Skip to content

Update quick-tune lists for Strix and Navi48 #2073

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.

Sign up for GitHub

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

Open
wants to merge 5 commits into
base: develop
Choose a base branch
from
+655 −196
Open

Update quick-tune lists for Strix and Navi48 #2073

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
811f610
Update gfx1201 GEMM and CONV quick-tune lists.
mirza-halilcevic Oct 29, 2025
d154f50
Update gfx1151 GEMM and CONV quick-tune lists.
mirza-halilcevic Nov 4, 2025
492c068
Update gfx1150 GEMM and CONV quick-tune lists.
mirza-halilcevic Nov 4, 2025
726f500
Update perf configs to v4.
mirza-halilcevic Nov 27, 2025
992918e
Fix tests.
mirza-halilcevic Nov 6, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
797 changes: 628 additions & 169 deletions mlir/include/mlir/Dialect/Rock/Tuning/QuickTuningPerfconfigs.inc
View file
Open in desktop

Large diffs are not rendered by default.

28 changes: 14 additions & 14 deletions mlir/test/Dialect/Rock/affix_tuning_params.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,9 +10,9 @@
// GRID-LABEL: rock_conv
func.func @rock_conv(%filter : memref<1x128x8x3x3xf32>, %input : memref<128x1x8x32x32xf32>, %output : memref<128x1x128x30x30xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906"} {
// CHECK: rock.conv
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 8, mPerBlock = 128, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 4, mPerBlock = 64, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 900
// GRID-SAME: gridSize = 1800
rock.conv(%filter, %input, %output) features = none {
filter_layout = ["g", "k", "c", "0", "1"],
input_layout = ["ni", "gi", "ci", "0i", "1i"],
Expand All @@ -28,9 +28,9 @@ func.func @rock_conv(%filter : memref<1x128x8x3x3xf32>, %input : memref<128x1x8x
// GRID-LABEL: rock_conv_schedulev2
func.func @rock_conv_schedulev2(%filter : memref<1x128x8x3x3xf32>, %input : memref<128x1x8x32x32xf32>, %output : memref<128x1x128x30x30xf32>) attributes {schedule_version = #rock.schedule_version<2>, arch = "amdgcn-amd-amdhsa:gfx906"} {
// CHECK: rock.conv
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 8, mPerBlock = 128, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 2, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 4, mPerBlock = 64, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 2, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 900
// GRID-SAME: gridSize = 1800
rock.conv(%filter, %input, %output) features = none {
filter_layout = ["g", "k", "c", "0", "1"],
input_layout = ["ni", "gi", "ci", "0i", "1i"],
Expand All @@ -46,9 +46,9 @@ func.func @rock_conv_schedulev2(%filter : memref<1x128x8x3x3xf32>, %input : memr
// GRID-LABEL: func.func @rock_conv_f16
func.func @rock_conv_f16(%filter : memref<1x128x8x3x3xf16>, %input : memref<128x1x8x32x32xf16>, %output : memref<128x1x128x30x30xf16>) attributes {arch = "amdgcn-amd-amdhsa:gfx906"} {
// CHECK: rock.conv
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 8, mPerBlock = 128, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 4, mPerBlock = 64, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 900
// GRID-SAME: gridSize = 1800
rock.conv(%filter, %input, %output) features = none {
filter_layout = ["g", "k", "c", "0", "1"],
input_layout = ["ni", "gi", "ci", "0i", "1i"],
Expand Down Expand Up @@ -125,7 +125,7 @@ func.func @rock_conv_bwd_data_f16(%filter: memref<1x1024x1024x1x1xf16>, %input:
// GRID-LABEL: func.func @rock_conv_bwd_data_padMN
func.func @rock_conv_bwd_data_padMN(%filter : memref<1x64x3x1x1xf32>, %input : memref<11x1x3x15x15xf32>, %output : memref<11x1x64x15x15xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906"} {
// CHECK: rock.conv_bwd_data
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 16, mPerBlock = 32, nPerBlock = 64, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 8, mPerBlock = 32, nPerBlock = 64, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 39
rock.conv_bwd_data(%filter, %input, %output) features = none {
Expand All @@ -145,9 +145,9 @@ func.func @rock_conv_bwd_data_padMN(%filter : memref<1x64x3x1x1xf32>, %input : m
// GRID-LABEL: @rock_conv_bwd_data_padMK
func.func @rock_conv_bwd_data_padMK(%filter : memref<1x11x3x1x1xf32>, %input : memref<128x1x3x15x15xf32>, %output : memref<128x1x11x15x15xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906"} {
// CHECK: rock.conv_bwd_data
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 4, mPerBlock = 32, nPerBlock = 64, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 4, mPerBlock = 32, nPerBlock = 128, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 450
// GRID-SAME: gridSize = 225
rock.conv_bwd_data(%filter, %input, %output) features = none {
filter_layout = ["g", "k", "c", "0", "1"],
input_layout = ["ni", "gi", "ci", "0i", "1i"],
Expand All @@ -165,7 +165,7 @@ func.func @rock_conv_bwd_data_padMK(%filter : memref<1x11x3x1x1xf32>, %input : m
// GRID-LABEL: @rock_conv_bwd_weight
func.func @rock_conv_bwd_weight(%filter : memref<1x128x8x3x3xf32>, %input : memref<128x1x8x32x32xf32>, %output : memref<128x1x128x30x30xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906", numCU = 64 : i32} {
// CHECK: rock.conv_bwd_weight
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 16, mPerBlock = 64, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 8, mPerBlock = 64, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 6
rock.conv_bwd_weight(%filter, %input, %output) features = none {
Expand All @@ -183,7 +183,7 @@ func.func @rock_conv_bwd_weight(%filter : memref<1x128x8x3x3xf32>, %input : memr
// GRID-LABEL: @rock_conv_bwd_weight_f16
func.func @rock_conv_bwd_weight_f16(%filter : memref<1x128x8x3x3xf16>, %input : memref<128x1x8x32x32xf16>, %output : memref<128x1x128x30x30xf16>) attributes {arch = "amdgcn-amd-amdhsa:gfx906", numCU = 64 : i32} {
// CHECK: rock.conv_bwd_weight
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 16, mPerBlock = 64, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 8, mPerBlock = 64, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 6
rock.conv_bwd_weight(%filter, %input, %output) features = none {
Expand All @@ -201,7 +201,7 @@ func.func @rock_conv_bwd_weight_f16(%filter : memref<1x128x8x3x3xf16>, %input :
// GRID-LABEL: func.func @rock_conv_bwd_weight_padALL
func.func @rock_conv_bwd_weight_padALL(%filter : memref<1x20x8x3x3xf32>, %input : memref<7x1x8x32x32xf32>, %output : memref<7x1x20x30x30xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906", numCU = 64 : i32} {
// CHECK: rock.conv_bwd_weight
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 16, mPerBlock = 32, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 16, mPerBlock = 32, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 3
rock.conv_bwd_weight(%filter, %input, %output) features = none {
Expand All @@ -219,7 +219,7 @@ func.func @rock_conv_bwd_weight_padALL(%filter : memref<1x20x8x3x3xf32>, %input
// GRID-LABEL: @rock_conv_bwd_weight_padALL_f16
func.func @rock_conv_bwd_weight_padALL_f16(%filter : memref<1x20x8x3x3xf16>, %input : memref<7x1x8x32x32xf16>, %output : memref<7x1x20x30x30xf16>) attributes {arch = "amdgcn-amd-amdhsa:gfx906", numCU = 64 : i32} {
// CHECK: rock.conv_bwd_weight
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 64, kPerBlock = 16, mPerBlock = 32, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 16, mPerBlock = 32, nPerBlock = 32, kPerThread = 1, mPerThread = 2, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 3
rock.conv_bwd_weight(%filter, %input, %output) features = none {
Expand Down Expand Up @@ -340,7 +340,7 @@ func.func @rock_conv_bwd_data_7x7(%arg0: memref<1x64x3x7x7xf32>, %arg1: memref<2
// GRID-LABEL: @rock_gemm_from_conv
func.func @rock_gemm_from_conv(%a : memref<1x72x128xf32>, %b : memref<1x72x115200xf32>, %c : memref<1x128x115200xf32>) attributes {arch = "amdgcn-amd-amdhsa:gfx906", numCU = 64 : i32} {
// CHECK: rock.gemm
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 8, mPerBlock = 128, nPerBlock = 128, kPerThread = 1, mPerThread = 4, nPerThread = 4, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// CHECK-SAME: params = #rock.general_gemm_params<blockSize = 128, kPerBlock = 4, mPerBlock = 128, nPerBlock = 128, kPerThread = 1, mPerThread = 4, nPerThread = 2, kpack = 1, splitKFactor = 1, scheduleVersion = 1, outputSwizzle = 2>
// GRID: rock.gridwise_gemm
// GRID-SAME: gridSize = 900
rock.gemm %c = tr %a * %b features = none storeMethod = set
Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/noTransA-noTransB/broadcasted-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA false -transB false -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: bVectorLen: 2

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/noTransA-transB/broadcasted-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA false -transB true -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::K
// VECTORIZATION-NEXT: bVectorLen: 8

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/transA-noTransB/gemm-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA true -transB false -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: bVectorLen: 2

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/transA-noTransB/sliced-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA true -transB false -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: bVectorLen: 2

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/transA-transB/gemm-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA true -transB true -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::K
// VECTORIZATION-NEXT: bVectorLen: 8

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/pr-e2e/gemm-layouts/transA-transB/sliced-k-e2e.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@
// EMITKEY: -t f16 -out_datatype f16 -transA true -transB true -g 2 -m 4096 -n 640 -k 320

// VECTORIZATION: aVectorDim: GemmDimension::MorN
// VECTORIZATION-NEXT: aVectorLen: 2
// VECTORIZATION-NEXT: aVectorLen: 4
// VECTORIZATION: bVectorDim: GemmDimension::K
// VECTORIZATION-NEXT: bVectorLen: 8

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/tosa-to-rock-gemm-reshape-add.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@

// CHECK_LINALG_ALIGN-COUNT-2: rock.threadwise_read_into {{.*}}
// CHECK_LINALG_ALIGN: rock.threadwise_read_into {{.*}} -> [[lain:%.*]] :
// CHECK_LINALG_ALIGN: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<32xf32, #gpu.address_space<private>>)
// CHECK_LINALG_ALIGN: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<8xf32, #gpu.address_space<private>>)
// CHECK_LINALG_ALIGN: rock.threadwise_write_all {{.*}} %[[outBuf]] ->
// to test reshape is converted as transform and fused.

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/tosa-to-rock-tp-add-tp.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
// CHECK-DAG: #[[MAP2:.*]] = #rock.transform_map<{{.*}} by [<PassThrough ["dim0", "dim2", "dim3", "dim1"] at [0, 1, 2, 3] -> ["dim0", "dim2", "dim3", "dim1"] at [0, 2, 3, 1]>] bounds = [256, 28, 28, 64] -> [256, 64, 28, 28]>
// CHECK-COUNT-2: rock.threadwise_read_into {{.*}}
// CHECK: rock.threadwise_read_into {{.*}} -> [[lain:%.*]] :
// CHECK: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<32xf32, #gpu.address_space<private>>)
// CHECK: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<128xf32, #gpu.address_space<private>>)
// CHECK: rock.threadwise_write_all {{.*}} %[[outBuf]] ->
// to test transpose is converted as transform and fused.

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/fusion/tosa-to-rock-tp-add.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
// CHECK-DAG: #[[MAP2:.*]] = #rock.transform_map<#map{{.*}} by [<PassThrough ["{{.*}}", "{{.*}}", "{{.*}}", "{{.*}}"] at [0, 1, 2, 3] -> ["{{.*}}", "{{.*}}", "{{.*}}", "{{.*}}"] at [0, 2, 3, 1]>] bounds = [256, 28, 28, 64] -> [256, 64, 28, 28]>
// CHECK-COUNT-2: rock.threadwise_read_into {{.*}}
// CHECK: rock.threadwise_read_into {{.*}} -> [[lain:%.*]] :
// CHECK: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<32xf32, #gpu.address_space<private>>)
// CHECK: linalg.generic{{.*}} ins({{.*}}, [[lain]] :{{.*}}) outs(%[[outBuf:.*]] : memref<128xf32, #gpu.address_space<private>>)
// CHECK: rock.threadwise_write_all {{.*}} %[[outBuf]] ->
// to test transpose is converted as transform and fused.

Expand Down
2 changes: 1 addition & 1 deletion mlir/test/mlir-rock-lib/populate_bwd.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@

// ZEROINIT_KERNELCOUNT: Kernel count=1
// ZEROINIT_BIN: ELF
// ZEROINIT_TUNING: globalSize=100352, localSize=128
// ZEROINIT_TUNING: globalSize=100352, localSize=64
// ZEROINIT_DRIVER: %arg1: memref<{{.*}}xf16> {rock.prefill = 0.000000e+00 : f16}
// ZEROINIT_DRIVER-COUNT-3: rock.transform %{{.+}} by
// ZEROINIT_DRIVER-NEXT: rock.conv_bwd_data(%{{.+}}, %{{.+}}, %{{.+}}) features = dot {dilations = [1 : index, 1 : index], filter_layout = ["g", "k", "c", "0", "1"], input_layout = ["ni", "gi", "ci", "0i", "1i"], kernelId = 0 : index, output_layout = ["no", "go", "ko", "0o", "1o"], padding = [0 : index, 0 : index, 0 : index, 0 : index], strides = [2 : index, 2 : index], usesV4R1 = true} : memref<1x2048x1024x1x1xf16>, memref<256x1x1024x14x14xf16>, memref<256x1x2048x7x7xf16>
6 changes: 3 additions & 3 deletions mlir/unittests/Dialect/Rock/ParamLookupTableTests.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,8 +25,8 @@ TEST(FindFallbackTest, OldestRelative) {
}

TEST(FindFallbackTest, YoungestRelative) {
// gfx1200 is the youngest available relative for gfx1900
EXPECT_EQ("gfx1200_conv_f16", ParamLookupTable<InitParamsAccel>::findFallback(
// gfx1201 is the youngest available relative for gfx1900
EXPECT_EQ("gfx1201_conv_f16", ParamLookupTable<InitParamsAccel>::findFallback(
"gfx1900_conv_f16"));
}

Expand Down Expand Up @@ -63,6 +63,6 @@ TEST(FindFallbackTest, NoRelativesBySuffix) {
TEST(FindFallbackTest, AnyGfxForNonAccel) {
// Any gfx version is acceptable for non-accelerated operations
EXPECT_EQ(
"gfx1200_gemm_f32",
"gfx1201_gemm_f32",
ParamLookupTable<InitParamsNonAccel>::findFallback("gfx942_gemm_f32"));
}