ac4f ADDSD	(%R12,%RDX,8),%XMM1

    (4674) 0x55ac55 MOVSD	%XMM1,(%R12,%RDX,8)

    (4674) 0x55ac5b INC	%R9

    (4674) 0x55ac5e CMP	%R10,%R9

    (4674) 0x55ac61 JL	55ac2a

  (4673) 0x55ac63 JMP	55ab20

0x55ac70 LEA	0x1(%RDX),%R8

0x55ac74 MOV	-0x30(%RBP),%RCX

0x55ac78 MOV	-0x60(%RBP),%RSI

0x55ac7c MOV	(%RSI,%RDX,8),%RDX

0x55ac80 CMP	(%RSI,%R8,8),%RDX

0x55ac84 JGE	55aab0

0x55ac8a JMP	55aca9

0x55ac90 MOV	%RSI,%RDX

0x55ac93 MOV	-0x38(%RBP),%R8

0x55ac97 MOV	-0x60(%RBP),%RSI

0x55ac9b MOV	(%RSI,%RDX,8),%RDX

0x55ac9f CMP	(%RSI,%R8,8),%RDX

0x55aca3 JGE	55aab0

0x55aca9 MOV	%R8,-0x38(%RBP)

0x55acad JMP	55acc9

  (4670) 0x55acb0 MOV	-0x40(%RBP),%RDX

  (4670) 0x55acb4 INC	%RDX

  (4670) 0x55acb7 MOV	-0x60(%RBP),%RSI

  (4670) 0x55acbb MOV	-0x38(%RBP),%R8

  (4670) 0x55acbf CMP	(%RSI,%R8,8),%RDX

  (4670) 0x55acc3 JGE	55aab0

  (4670) 0x55acc9 MOV	-0xc8(%RBP),%RSI

  (4670) 0x55acd0 MOV	(%RSI,%RDX,8),%RSI

  (4670) 0x55acd4 MOV	-0xc0(%RBP),%R8

  (4670) 0x55acdb MOV	%RDX,-0x40(%RBP)

  (4670) 0x55acdf MOVSD	(%R8,%RDX,8),%XMM0

  (4670) 0x55ace5 MOV	-0x80(%RBP),%RDX

  (4670) 0x55ace9 MOV	(%RDX,%RSI,8),%R8

  (4670) 0x55aced MOV	0x8(%RDX,%RSI,8),%R9

  (4670) 0x55acf2 CMP	%R9,%R8

  (4670) 0x55acf5 JL	55ad26

  (4670) 0x55acf7 JMP	55ad5f

    (4672) 0x55ad00 MOV	%RAX,(%RDI,%R10,8)

    (4672) 0x55ad04 MOVSD	%XMM1,(%R12,%RAX,8)

    (4672) 0x55ad0a MOV	-0x50(%RBP),%R9

    (4672) 0x55ad0e MOV	%R10,(%R9,%RAX,8)

    (4672) 0x55ad12 INC	%RAX

    (4672) 0x55ad15 MOV	-0x80(%RBP),%RDX

    (4672) 0x55ad19 MOV	0x8(%RDX,%RSI,8),%R9

    (4672) 0x55ad1e INC	%R8

    (4672) 0x55ad21 CMP	%R9,%R8

    (4672) 0x55ad24 JGE	55ad5f

    (4672) 0x55ad26 MOV	-0x110(%RBP),%RDX

    (4672) 0x55ad2d MOV	(%RDX,%R8,8),%R10

    (4672) 0x55ad31 MOV	(%RDI,%R10,8),%R11

    (4672) 0x55ad35 MOV	-0x108(%RBP),%RDX

    (4672) 0x55ad3c MOVSD	(%RDX,%R8,8),%XMM1

    (4672) 0x55ad42 MULSD	%XMM0,%XMM1

    (4672) 0x55ad46 CMP	%RBX,%R11

    (4672) 0x55ad49 JL	55ad00

    (4672) 0x55ad4b ADDSD	(%R12,%R11,8),%XMM1

    (4672) 0x55ad51 MOVSD	%XMM1,(%R12,%R11,8)

    (4672) 0x55ad57 INC	%R8

    (4672) 0x55ad5a CMP	%R9,%R8

    (4672) 0x55ad5d JL	55ad26

  (4670) 0x55ad5f CMPQ	$0,-0xb8(%RBP)

  (4670) 0x55ad67 JE	55acb0

  (4670) 0x55ad6d MOV	-0x68(%RBP),%R9

  (4670) 0x55ad71 MOV	(%R9,%RSI,8),%R8

  (4670) 0x55ad75 MOV	0x8(%R9,%RSI,8),%R9

  (4670) 0x55ad7a JMP	55ad94

    (4671) 0x55ad80 ADDSD	(%R13,%R11,8),%XMM1

    (4671) 0x55ad87 MOVSD	%XMM1,(%R13,%R11,8)

    (4671) 0x55ad8e MOV	%R14,%R15

    (4671) 0x55ad91 INC	%R8

    (4671) 0x55ad94 CMP	%R9,%R8

    (4671) 0x55ad97 JGE	55acb0

    (4671) 0x55ad9d MOV	-0xd8(%RBP),%R10

    (4671) 0x55ada4 MOV	(%R10,%R8,8),%R10

    (4671) 0x55ada8 MOV	-0xe0(%RBP),%R11

    (4671) 0x55adaf MOV	(%R11,%R10,8),%R10

    (4671) 0x55adb3 ADD	%R10,%R15

    (4671) 0x55adb6 MOV	(%RDI,%R15,8),%R11

    (4671) 0x55adba MOV	-0xd0(%RBP),%RDX

    (4671) 0x55adc1 MOVSD	(%RDX,%R8,8),%XMM1

    (4671) 0x55adc7 MULSD	%XMM0,%XMM1

    (4671) 0x55adcb CMP	-0x30(%RBP),%R11

    (4671) 0x55adcf JGE	55ad80

    (4671) 0x55add1 MOV	%RCX,(%RDI,%R15,8)

    (4671) 0x55add5 MOVSD	%XMM1,(%R13,%RCX,8)

    (4671) 0x55addc MOV	-0x88(%RBP),%R9

    (4671) 0x55ade3 MOV	%R10,(%R9,%RCX,8)

    (4671) 0x55ade7 INC	%RCX

    (4671) 0x55adea MOV	-0x68(%RBP),%R9

    (4671) 0x55adee MOV	0x8(%R9,%RSI,8),%R9

    (4671) 0x55adf3 JMP	55ad8e

/home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/parcsr_mv/par_csr_matop.c: 865 - 989

--------------------------------------------------------------------------------

865:     for (i1 = ns; i1 < ne; i1++)

[...]

874:       if ( allsquare )

875:       {

876:          B_marker[i1] = jj_count_diag;

877:          C_diag_data[jj_count_diag] = zero;

878:          C_diag_j[jj_count_diag] = i1;

879:          jj_count_diag++;

[...]

886:       if (num_cols_offd_A)

887:       {

888:          for (jj2 = A_offd_i[i1]; jj2 < A_offd_i[i1+1]; jj2++)

889:          {

890:             i2 = A_offd_j[jj2];

891:             a_entry = A_offd_data[jj2];

[...]

897:             for (jj3 = B_ext_offd_i[i2]; jj3 < B_ext_offd_i[i2+1]; jj3++)

898:             {

899:                i3 = num_cols_diag_B+B_ext_offd_j[jj3];

[...]

907:                if (B_marker[i3] < jj_row_begin_offd)

908:                {

909:                   B_marker[i3] = jj_count_offd;

910:                   C_offd_data[jj_count_offd] = a_entry*B_ext_offd_data[jj3];

911:                   C_offd_j[jj_count_offd] = i3-num_cols_diag_B;

912:                   jj_count_offd++;

913:                }

914:                else

915:                   C_offd_data[B_marker[i3]] += a_entry*B_ext_offd_data[jj3];

916:             }

917:             for (jj3 = B_ext_diag_i[i2]; jj3 < B_ext_diag_i[i2+1]; jj3++)

918:             {

919:                i3 = B_ext_diag_j[jj3];

920:                if (B_marker[i3] < jj_row_begin_diag)

921:                {

922:                   B_marker[i3] = jj_count_diag;

923:                   C_diag_data[jj_count_diag] = a_entry*B_ext_diag_data[jj3];

924:                   C_diag_j[jj_count_diag] = i3;

925:                   jj_count_diag++;

926:                }

927:                else

928:                   C_diag_data[B_marker[i3]] += a_entry*B_ext_diag_data[jj3];

[...]

937:       for (jj2 = A_diag_i[i1]; jj2 < A_diag_i[i1+1]; jj2++)

938:       {

939:          i2 = A_diag_j[jj2];

940:          a_entry = A_diag_data[jj2];

[...]

946:          for (jj3 = B_diag_i[i2]; jj3 < B_diag_i[i2+1]; jj3++)

947:          {

948:             i3 = B_diag_j[jj3];

[...]

956:             if (B_marker[i3] < jj_row_begin_diag)

957:             {

958:                B_marker[i3] = jj_count_diag;

959:                C_diag_data[jj_count_diag] = a_entry*B_diag_data[jj3];

960:                C_diag_j[jj_count_diag] = i3;

961:                jj_count_diag++;

962:             }

963:             else

964:             {

965:                C_diag_data[B_marker[i3]] += a_entry*B_diag_data[jj3];

966:             }

967:          }

968:          if (num_cols_offd_B)

969:          {

970:             for (jj3 = B_offd_i[i2]; jj3 < B_offd_i[i2+1]; jj3++)

971:             {

972:                i3 = num_cols_diag_B+map_B_to_C[B_offd_j[jj3]];

[...]

980:                if (B_marker[i3] < jj_row_begin_offd)

981:                {

982:                   B_marker[i3] = jj_count_offd;

983:                   C_offd_data[jj_count_offd] = a_entry*B_offd_data[jj3];

984:                   C_offd_j[jj_count_offd] = i3-num_cols_diag_B;

985:                   jj_count_offd++;

986:                }

987:                else

988:                {

989:                   C_offd_data[B_marker[i3]] += a_entry*B_offd_data[jj3];

Coverage (%)	Name	Source Location	Module
►96.67+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_launch_thread		libiomp5.so
○	_INTERNALea951ae6::__kmp_launc[...]		libiomp5.so
○	start_thread		libc.so.6
○	__GI___clone3		libc.so.6
►2.96+	__kmp_invoke_microtask		libiomp5.so
○	__kmp_invoke_task_func		libiomp5.so
○	__kmp_fork_call		libiomp5.so
○	__kmpc_fork_call		libiomp5.so
○	hypre_ParMatmul	par_csr_matop.c:829	exec
○	hypre_BoomerAMGSetup	par_amg_setup.c:1226	exec
○	hypre_PCGSetup	pcg.c:234	exec
○	main	amg.c:398	exec
○	__libc_start_call_main		libc.so.6
○	__libc_start_main		libc.so.6
○	_start		exec

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

min	med	avg	max

Percentile Index	10	20	30	40	50	60	70	80	90	100
Value

Path /

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	10.51
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.21
Bottlenecks	P6, P7, P8, P9,
Function	hypre_ParMatmul.extracted.12
Source	par_csr_matop.c:865-865,par_csr_matop.c:874-879,par_csr_matop.c:886-888,par_csr_matop.c:937-937
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	5.75
CQA cycles if no scalar integer	5.75
CQA cycles if FP arith vectorized	5.75
CQA cycles if fully vectorized	0.55
Front-end cycles	4.75
P0 cycles	3.00
P1 cycles	3.00
P2 cycles	3.00
P3 cycles	3.00
P4 cycles	3.00
P5 cycles	3.00
P6 cycles	5.75
P7 cycles	5.75
P8 cycles	5.75
P9 cycles	5.75
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	38.00
Nb uops	38.00
Nb loads	16.00
Nb stores	7.00
Nb stack references	9.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	32.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	56.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	12.02
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	11.61
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.50

Metric	Value
CQA speedup if no scalar integer	1.00
CQA speedup if FP arith vectorized	1.00
CQA speedup if fully vectorized	10.51
CQA speedup if no inter-iteration dependency	NA
CQA speedup if next bottleneck killed	1.21
Bottlenecks	P6, P7, P8, P9,
Function	hypre_ParMatmul.extracted.12
Source	par_csr_matop.c:865-865,par_csr_matop.c:874-879,par_csr_matop.c:886-888,par_csr_matop.c:937-937
Source loop unroll info	NA
Source loop unroll confidence level	NA
Unroll/vectorization loop type	NA
Unroll factor	NA
CQA cycles	5.75
CQA cycles if no scalar integer	5.75
CQA cycles if FP arith vectorized	5.75
CQA cycles if fully vectorized	0.55
Front-end cycles	4.75
P0 cycles	3.00
P1 cycles	3.00
P2 cycles	3.00
P3 cycles	3.00
P4 cycles	3.00
P5 cycles	3.00
P6 cycles	5.75
P7 cycles	5.75
P8 cycles	5.75
P9 cycles	5.75
P10 cycles	0.00
P11 cycles	0.00
P12 cycles	0.00
P13 cycles	0.00
P14 cycles	0.00
P15 cycles	0.00
DIV/SQRT cycles	0.00
Inter-iter dependencies cycles	NA
FE+BE cycles (UFS)	NA
Stall cycles (UFS)	NA
Nb insns	38.00
Nb uops	38.00
Nb loads	16.00
Nb stores	7.00
Nb stack references	9.00
FLOP/cycle	0.00
Nb FLOP add-sub	0.00
Nb FLOP mul	0.00
Nb FLOP fma	0.00
Nb FLOP div	0.00
Nb FLOP rcp	0.00
Nb FLOP sqrt	0.00
Nb FLOP rsqrt	0.00
Bytes/cycle	32.00
Bytes prefetched	0.00
Bytes loaded	128.00
Bytes stored	56.00
Stride 0	NA
Stride 1	NA
Stride n	NA
Stride unknown	NA
Stride indirect	NA
Vectorization ratio all	0.00
Vectorization ratio load	0.00
Vectorization ratio store	0.00
Vectorization ratio mul	NA
Vectorization ratio add_sub	NA
Vectorization ratio fma	NA
Vectorization ratio div_sqrt	NA
Vectorization ratio other	0.00
Vector-efficiency ratio all	12.02
Vector-efficiency ratio load	12.50
Vector-efficiency ratio store	11.61
Vector-efficiency ratio mul	NA
Vector-efficiency ratio add_sub	NA
Vector-efficiency ratio fma	NA
Vector-efficiency ratio div_sqrt	NA
Vector-efficiency ratio other	12.50

Path /

Average path: Display a virtual path defined by average values of all real paths

Function	hypre_ParMatmul.extracted.12
Source file and lines	par_csr_matop.c:865-989
Module	exec

The loop is defined in /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/parcsr_mv/par_csr_matop.c:865,874-879,886-891,897-899,907-928,937-940,946-948,956-972,980-989.

Analyzed code is defined in /home/eoseret/qaas_runs_ZEN5/173-989-6379/intel/AMG/build/AMG/AMG/parcsr_mv/par_csr_matop.c:865,874-879,886-888,937.

Warnings:

Non-innermost loop: analyzing only self part (ignoring child loops).
Ignoring paths for analysis
Too many paths. If you really need to analyze all of the 134 paths individually, rerun with max-paths=134
RecMII not computed since number of paths is unknown or > max_paths
Streams not analyzed since number of paths is unknown or > max_paths

Try to simplify control and/or increase the maximum number of paths per function/loop through the 'max-paths-nb' option.

This loop has 134 execution paths.

The presence of multiple execution paths is typically the main/first bottleneck.
Try to simplify control inside loop: ideally, try to remove all conditional expressions, for example by (if applicable):

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Your loop is not vectorized. Only 12% of vector register length is used (average across all SSE/AVX instructions). By vectorizing your loop, you can lower the cost of an iteration from 5.75 to 0.55 cycles (10.51x speedup).

Details

All SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 5.75 to 4.75 cycles (1.21x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

No data for this section

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

Matching between your loop (in the source code) and the binary loop

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 128 bytes. The binary loop is storing 56 bytes.

General properties

nb instructions	38
nb uops	38
loop length	167
used x86 registers	10
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	9

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.75 cycles
front end	4.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	3.00	3.00	3.00	3.00	3.00	3.00	5.75	5.75	5.75	5.75	0.00	0.00	0.00	0.00	0.00	0.00
cycles	3.00	3.00	3.00	3.00	3.00	3.00	5.75	5.75	5.75	5.75	0.00	0.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.75
Dispatch	5.75
Overall L1	5.75

Vectorization ratios

all	0%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

all	12%
load	12%
store	11%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	12%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 5.75 cycles. At this rate:

8% of peak load performance is reached (22.26 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
7% of peak store performance is reached (9.74 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 55aac1

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	Latency	Recip. throughput	Vectorization
MOV %RCX,-0x30(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOV %RAX,%RBX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
CMP -0x48(%RBP),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55adf5 <hypre_ParMatmul.extracted.12+0x585>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,%RDX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
CMPQ $0,-0x98(%RBP)	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JE 55aae6 <hypre_ParMatmul.extracted.12+0x276>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %RBX,(%RDI,%RDX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOVQ $0,(%R12,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (6.3%)
MOV -0x50(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %RDX,(%RAX,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
LEA 0x1(%RBX),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
CMPQ $0,-0x90(%RBP)	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JE 55ac70 <hypre_ParMatmul.extracted.12+0x400>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV -0x58(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RCX,%RDX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	scal (12.5%)
LEA 0x1(%RDX),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
CMP 0x8(%RCX,%RDX,8),%R9	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55ac74 <hypre_ParMatmul.extracted.12+0x404>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOV -0x30(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %RDX,-0xb0(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
JMP 55ab3d <hypre_ParMatmul.extracted.12+0x2cd>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A
LEA 0x1(%RDX),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
MOV -0x30(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV -0x60(%RBP),%RSI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RSI,%RDX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
CMP (%RSI,%R8,8),%RDX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55aab0 <hypre_ParMatmul.extracted.12+0x240>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
JMP 55aca9 <hypre_ParMatmul.extracted.12+0x439>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A
MOV %RSI,%RDX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
MOV -0x38(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV -0x60(%RBP),%RSI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RSI,%RDX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
CMP (%RSI,%R8,8),%RDX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	N/A
JGE 55aab0 <hypre_ParMatmul.extracted.12+0x240>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
JMP 55acc9 <hypre_ParMatmul.extracted.12+0x459>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A

Function	hypre_ParMatmul.extracted.12
Source file and lines	par_csr_matop.c:865-989
Module	exec

Non-innermost loop: analyzing only self part (ignoring child loops).
Ignoring paths for analysis
Too many paths. If you really need to analyze all of the 134 paths individually, rerun with max-paths=134
RecMII not computed since number of paths is unknown or > max_paths
Streams not analyzed since number of paths is unknown or > max_paths

hoisting them (moving them outside the loop)
turning them into conditional moves, MIN or MAX

Ex: if (x<0) x=0 => x = (x<0 ? 0 : x) (or MAX(0,x) after defining the corresponding macro)

gain
potential
hint
expert

Vectorization

Details

All SSE/AVX instructions are used in scalar version (process only one data element in vector registers). Since your execution units are vector units, only a vectorized loop can use their full power.

Workaround

Try another compiler or update/tune your current one:
- recompile with ffast-math (included in Ofast) to extend loop vectorization to FP reductions.
Remove inter-iterations dependences from your loop and make it unit-stride:
- If your arrays have 2 or more dimensions, check whether elements are accessed contiguously and, otherwise, try to permute loops accordingly: C storage order is row-major: for(i) for(j) a[j][i] = b[j][i]; (slow, non stride 1) => for(i) for(j) a[i][j] = b[i][j]; (fast, stride 1)
- If your loop streams arrays of structures (AoS), try to use structures of arrays instead (SoA): for(i) a[i].x = b[i].x; (slow, non stride 1) => for(i) a.x[i] = b.x[i]; (fast, stride 1)

Execution units bottlenecks

Performance is limited by:

reading data from caches/RAM (load units are a bottleneck)
writing data to caches/RAM (the store unit is a bottleneck)

By removing all these bottlenecks, you can lower the cost of an iteration from 5.75 to 4.75 cycles (1.21x speedup).

Workaround

Read less array elements
Write less array elements
Provide more information to your compiler:
- hardcode the bounds of the corresponding 'for' loop
- use the 'restrict' C99 keyword

No data for this section

Type of elements and instruction set

No instructions are processing arithmetic or math operations on FP elements. This loop is probably writing/copying data or processing integer elements.

Matching between your loop (in the source code) and the binary loop

The binary loop does not contain any FP arithmetical operations. The binary loop is loading 128 bytes. The binary loop is storing 56 bytes.

General properties

nb instructions	38
nb uops	38
loop length	167
used x86 registers	10
used mmx registers	0
used xmm registers	0
used ymm registers	0
used zmm registers	0
nb stack references	9

Front-end

MACRO FUSION NOT POSSIBLE FIT IN UOP CACHE

micro-operation queue	4.75 cycles
front end	4.75 cycles

Back-end

	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	P10	P11	P12	P13	P14	P15
uops	3.00	3.00	3.00	3.00	3.00	3.00	5.75	5.75	5.75	5.75	0.00	0.00	0.00	0.00	0.00	0.00
cycles	3.00	3.00	3.00	3.00	3.00	3.00	5.75	5.75	5.75	5.75	0.00	0.00	0.00	0.00	0.00	0.00

Execution ports to units layout:

P0: ALU
P1: ALU
P2: ALU
P3: ALU, BRU
P4: ALU, BRU
P5: ALU, BRU
P6 (512 bits): store address, load
P7 (512 bits): store address, load
P8 (512 bits): store address, load
P9 (512 bits): store address, load
P10 (512 bits): VPU, DIV/SQRT
P11 (512 bits): VPU, DIV/SQRT
P12 (512 bits): VPU
P13 (512 bits): VPU
P14 (512 bits): FP store data
P15 (512 bits): FP store data

Cycles executing div or sqrt instructions

Cycles summary

Front-end	4.75
Dispatch	5.75
Overall L1	5.75

Vectorization ratios

all	0%
load	0%
store	0%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	0%

Vector efficiency ratios

all	12%
load	12%
store	11%
mul	NA (no mul vectorizable/vectorized instructions)
add-sub	NA (no add-sub vectorizable/vectorized instructions)
fma	NA (no fma vectorizable/vectorized instructions)
div/sqrt	NA (no div/sqrt vectorizable/vectorized instructions)
other	12%

Cycles and memory resources usage

Assuming all data fit into the L1 cache, each iteration of the binary loop takes 5.75 cycles. At this rate:

8% of peak load performance is reached (22.26 out of 256.00 bytes loaded per cycle (GB/s @ 1GHz))
7% of peak store performance is reached (9.74 out of 128.00 bytes stored per cycle (GB/s @ 1GHz))

Front-end bottlenecks

Found no such bottlenecks.

ASM code

In the binary file, the address of the loop is: 55aac1

Instruction	Nb FU	P0	P1	P2	P3	P4	P5	P6	P7	P8	P9	Latency	Recip. throughput	Vectorization
MOV %RCX,-0x30(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOV %RAX,%RBX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
CMP -0x48(%RBP),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55adf5 <hypre_ParMatmul.extracted.12+0x585>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,%RDX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
CMPQ $0,-0x98(%RBP)	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JE 55aae6 <hypre_ParMatmul.extracted.12+0x276>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %RBX,(%RDI,%RDX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOVQ $0,(%R12,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (6.3%)
MOV -0x50(%RBP),%RAX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %RDX,(%RAX,%RBX,8)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
LEA 0x1(%RBX),%RAX	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
CMPQ $0,-0x90(%RBP)	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JE 55ac70 <hypre_ParMatmul.extracted.12+0x400>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV -0x58(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RCX,%RDX,8),%R9	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	scal (12.5%)
LEA 0x1(%RDX),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
CMP 0x8(%RCX,%RDX,8),%R9	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55ac74 <hypre_ParMatmul.extracted.12+0x404>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
MOV -0x30(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV %RDX,-0xb0(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
JMP 55ab3d <hypre_ParMatmul.extracted.12+0x2cd>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A
LEA 0x1(%RDX),%R8	1	0.17	0.17	0.17	0.17	0.17	0.17	0	0	0	0	1	0.25	N/A
MOV -0x30(%RBP),%RCX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV -0x60(%RBP),%RSI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RSI,%RDX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
CMP (%RSI,%R8,8),%RDX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	scal (12.5%)
JGE 55aab0 <hypre_ParMatmul.extracted.12+0x240>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
JMP 55aca9 <hypre_ParMatmul.extracted.12+0x439>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A
MOV %RSI,%RDX	1	0	0	0	0	0	0	0	0	0	0	0	0.17	N/A
MOV -0x38(%RBP),%R8	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV -0x60(%RBP),%RSI	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
MOV (%RSI,%RDX,8),%RDX	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	3	0.25	N/A
CMP (%RSI,%R8,8),%RDX	1	0.17	0.17	0.17	0.17	0.17	0.17	0.25	0.25	0.25	0.25	1	0.25	N/A
JGE 55aab0 <hypre_ParMatmul.extracted.12+0x240>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	1	0.50-1	N/A
MOV %R8,-0x38(%RBP)	1	0	0	0	0	0	0	0.25	0.25	0.25	0.25	4	0.50	scal (12.5%)
JMP 55acc9 <hypre_ParMatmul.extracted.12+0x459>	1	0	0	0	0.33	0.33	0.33	0	0	0	0	0	1	N/A

Report Configuration

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

Type of elements and instruction set

Matching between your loop (in the source code) and the binary loop

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code

Vectorization

Details

Workaround

Execution units bottlenecks

Workaround

Type of elements and instruction set

Matching between your loop (in the source code) and the binary loop

General properties

Front-end

Back-end

Cycles summary

Vectorization ratios

Vector efficiency ratios

Cycles and memory resources usage

Front-end bottlenecks

ASM code