-
Notifications
You must be signed in to change notification settings - Fork 344
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add RVV qs8/qu8 gavgpool kernels, configs and tests. #7031
Open
oliIMG
wants to merge
7
commits into
google:master
Choose a base branch
from
imaginationtech:img_patch25_qs8_gavgpool
base: master
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
+3,108
−3
Open
Changes from 2 commits
Commits
Show all changes
7 commits
Select commit
Hold shift + click to select a range
bc650ef
Add qs8/qu8 gavgpool kernels, configs and tests.
oliIMG 12f71b3
Merge branch 'google:master' into img_patch25_qs8_gavgpool
oliIMG b0cfeb2
Adhere to standard RVV naming.
oliIMG 8c941c0
Remove old unipass reminants.
oliIMG 8304b4d
Regenerate make files.
oliIMG e24e30d
Merge branch 'google:master' into img_patch25_qs8_gavgpool
KaustubhIMG 8db8f2f
Merge branch 'google:master' into img_patch25_qs8_gavgpool
KaustubhIMG File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
175 changes: 175 additions & 0 deletions
175
src/qs8-gavgpool/gen/qs8-gavgpool-7p7x-minmax-fp32-rvv-c1v.c
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,175 @@ | ||
| // Auto-generated file. Do not edit! | ||
| // Template: src/qs8-gavgpool/multipass-rvv.c.in | ||
| // Generator: tools/xngen | ||
| // | ||
| // Copyright 2024 Imagination Technologies, inc. | ||
| // | ||
| // This source code is licensed under the BSD-style license found in the | ||
| // LICENSE file in the root directory of this source tree. | ||
|
|
||
| #include <assert.h> | ||
|
|
||
| #include <riscv_vector.h> | ||
|
|
||
| #include "xnnpack/gavgpool.h" | ||
| #include "xnnpack/math.h" | ||
|
|
||
|
|
||
| void xnn_qs8_gavgpool_minmax_fp32_ukernel_7p7x__rvv_c1v( | ||
| size_t rows, | ||
| size_t channels, | ||
| const int8_t* input, | ||
| size_t input_stride, | ||
| const int8_t* zero, | ||
| int32_t* buffer, | ||
| int8_t* output, | ||
| const union xnn_qs8_avgpool_minmax_params params[restrict XNN_MIN_ELEMENTS(1)]) | ||
| { | ||
| assert(rows > 7); | ||
| assert(channels != 0); | ||
|
|
||
| const int8_t* i0 = input; | ||
| const int8_t* i1 = (const int8_t*) ((uintptr_t) i0 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows < 2) { | ||
| i1 = zero; | ||
| } | ||
| const int8_t* i2 = (const int8_t*) ((uintptr_t) i1 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows <= 2) { | ||
| i2 = zero; | ||
| } | ||
| const int8_t* i3 = (const int8_t*) ((uintptr_t) i2 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows < 4) { | ||
| i3 = zero; | ||
| } | ||
| const int8_t* i4 = (const int8_t*) ((uintptr_t) i3 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows <= 4) { | ||
| i4 = zero; | ||
| } | ||
| const int8_t* i5 = (const int8_t*) ((uintptr_t) i4 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows < 6) { | ||
| i5 = zero; | ||
| } | ||
| const int8_t* i6 = (const int8_t*) ((uintptr_t) i5 + input_stride); | ||
| if XNN_UNPREDICTABLE(rows <= 6) { | ||
| i6 = zero; | ||
| } | ||
| const size_t input_increment = 7 * input_stride - channels; | ||
|
|
||
| const int32_t init_bias = params->fp32_scalar_fmagic.init_bias; | ||
| int32_t* b = buffer; | ||
| int32_t c = (int32_t) channels; | ||
| do { | ||
| int32_t n = __riscv_vsetvl_e8m1(c); c -= n; | ||
| vint8m1_t i0_i8v = __riscv_vle8_v_i8m1(i0, n); i0 += n; | ||
| vint8m1_t i1_i8v = __riscv_vle8_v_i8m1(i1, n); i1 += n; | ||
| vint16m2_t acc_i16v = __riscv_vwadd_vv_i16m2(i0_i8v, i1_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i2, n); i2 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i3, n); i3 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i4, n); i4 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i5, n); i5 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i6, n); i6 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
|
|
||
| vint32m4_t acc_i32v = __riscv_vwadd_vx_i32m4(acc_i16v, init_bias, n); | ||
| __riscv_vse32_v_i32m4(b, acc_i32v, n); b += n; | ||
| } while (c > 0); | ||
|
|
||
| for (rows -= 7; rows > 7; rows -= 7) { | ||
| i0 = (const int8_t*) ((uintptr_t) i0 + input_increment); | ||
| i1 = (const int8_t*) ((uintptr_t) i1 + input_increment); | ||
| i2 = (const int8_t*) ((uintptr_t) i2 + input_increment); | ||
| i3 = (const int8_t*) ((uintptr_t) i3 + input_increment); | ||
| i4 = (const int8_t*) ((uintptr_t) i4 + input_increment); | ||
| i5 = (const int8_t*) ((uintptr_t) i5 + input_increment); | ||
| i6 = (const int8_t*) ((uintptr_t) i6 + input_increment); | ||
|
|
||
| int32_t* b = buffer; | ||
| int32_t c = (int32_t) channels; | ||
| do { | ||
| int32_t n = __riscv_vsetvl_e8m1(c); c -= n; | ||
| vint8m1_t i0_i8v = __riscv_vle8_v_i8m1(i0, n); i0 += n; | ||
| vint8m1_t i1_i8v = __riscv_vle8_v_i8m1(i1, n); i1 += n; | ||
| vint16m2_t acc_i16v = __riscv_vwadd_vv_i16m2(i0_i8v, i1_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i2, n); i2 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i3, n); i3 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i4, n); i4 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i5, n); i5 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i6, n); i6 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
|
|
||
| vint32m4_t acc_i32v = __riscv_vle32_v_i32m4(b, n); | ||
| acc_i32v = __riscv_vwadd_wv_i32m4(acc_i32v, acc_i16v, n); | ||
| __riscv_vse32_v_i32m4(b, acc_i32v, n); b += n; | ||
| } while (c > 0); | ||
| } | ||
|
|
||
| i0 = (const int8_t*) ((uintptr_t) i0 + input_increment); | ||
| i1 = (const int8_t*) ((uintptr_t) i1 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows < 2) { | ||
| i1 = zero; | ||
| } | ||
| i2 = (const int8_t*) ((uintptr_t) i2 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows <= 2) { | ||
| i2 = zero; | ||
| } | ||
| i3 = (const int8_t*) ((uintptr_t) i3 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows < 4) { | ||
| i3 = zero; | ||
| } | ||
| i4 = (const int8_t*) ((uintptr_t) i4 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows <= 4) { | ||
| i4 = zero; | ||
| } | ||
| i5 = (const int8_t*) ((uintptr_t) i5 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows < 6) { | ||
| i5 = zero; | ||
| } | ||
| i6 = (const int8_t*) ((uintptr_t) i6 + input_increment); | ||
| if XNN_UNPREDICTABLE(rows <= 6) { | ||
| i6 = zero; | ||
| } | ||
|
|
||
| const float scale = params->fp32_scalar_fmagic.scale; | ||
| const float output_min_less_zero_point = params->fp32_scalar_fmagic.output_min_less_zero_point; | ||
| const float output_max_less_zero_point = params->fp32_scalar_fmagic.output_max_less_zero_point; | ||
| const float magic_bias = params->fp32_scalar_fmagic.magic_bias; | ||
| const int32_t magic_bias_less_output_zero_point = params->fp32_scalar_fmagic.magic_bias_less_output_zero_point; | ||
|
|
||
| do { | ||
| int32_t n = __riscv_vsetvl_e8m1(channels); channels -= n; | ||
| vint8m1_t i0_i8v = __riscv_vle8_v_i8m1(i0, n); i0 += n; | ||
| vint8m1_t i1_i8v = __riscv_vle8_v_i8m1(i1, n); i1 += n; | ||
| vint16m2_t acc_i16v = __riscv_vwadd_vv_i16m2(i0_i8v, i1_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i2, n); i2 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i3, n); i3 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i4, n); i4 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i5, n); i5 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
| i0_i8v = __riscv_vle8_v_i8m1(i6, n); i6 += n; | ||
| acc_i16v = __riscv_vwadd_wv_i16m2(acc_i16v, i0_i8v, n); | ||
|
|
||
| vint32m4_t acc_i32v = __riscv_vle32_v_i32m4(buffer, n); buffer += n; | ||
| acc_i32v = __riscv_vwadd_wv_i32m4(acc_i32v, acc_i16v, n); | ||
| vfloat32m4_t acc_f32v = __riscv_vfcvt_f_x_v_f32m4(acc_i32v, n); | ||
| acc_f32v = __riscv_vfmul_vf_f32m4(acc_f32v, scale, n); | ||
| acc_f32v = __riscv_vfmin_vf_f32m4(__riscv_vfmax_vf_f32m4(acc_f32v, output_min_less_zero_point, n), output_max_less_zero_point, n); | ||
| acc_f32v = __riscv_vfadd_vf_f32m4(acc_f32v, magic_bias, n); | ||
|
|
||
| vint32m4_t out_i32v = __riscv_vfcvt_x_f_v_i32m4(acc_f32v, n); | ||
| vint16m2_t out_i16v = __riscv_vncvt_x_x_w_i16m2(out_i32v, n); | ||
| out_i16v = __riscv_vsub_vx_i16m2(out_i16v, magic_bias_less_output_zero_point, n); | ||
| vint8m1_t out_i8v = __riscv_vncvt_x_x_w_i8m1(out_i16v, n); | ||
| __riscv_vse8_v_i8m1(output, out_i8v, n); output += n; | ||
| } while (channels != 0); | ||
| } |
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
c2v function/file naming inconsistent with rvv kernels
should be u2v meaning unrolled to 2 vectors.. aka m2
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
f32 gavgpool uses the old style naming as well, should get fixed