[AMDGPU] New GFX11+ aliases v_dot4_i32_i8 and v_dot8_i32_i4 #118997
Merged
Conversation
This file contains hidden or 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
Previously we decided not to support these aliases because in other generations they are real instructions with different behavior. Now I am inclined to support them anyway for compatibility with SP3.
Member
|
@llvm/pr-subscribers-backend-amdgpu Author: Jay Foad (jayfoad) ChangesPreviously we decided not to support these aliases because in other Full diff: https://github.com/llvm/llvm-project/pull/118997.diff 5 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index bae37358ffe0c7..7638879afcb56c 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1845,6 +1845,11 @@ defm V_DOT4_I32_IU8 : VOP3P_Real_gfx11_gfx12<0x16>;
defm V_DOT8_I32_IU4 : VOP3P_Real_gfx11_gfx12<0x18>;
defm V_DOT2_F32_BF16 : VOP3P_Real_gfx11_gfx12<0x1a>;
+let AssemblerPredicate = isGFX11Plus in {
+ def : AMDGPUMnemonicAlias<"v_dot4_i32_i8", "v_dot4_i32_iu8">;
+ def : AMDGPUMnemonicAlias<"v_dot8_i32_i4", "v_dot8_i32_iu4">;
+}
+
multiclass VOP3P_Real_WMMA <bits<7> op> {
let WaveSizePredicate = isWave32, DecoderNamespace = "GFX11" in {
defm _twoaddr_w32 : VOP3P_Real_Base <GFX11Gen, op>;
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_err.s b/llvm/test/MC/AMDGPU/gfx11_asm_err.s
index 68442b01bf7d90..1b1d1c04359bdf 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_err.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_err.s
@@ -75,20 +75,6 @@ v_cvt_f16_u16_e64_dpp v5, s1 row_shl:1 row_mask:0xf bank_mask:0xf
v_dual_mul_f32 v0, v0, v2 : : v_dual_mul_f32 v1, v1, v3
// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: unknown token in expression
-// On GFX11, v_dot8_i32_i4 is a valid SP3 alias for v_dot8_i32_iu4.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot8_i32_i4 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot8_i32_i4 v0, v1, v2, v3
-// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
-// On GFX11, v_dot4_i32_i8 is a valid SP3 alias for v_dot4_i32_iu8.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot4_i32_i8 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot4_i32_i8 v0, v1, v2, v3
-// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
v_dot4c_i32_i8 v0, v1, v2
// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s
new file mode 100644
index 00000000000000..d2fdf013939fb5
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s
@@ -0,0 +1,7 @@
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -show-encoding %s | FileCheck -check-prefix=GFX11 %s
+
+v_dot4_i32_i8 v5, v1, v2, s3
+// GFX11: v_dot4_i32_iu8 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x16,0xcc,0x01,0x05,0x0e,0x18]
+
+v_dot8_i32_i4 v5, v1, v2, s3
+// GFX11: v_dot8_i32_iu4 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x18,0xcc,0x01,0x05,0x0e,0x18]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
index 5915cbc011863a..fadd24283d211f 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
@@ -5,3 +5,9 @@ v_pk_min_f16 v0, v1, v2
v_pk_max_f16 v0, v1, v2
// GFX12: v_pk_max_num_f16 v0, v1, v2 ; encoding: [0x00,0x40,0x1c,0xcc,0x01,0x05,0x02,0x18]
+
+v_dot4_i32_i8 v5, v1, v2, s3
+// GFX12: v_dot4_i32_iu8 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x16,0xcc,0x01,0x05,0x0e,0x18]
+
+v_dot8_i32_i4 v5, v1, v2, s3
+// GFX12: v_dot8_i32_iu4 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x18,0xcc,0x01,0x05,0x0e,0x18]
diff --git a/llvm/test/MC/AMDGPU/gfx12_err.s b/llvm/test/MC/AMDGPU/gfx12_err.s
index d8578d87279d1a..d55b86b54ec7d8 100644
--- a/llvm/test/MC/AMDGPU/gfx12_err.s
+++ b/llvm/test/MC/AMDGPU/gfx12_err.s
@@ -22,20 +22,6 @@ v_cvt_f16_u16_e64_dpp v5, s1 row_shl:1 row_mask:0xf bank_mask:0xf
v_dual_mul_f32 v0, v0, v2 : : v_dual_mul_f32 v1, v1, v3
// GFX12-ERR: [[@LINE-1]]:{{[0-9]+}}: error: unknown token in expression
-// On GFX12, v_dot8_i32_i4 is a valid SP3 alias for v_dot8_i32_iu4.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot8_i32_i4 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot8_i32_i4 v0, v1, v2, v3
-// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
-// On GFX12, v_dot4_i32_i8 is a valid SP3 alias for v_dot4_i32_iu8.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot4_i32_i8 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot4_i32_i8 v0, v1, v2, v3
-// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
v_dot4c_i32_i8 v0, v1, v2
// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
|
Member
|
@llvm/pr-subscribers-mc Author: Jay Foad (jayfoad) ChangesPreviously we decided not to support these aliases because in other Full diff: https://github.com/llvm/llvm-project/pull/118997.diff 5 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index bae37358ffe0c7..7638879afcb56c 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1845,6 +1845,11 @@ defm V_DOT4_I32_IU8 : VOP3P_Real_gfx11_gfx12<0x16>;
defm V_DOT8_I32_IU4 : VOP3P_Real_gfx11_gfx12<0x18>;
defm V_DOT2_F32_BF16 : VOP3P_Real_gfx11_gfx12<0x1a>;
+let AssemblerPredicate = isGFX11Plus in {
+ def : AMDGPUMnemonicAlias<"v_dot4_i32_i8", "v_dot4_i32_iu8">;
+ def : AMDGPUMnemonicAlias<"v_dot8_i32_i4", "v_dot8_i32_iu4">;
+}
+
multiclass VOP3P_Real_WMMA <bits<7> op> {
let WaveSizePredicate = isWave32, DecoderNamespace = "GFX11" in {
defm _twoaddr_w32 : VOP3P_Real_Base <GFX11Gen, op>;
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_err.s b/llvm/test/MC/AMDGPU/gfx11_asm_err.s
index 68442b01bf7d90..1b1d1c04359bdf 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_err.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_err.s
@@ -75,20 +75,6 @@ v_cvt_f16_u16_e64_dpp v5, s1 row_shl:1 row_mask:0xf bank_mask:0xf
v_dual_mul_f32 v0, v0, v2 : : v_dual_mul_f32 v1, v1, v3
// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: unknown token in expression
-// On GFX11, v_dot8_i32_i4 is a valid SP3 alias for v_dot8_i32_iu4.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot8_i32_i4 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot8_i32_i4 v0, v1, v2, v3
-// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
-// On GFX11, v_dot4_i32_i8 is a valid SP3 alias for v_dot4_i32_iu8.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot4_i32_i8 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot4_i32_i8 v0, v1, v2, v3
-// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
v_dot4c_i32_i8 v0, v1, v2
// GFX11: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s
new file mode 100644
index 00000000000000..d2fdf013939fb5
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_alias.s
@@ -0,0 +1,7 @@
+// RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -show-encoding %s | FileCheck -check-prefix=GFX11 %s
+
+v_dot4_i32_i8 v5, v1, v2, s3
+// GFX11: v_dot4_i32_iu8 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x16,0xcc,0x01,0x05,0x0e,0x18]
+
+v_dot8_i32_i4 v5, v1, v2, s3
+// GFX11: v_dot8_i32_iu4 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x18,0xcc,0x01,0x05,0x0e,0x18]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
index 5915cbc011863a..fadd24283d211f 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_aliases.s
@@ -5,3 +5,9 @@ v_pk_min_f16 v0, v1, v2
v_pk_max_f16 v0, v1, v2
// GFX12: v_pk_max_num_f16 v0, v1, v2 ; encoding: [0x00,0x40,0x1c,0xcc,0x01,0x05,0x02,0x18]
+
+v_dot4_i32_i8 v5, v1, v2, s3
+// GFX12: v_dot4_i32_iu8 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x16,0xcc,0x01,0x05,0x0e,0x18]
+
+v_dot8_i32_i4 v5, v1, v2, s3
+// GFX12: v_dot8_i32_iu4 v5, v1, v2, s3 ; encoding: [0x05,0x40,0x18,0xcc,0x01,0x05,0x0e,0x18]
diff --git a/llvm/test/MC/AMDGPU/gfx12_err.s b/llvm/test/MC/AMDGPU/gfx12_err.s
index d8578d87279d1a..d55b86b54ec7d8 100644
--- a/llvm/test/MC/AMDGPU/gfx12_err.s
+++ b/llvm/test/MC/AMDGPU/gfx12_err.s
@@ -22,20 +22,6 @@ v_cvt_f16_u16_e64_dpp v5, s1 row_shl:1 row_mask:0xf bank_mask:0xf
v_dual_mul_f32 v0, v0, v2 : : v_dual_mul_f32 v1, v1, v3
// GFX12-ERR: [[@LINE-1]]:{{[0-9]+}}: error: unknown token in expression
-// On GFX12, v_dot8_i32_i4 is a valid SP3 alias for v_dot8_i32_iu4.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot8_i32_i4 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot8_i32_i4 v0, v1, v2, v3
-// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
-// On GFX12, v_dot4_i32_i8 is a valid SP3 alias for v_dot4_i32_iu8.
-// However, we intentionally leave it unimplemented because on other
-// processors v_dot4_i32_i8 denotes an instruction of a different
-// behaviour, which is considered potentially dangerous.
-v_dot4_i32_i8 v0, v1, v2, v3
-// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-
v_dot4c_i32_i8 v0, v1, v2
// GFX12-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
|
Contributor
Author
|
@Sisyph @mbrkusanin any thoughts about this? My immediate motivation is to make it simpler to do automated testing against SP3. |
rampitec
approved these changes
Dec 6, 2024
Collaborator
rampitec
left a comment
rampitec
left a comment
There was a problem hiding this comment.
That's always a bad idea to repurpose instructions' names, but it is already done. LGTM for the purpose of SP3 compatibility.
|
wouldn't it make sense to set |
Sisyph
approved these changes
Dec 6, 2024
Contributor
Sisyph
left a comment
Sisyph
left a comment
There was a problem hiding this comment.
LGTM. I'm having a hard time even picturing a scenario where having these aliases would confuse someone. Emitting textual assembly on an old target and reassemble it on gfx11? Porting inline asm?
Contributor
Author
Maybe, if the goal was to provide backwards compatibility of handwritten assembly code. But that's really not a goal of the LLVM assembler. The only reason for implementing these aliases is for compatibility with the proprietary assembler. |
searlmc1
pushed a commit
to ROCm/llvm-project
that referenced
this pull request
Feb 3, 2025
) (#282) Previously we decided not to support these aliases because in other generations they are real instructions with different behavior. Now I am inclined to support them anyway for compatibility with SP3. Co-authored-by: Jay Foad <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Previously we decided not to support these aliases because in other
generations they are real instructions with different behavior. Now I am
inclined to support them anyway for compatibility with SP3.