AMDGPU: Handle true16 disassembly of ds_write_b8/b16 #156406

arsenm · 2025-09-02T06:07:53Z

This avoids an inconsistency in pseudo definitions I
ran into for a later patch.

arsenm · 2025-09-02T06:08:12Z

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-09-02T06:08:23Z

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

This avoids an inconsistency in pseudo definitions I
ran into for a later patch.

Patch is 25.26 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/156406.diff

5 Files Affected:

(modified) llvm/lib/Target/AMDGPU/DSInstructions.td (+37-2)
(modified) llvm/lib/Target/AMDGPU/SIInstrInfo.td (+3-1)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt (+48-48)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_ds.txt (+6-6)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_ds.txt (+24-24)

diff --git a/llvm/lib/Target/AMDGPU/DSInstructions.td b/llvm/lib/Target/AMDGPU/DSInstructions.td
index a9376250931b6..5e4c19e389cf6 100644
--- a/llvm/lib/Target/AMDGPU/DSInstructions.td
+++ b/llvm/lib/Target/AMDGPU/DSInstructions.td
@@ -1380,6 +1380,21 @@ multiclass DS_Real_gfx12_with_name<bits<8> op, string name> {
   defm "" : DS_Real_gfx12<op, !cast<DS_Pseudo>(NAME), name>;
 }
 
+multiclass DS_Real_gfx12_with_t16<bits<8> op,
+                                  string name = !tolower(NAME)> {
+  let DecoderNamespace = "GFX12_FAKE16" in {
+    defm "" : DS_Real_gfx12<op, !cast<DS_Pseudo>(NAME#"_gfx9"), name>;
+  }
+
+  let DecoderNamespace = "GFX12" in {
+    def _gfx12_t16 :
+      Base_DS_Real_gfx6_gfx7_gfx10_gfx11_gfx12<op,
+          !cast<DS_Pseudo>(NAME#"_t16"),
+          SIEncodingFamily.GFX12, name>,
+      True16D16Table<NAME#"_D16_HI", NAME>;
+  }
+}
+
 defm DS_MIN_F32           : DS_Real_gfx12_with_name<0x012, "ds_min_num_f32">;
 defm DS_MAX_F32           : DS_Real_gfx12_with_name<0x013, "ds_max_num_f32">;
 defm DS_MIN_RTN_F32       : DS_Real_gfx12_with_name<0x032, "ds_min_num_rtn_f32">;
@@ -1444,17 +1459,37 @@ multiclass DS_Real_gfx11<bits<8> op, DS_Pseudo ps = !cast<DS_Pseudo>(NAME),
   } // End AssemblerPredicate
 }
 
+multiclass DS_Real_gfx11_with_t16<bits<8> op,
+                                  string name = !tolower(NAME)> {
+  let DecoderNamespace = "GFX11_FAKE16" in {
+    defm "" : DS_Real_gfx11<op, !cast<DS_Pseudo>(NAME#"_gfx9"), name>;
+  }
+
+  let DecoderNamespace = "GFX11" in {
+    def _gfx11_t16 :
+      Base_DS_Real_gfx6_gfx7_gfx10_gfx11_gfx12<op,
+          !cast<DS_Pseudo>(NAME#"_t16"),
+          SIEncodingFamily.GFX11, name>,
+      True16D16Table<NAME#"_D16_HI", NAME#"_gfx9">;
+  }
+}
+
 multiclass DS_Real_gfx11_gfx12<bits<8> op,
                                string name = !tolower(NAME),
                                DS_Pseudo ps = !cast<DS_Pseudo>(NAME)>
   : DS_Real_gfx11<op, ps, name>,
     DS_Real_gfx12<op, ps, name>;
 
+multiclass DS_Real_gfx11_gfx12_with_t16<bits<8> op,
+                                        string name = !tolower(NAME)>
+  : DS_Real_gfx11_with_t16<op, name>,
+    DS_Real_gfx12_with_t16<op, name>;
+
 defm DS_WRITE_B32           : DS_Real_gfx11_gfx12<0x00d, "ds_store_b32">;
 defm DS_WRITE2_B32          : DS_Real_gfx11_gfx12<0x00e, "ds_store_2addr_b32">;
 defm DS_WRITE2ST64_B32      : DS_Real_gfx11_gfx12<0x00f, "ds_store_2addr_stride64_b32">;
-defm DS_WRITE_B8            : DS_Real_gfx11_gfx12<0x01e, "ds_store_b8">;
-defm DS_WRITE_B16           : DS_Real_gfx11_gfx12<0x01f, "ds_store_b16">;
+defm DS_WRITE_B8            : DS_Real_gfx11_gfx12_with_t16<0x01e, "ds_store_b8">;
+defm DS_WRITE_B16           : DS_Real_gfx11_gfx12_with_t16<0x01f, "ds_store_b16">;
 defm DS_WRXCHG_RTN_B32      : DS_Real_gfx11_gfx12<0x02d, "ds_storexchg_rtn_b32">;
 defm DS_WRXCHG2_RTN_B32     : DS_Real_gfx11_gfx12<0x02e, "ds_storexchg_2addr_rtn_b32">;
 defm DS_WRXCHG2ST64_RTN_B32 : DS_Real_gfx11_gfx12<0x02f, "ds_storexchg_2addr_stride64_rtn_b32">;
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.td b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
index 63c938b259f35..cf9fe74f977ea 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.td
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.td
@@ -12,7 +12,9 @@ def isWave64 : Predicate<"Subtarget->isWave64()">,
   AssemblerPredicate <(all_of FeatureWavefrontSize64)>;
 
 class AMDGPUMnemonicAlias<string From, string To, string VariantName = "">
-    : MnemonicAlias<From, To, VariantName>, PredicateControl;
+    : MnemonicAlias<From, To, VariantName>, PredicateControl {
+  string DecoderNamespace = ?; // Dummy field
+}
 
 // Except for the NONE field, this must be kept in sync with the
 // SIEncodingFamily enum in SIInstrInfo.cpp and the columns of the
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt
index 448cfc95de095..2730b5be92f58 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt
@@ -4447,76 +4447,76 @@
 # GFX11: ds_store_b128 v255, v[2:5] offset:65535 ; encoding: [0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00]
 0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1                     ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l                   ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1 gds                 ; encoding: [0x00,0x00,0x7e,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l gds               ; encoding: [0x00,0x00,0x7e,0xd8,0x00,0x01,0x00,0x00]
 0x00,0x00,0x7e,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1 offset:4660         ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l offset:4660       ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1 offset:4660 gds     ; encoding: [0x34,0x12,0x7e,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l offset:4660 gds   ; encoding: [0x34,0x12,0x7e,0xd8,0x00,0x01,0x00,0x00]
 0x34,0x12,0x7e,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1 offset:65535        ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l offset:65535      ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v1 offset:65535 gds    ; encoding: [0xff,0xff,0x7e,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l offset:65535 gds  ; encoding: [0xff,0xff,0x7e,0xd8,0x00,0x01,0x00,0x00]
 0xff,0xff,0x7e,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254                   ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l                 ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254 gds               ; encoding: [0x00,0x00,0x7e,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l gds             ; encoding: [0x00,0x00,0x7e,0xd8,0x00,0xfe,0x00,0x00]
 0x00,0x00,0x7e,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254 offset:4660       ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l offset:4660     ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0xfe,0x00,0x00]
 0x34,0x12,0x7c,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254 offset:4660 gds   ; encoding: [0x34,0x12,0x7e,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l offset:4660 gds ; encoding: [0x34,0x12,0x7e,0xd8,0x00,0xfe,0x00,0x00]
 0x34,0x12,0x7e,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254 offset:65535      ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l offset:65535    ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0xfe,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v0, v254 offset:65535 gds  ; encoding: [0xff,0xff,0x7e,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v0, v254.l offset:65535 gds ; encoding: [0xff,0xff,0x7e,0xd8,0x00,0xfe,0x00,0x00]
 0xff,0xff,0x7e,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1                   ; encoding: [0x00,0x00,0x7c,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l                 ; encoding: [0x00,0x00,0x7c,0xd8,0xff,0x01,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1 gds               ; encoding: [0x00,0x00,0x7e,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l gds             ; encoding: [0x00,0x00,0x7e,0xd8,0xff,0x01,0x00,0x00]
 0x00,0x00,0x7e,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1 offset:4660       ; encoding: [0x34,0x12,0x7c,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l offset:4660     ; encoding: [0x34,0x12,0x7c,0xd8,0xff,0x01,0x00,0x00]
 0x34,0x12,0x7c,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1 offset:4660 gds   ; encoding: [0x34,0x12,0x7e,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l offset:4660 gds ; encoding: [0x34,0x12,0x7e,0xd8,0xff,0x01,0x00,0x00]
 0x34,0x12,0x7e,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1 offset:65535      ; encoding: [0xff,0xff,0x7c,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l offset:65535    ; encoding: [0xff,0xff,0x7c,0xd8,0xff,0x01,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v1 offset:65535 gds  ; encoding: [0xff,0xff,0x7e,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v255, v1.l offset:65535 gds ; encoding: [0xff,0xff,0x7e,0xd8,0xff,0x01,0x00,0x00]
 0xff,0xff,0x7e,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254                 ; encoding: [0x00,0x00,0x7c,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l               ; encoding: [0x00,0x00,0x7c,0xd8,0xff,0xfe,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254 gds             ; encoding: [0x00,0x00,0x7e,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l gds           ; encoding: [0x00,0x00,0x7e,0xd8,0xff,0xfe,0x00,0x00]
 0x00,0x00,0x7e,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254 offset:4660     ; encoding: [0x34,0x12,0x7c,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l offset:4660   ; encoding: [0x34,0x12,0x7c,0xd8,0xff,0xfe,0x00,0x00]
 0x34,0x12,0x7c,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254 offset:4660 gds ; encoding: [0x34,0x12,0x7e,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l offset:4660 gds ; encoding: [0x34,0x12,0x7e,0xd8,0xff,0xfe,0x00,0x00]
 0x34,0x12,0x7e,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254 offset:65535    ; encoding: [0xff,0xff,0x7c,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l offset:65535  ; encoding: [0xff,0xff,0x7c,0xd8,0xff,0xfe,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b16 v255, v254 offset:65535 gds ; encoding: [0xff,0xff,0x7e,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b16 v255, v254.l offset:65535 gds ; encoding: [0xff,0xff,0x7e,0xd8,0xff,0xfe,0x00,0x00]
 0xff,0xff,0x7e,0xd8,0xff,0xfe,0x00,0x00
 
 # GFX11: ds_store_b16_d16_hi v1, v2              ; encoding: [0x00,0x00,0x84,0xda,0x01,0x02,0x00,0x00]
@@ -4627,76 +4627,76 @@
 # GFX11: ds_store_b64 v255, v[2:3] offset:65535  ; encoding: [0xff,0xff,0x34,0xd9,0xff,0x02,0x00,0x00]
 0xff,0xff,0x34,0xd9,0xff,0x02,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1                      ; encoding: [0x00,0x00,0x78,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l                    ; encoding: [0x00,0x00,0x78,0xd8,0x00,0x01,0x00,0x00]
 0x00,0x00,0x78,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1 gds                  ; encoding: [0x00,0x00,0x7a,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l gds                ; encoding: [0x00,0x00,0x7a,0xd8,0x00,0x01,0x00,0x00]
 0x00,0x00,0x7a,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1 offset:4660          ; encoding: [0x34,0x12,0x78,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l offset:4660        ; encoding: [0x34,0x12,0x78,0xd8,0x00,0x01,0x00,0x00]
 0x34,0x12,0x78,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1 offset:4660 gds      ; encoding: [0x34,0x12,0x7a,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l offset:4660 gds    ; encoding: [0x34,0x12,0x7a,0xd8,0x00,0x01,0x00,0x00]
 0x34,0x12,0x7a,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1 offset:65535         ; encoding: [0xff,0xff,0x78,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l offset:65535       ; encoding: [0xff,0xff,0x78,0xd8,0x00,0x01,0x00,0x00]
 0xff,0xff,0x78,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v1 offset:65535 gds     ; encoding: [0xff,0xff,0x7a,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v0, v1.l offset:65535 gds   ; encoding: [0xff,0xff,0x7a,0xd8,0x00,0x01,0x00,0x00]
 0xff,0xff,0x7a,0xd8,0x00,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254                    ; encoding: [0x00,0x00,0x78,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l                  ; encoding: [0x00,0x00,0x78,0xd8,0x00,0xfe,0x00,0x00]
 0x00,0x00,0x78,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254 gds                ; encoding: [0x00,0x00,0x7a,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l gds              ; encoding: [0x00,0x00,0x7a,0xd8,0x00,0xfe,0x00,0x00]
 0x00,0x00,0x7a,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254 offset:4660        ; encoding: [0x34,0x12,0x78,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l offset:4660      ; encoding: [0x34,0x12,0x78,0xd8,0x00,0xfe,0x00,0x00]
 0x34,0x12,0x78,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254 offset:4660 gds    ; encoding: [0x34,0x12,0x7a,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l offset:4660 gds  ; encoding: [0x34,0x12,0x7a,0xd8,0x00,0xfe,0x00,0x00]
 0x34,0x12,0x7a,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254 offset:65535       ; encoding: [0xff,0xff,0x78,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l offset:65535     ; encoding: [0xff,0xff,0x78,0xd8,0x00,0xfe,0x00,0x00]
 0xff,0xff,0x78,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v0, v254 offset:65535 gds   ; encoding: [0xff,0xff,0x7a,0xd8,0x00,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v0, v254.l offset:65535 gds ; encoding: [0xff,0xff,0x7a,0xd8,0x00,0xfe,0x00,0x00]
 0xff,0xff,0x7a,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1                    ; encoding: [0x00,0x00,0x78,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l                  ; encoding: [0x00,0x00,0x78,0xd8,0xff,0x01,0x00,0x00]
 0x00,0x00,0x78,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1 gds                ; encoding: [0x00,0x00,0x7a,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l gds              ; encoding: [0x00,0x00,0x7a,0xd8,0xff,0x01,0x00,0x00]
 0x00,0x00,0x7a,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1 offset:4660        ; encoding: [0x34,0x12,0x78,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l offset:4660      ; encoding: [0x34,0x12,0x78,0xd8,0xff,0x01,0x00,0x00]
 0x34,0x12,0x78,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1 offset:4660 gds    ; encoding: [0x34,0x12,0x7a,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l offset:4660 gds  ; encoding: [0x34,0x12,0x7a,0xd8,0xff,0x01,0x00,0x00]
 0x34,0x12,0x7a,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1 offset:65535       ; encoding: [0xff,0xff,0x78,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l offset:65535     ; encoding: [0xff,0xff,0x78,0xd8,0xff,0x01,0x00,0x00]
 0xff,0xff,0x78,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v1 offset:65535 gds   ; encoding: [0xff,0xff,0x7a,0xd8,0xff,0x01,0x00,0x00]
+# GFX11: ds_store_b8 v255, v1.l offset:65535 gds ; encoding: [0xff,0xff,0x7a,0xd8,0xff,0x01,0x00,0x00]
 0xff,0xff,0x7a,0xd8,0xff,0x01,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254                  ; encoding: [0x00,0x00,0x78,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l                ; encoding: [0x00,0x00,0x78,0xd8,0xff,0xfe,0x00,0x00]
 0x00,0x00,0x78,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254 gds              ; encoding: [0x00,0x00,0x7a,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l gds            ; encoding: [0x00,0x00,0x7a,0xd8,0xff,0xfe,0x00,0x00]
 0x00,0x00,0x7a,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254 offset:4660      ; encoding: [0x34,0x12,0x78,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l offset:4660    ; encoding: [0x34,0x12,0x78,0xd8,0xff,0xfe,0x00,0x00]
 0x34,0x12,0x78,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254 offset:4660 gds  ; encoding: [0x34,0x12,0x7a,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l offset:4660 gds ; encoding: [0x34,0x12,0x7a,0xd8,0xff,0xfe,0x00,0x00]
 0x34,0x12,0x7a,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254 offset:65535     ; encoding: [0xff,0xff,0x78,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l offset:65535   ; encoding: [0xff,0xff,0x78,0xd8,0xff,0xfe,0x00,0x00]
 0xff,0xff,0x78,0xd8,0xff,0xfe,0x00,0x00
 
-# GFX11: ds_store_b8 v255, v254 offset:65535 gds ; encoding: [0xff,0xff,0x7a,0xd8,0xff,0xfe,0x00,0x00]
+# GFX11: ds_store_b8 v255, v254.l offset:65535 gds ; encoding: [0xff,0xff,0x7a,0xd8,0xff,0xfe,0x00,0x00]
 0xff,0xff,0x7a,0xd8,0xff,0xfe,0x00,0x00
 
 # GFX11: ds_store_b8_d16_hi v1, v2               ; encoding: [0x00,0x00,0x80,0xda,0x01,0x02,0x00,0x00]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_ds.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_ds.txt
index 13440a06032be..6e4b6f051b39d 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_ds.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx1250_dasm_ds.txt
@@ -873,13 +873,13 @@
 # GFX1250: ds_store_b128 v255, v[252:255] offset:4 ; encoding: [0x04,0x00,0x7c,0xdb,0xff,0xfc,0x00,0x00]
 0x04,0x00,0x7c,0xdb,0xff,0xfc,0x00,0x00
 
-# GFX1250: ds_store_b16 v1, v2                     ; encoding: [0x00,0x00,0x7c,0xd8,0x01,0x02,0x00,0x00]
+# GFX1250: ds_store_b16 v1, v2.l                   ; encoding: [0x00,0x00,0x7c,0xd8,0x01,0x02,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0x01,0x02,0x00,0x00
 
-# GFX1250: ds_store_b16 v1, v2 offset:65535        ; encoding: [0xff,0xff,0x7c,0xd8,0x01,0x02,0x00,0x00]
+# GFX1250: ds_store_b16 v1, v2.l offset:65535      ; encoding: [0xff,0xff,0x7c,0xd8,0x01,0x02,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0x01,0x02,0x00,0x00
 
-# GFX1250: ds_store_b16 v255, v255 offset:4        ; encoding: [0x04,0x00,0x7c,0xd8,0xff,0xff,0x00,0x00]
+# GFX1250: ds_store_b16 v255, v255.l offset:4      ; encoding: [0x04,0x00,0x7c,0xd8,0xff,0xff,0x00,0x00]
 0x04,0x00,0x7c,0xd8,0xff,0xff,0x00,0x00
 
 # GFX1250: ds_store_b16_d16_hi v1, v2              ; encoding: [0x00,0x00,0x84,0xda,0x01,0x02,0x00,0x00]
@@ -909,13 +909,13 @@
 # GFX1250: ds_store_b64 v255, v[254:255] offset:4  ; encoding: [0x04,0x00,0x34,0xd9,0xff,0xfe,0x00,0x00]
 0x04,0x00,0x34,0xd9,0xff,0xfe,0x00,0x00
 
-# GFX1250: ds_store_b8 v1, v2                      ; encoding: [0x00,0x00,0x78,0xd8,0x01,0x02,0x00,0x00]
+# GFX1250: ds_store_b8 v1, v2.l                    ; encoding: [0x00,0x00,0x78,0xd8,0x01,0x02,0x00,0x00]
 0x00,0x00,0x78,0xd8,0x01,0x02,0x00,0x00
 
-# GFX1250: ds_store_b8 v1, v2 offset:65535         ; encoding: [0xff,0xff,0x78,0xd8,0x01,0x02,0x00,0x00]
+# GFX1250: ds_store_b8 v1, v2.l offset:65535       ; encoding: [0xff,0xff,0x78,0xd8,0x01,0x02,0x00,0x00]
 0xff,0xff,0x78,0xd8,0x01,0x02,0x00,0x00
 
-# GFX1250: ds_store_b8 v255, v255 offset:4         ; encoding: [0x04,0x00,0x78,0xd8,0xff,0xff,0x00,0x00]
+# GFX1250: ds_store_b8 v255, v255.l offset:4       ; encoding: [0x04,0x00,0x78,0xd8,0xff,0xff,0x00,0x00]
 0x04,0x00,0x78,0xd8,0xff,0xff,0x00,0x00
 
 # GFX1250: ds_store_b8_d16_hi v1, v2               ; encoding: [0x00,0x00,0x80,0xda,0x01,0x02,0x00,0x00]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_ds.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_ds.txt
index d9381b50ca29f..8f75458419202 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_ds.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_ds.txt
@@ -2781,40 +2781,40 @@
 # GFX12: ds_store_b128 v255, v[2:5] offset:65535 ; encoding: [0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00]
 0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00
 
-# GFX12: ds_store_b16 v0, v1                     ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX12: ds_store_b16 v0, v1.l                   ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX12: ds_store_b16 v0, v1 offset:4660         ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX12: ds_store_b16 v0, v1.l offset:4660       ; encoding: [0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0x34,0x12,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX12: ds_store_b16 v0, v1 offset:65535        ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX12: ds_store_b16 v0, v1.l offset:65535      ; encoding: [0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00]
 0xff,0xff,0x7c,0xd8,0x00,0x01,0x00,0x00
 
-# GFX12: ds_store_b16 v0, v254                   ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00]
+# GFX12: ds_store_b16 v0, v254.l                 ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00]
 0x00,0x00,0x7c,0xd8,0x00,0xfe,0x00,0x00
 
-# GFX12: ds...
[truncated]

This avoids an inconsistency in pseudo definitions I ran into for a later patch.

jayfoad · 2025-09-02T09:20:41Z

llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_ds.txt

@@ -4447,76 +4447,76 @@
 # GFX11: ds_store_b128 v255, v[2:5] offset:65535 ; encoding: [0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00]
 0xff,0xff,0x7c,0xdb,0xff,0x02,0x00,0x00

-# GFX11: ds_store_b16 v0, v1                     ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]
+# GFX11: ds_store_b16 v0, v1.l                   ; encoding: [0x00,0x00,0x7c,0xd8,0x00,0x01,0x00,0x00]


Using the .l syntax is not right here. This is not a true16 operand, cos you can't specify v0.h but you can specify v255.l (with an 8-bit instruction field). So it's a normal 32-bit operand but the instruction happens to only use the low 16 bits of data from it.

Yes I think Jay is right, though the docs here are somewhat misleading. The operand is labeled as a vgpr with 16 bits. However, there is no way to encode the swizzle (ie v0.h) in DS. At least I can't find any reference that these would use the hi bit of the operand field the way vop1/2/C does.

arsenm · 2025-09-02T15:40:32Z

Seems #156419 is sufficient to fix the issue

arsenm added the backend:AMDGPU label Sep 2, 2025 — with Graphite App

arsenm requested review from jayfoad, kosarev, mariusz-sikora-at-amd, mbrkusanin, rampitec, shiltian and Sisyph September 2, 2025 06:08

This was referenced Sep 2, 2025

AMDGPU: Fix adding m0 uses to gfx12 ds atomics #156399

Merged

AMDGPU: Fix adding m0 uses to gfx94/gfx12 ds atomics #156402

Merged

arsenm mentioned this pull request Sep 2, 2025

AMDGPU: Reorder arguments of DS_Real_gfx12 (NFC) #156405

Merged

arsenm marked this pull request as ready for review September 2, 2025 06:08

AMDGPU: Handle true16 disassembly of ds_write_b8/b16

5892af7

This avoids an inconsistency in pseudo definitions I ran into for a later patch.

arsenm force-pushed the users/arsenm/amdgpu/true16-handle-ds-write-b8-b16 branch from 6ae597b to 5892af7 Compare September 2, 2025 06:17

This was referenced Sep 2, 2025

AMDGPU: Fix true16 d16 entry table for DS pseudos #156419

Merged

AMDGPU: Add agpr variants of multi-data DS instructions #156420

Merged

jayfoad requested a review from broxigarchen September 2, 2025 09:13

jayfoad reviewed Sep 2, 2025

View reviewed changes

arsenm closed this Sep 2, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

AMDGPU: Handle true16 disassembly of ds_write_b8/b16 #156406

AMDGPU: Handle true16 disassembly of ds_write_b8/b16 #156406

arsenm commented Sep 2, 2025

Uh oh!

arsenm commented Sep 2, 2025 •

edited

Loading

Uh oh!

llvmbot commented Sep 2, 2025

Uh oh!

jayfoad Sep 2, 2025

Uh oh!

Sisyph Sep 2, 2025

Uh oh!

arsenm commented Sep 2, 2025

Uh oh!

Uh oh!

AMDGPU: Handle true16 disassembly of ds_write_b8/b16 #156406

AMDGPU: Handle true16 disassembly of ds_write_b8/b16 #156406

Conversation

arsenm commented Sep 2, 2025

Uh oh!

arsenm commented Sep 2, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Sep 2, 2025

Uh oh!

jayfoad Sep 2, 2025

Choose a reason for hiding this comment

Uh oh!

Sisyph Sep 2, 2025

Choose a reason for hiding this comment

Uh oh!

arsenm commented Sep 2, 2025

Uh oh!

Uh oh!

arsenm commented Sep 2, 2025 •

edited

Loading