AMDGPU: Remove Formatted MUBUF instructions from gfx1250 support

[changpeng]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

@llvm/pr-subscribers-mc

Author: Changpeng Fang (changpeng)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/145590.diff

4 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPU.td (+3)
• (modified) llvm/lib/Target/AMDGPU/BUFInstructions.td (+18-8)
• (modified) llvm/lib/Target/AMDGPU/GCNSubtarget.h (+2)
• (modified) llvm/test/MC/AMDGPU/gfx1250_asm_unsupported.s (+86)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index 8ddddeb87fbb5..ed0617048863f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -2357,6 +2357,9 @@ def NotLDSRequiresM0Init : Predicate<"!Subtarget->ldsRequiresM0Init()">;
 def HasMTBUFInsts : Predicate<"Subtarget->hasMTBUFInsts()">,
   AssemblerPredicate<(all_of (not FeatureGFX1250Insts))>;
 
+def HasFormattedMUBUFInsts : Predicate<"Subtarget->hasFormattedMUBUFInsts()">,
+  AssemblerPredicate<(all_of (not FeatureGFX1250Insts))>;
+
 def HasExportInsts : Predicate<"Subtarget->hasExportInsts()">,
   AssemblerPredicate<(all_of (not FeatureGFX90AInsts))>;
 
diff --git a/llvm/lib/Target/AMDGPU/BUFInstructions.td b/llvm/lib/Target/AMDGPU/BUFInstructions.td
index 0e8140f464c74..0caabe41e9b79 100644
--- a/llvm/lib/Target/AMDGPU/BUFInstructions.td
+++ b/llvm/lib/Target/AMDGPU/BUFInstructions.td
@@ -845,9 +845,12 @@ multiclass MUBUF_Pseudo_Atomics <string opName,
 // MUBUF Instructions
 //===----------------------------------------------------------------------===//
 
+let OtherPredicates = [HasFormattedMUBUFInsts] in
 defm BUFFER_LOAD_FORMAT_X : MUBUF_Pseudo_Loads_Lds <
   "buffer_load_format_x", f32
 >;
+
+let SubtargetPredicate = HasFormattedMUBUFInsts in {
 defm BUFFER_LOAD_FORMAT_XY : MUBUF_Pseudo_Loads <
   "buffer_load_format_xy", v2f32
 >;
@@ -927,6 +930,7 @@ let TiedSourceNotRead = 1 in {
     "buffer_store_format_d16_xyzw", v4f16
   >;
 } // End OtherPredicates = [HasPackedD16VMem], D16Buf = 1.
+} // End SubtargetPredicate = HasFormattedMUBUFInsts.
 
 defm BUFFER_LOAD_UBYTE : MUBUF_Pseudo_Loads_Lds <
   "buffer_load_ubyte", i32
@@ -1172,6 +1176,7 @@ defm BUFFER_LOAD_SHORT_D16_HI : MUBUF_Pseudo_Loads <
   "buffer_load_short_d16_hi", i32, 1
 >;
 
+let OtherPredicates = [HasFormattedMUBUFInsts] in
 defm BUFFER_LOAD_FORMAT_D16_HI_X : MUBUF_Pseudo_Loads <
   "buffer_load_format_d16_hi_x", i32
 >;
@@ -1185,6 +1190,7 @@ defm BUFFER_STORE_SHORT_D16_HI : MUBUF_Pseudo_Stores <
   "buffer_store_short_d16_hi", i32
 >;
 
+let OtherPredicates = [HasFormattedMUBUFInsts] in
 defm BUFFER_STORE_FORMAT_D16_HI_X : MUBUF_Pseudo_Stores <
   "buffer_store_format_d16_hi_x", i32
 >;
@@ -1366,6 +1372,7 @@ multiclass MUBUF_LoadIntrinsicPat<SDPatternOperator name, ValueType vt,
   defm : MUBUF_LoadIntrinsicPat_Common<name, vt, opcode # "_VBUFFER", memoryVt>;
 }
 
+let OtherPredicates = [HasFormattedMUBUFInsts] in {
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, f32, "BUFFER_LOAD_FORMAT_X">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, i32, "BUFFER_LOAD_FORMAT_X">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format, v2f32, "BUFFER_LOAD_FORMAT_XY">;
@@ -1379,17 +1386,18 @@ defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_tfe, v2i32, "BUFFER_LOAD_FORM
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_tfe, v3i32, "BUFFER_LOAD_FORMAT_XY_TFE">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_tfe, v4i32, "BUFFER_LOAD_FORMAT_XYZ_TFE">;
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_tfe, v5i32, "BUFFER_LOAD_FORMAT_XYZW_TFE">;
+} // End OtherPredicates = [HasFormattedMUBUFInsts].
 
-let OtherPredicates = [HasUnpackedD16VMem] in {
+let OtherPredicates = [HasUnpackedD16VMem, HasFormattedMUBUFInsts] in {
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, f16, "BUFFER_LOAD_FORMAT_D16_X_gfx80">;
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, i16, "BUFFER_LOAD_FORMAT_D16_X_gfx80">;
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, i32, "BUFFER_LOAD_FORMAT_D16_X_gfx80">;
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, v2i32, "BUFFER_LOAD_FORMAT_D16_XY_gfx80">;
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, v3i32, "BUFFER_LOAD_FORMAT_D16_XYZ_gfx80">;
   defm : MUBUF_LoadIntrinsicPat_Common<SIbuffer_load_format_d16, v4i32, "BUFFER_LOAD_FORMAT_D16_XYZW_gfx80">;
-} // End HasUnpackedD16VMem.
+} // End OtherPredicates = [HasUnpackedD16VMem, HasFormattedMUBUFInsts].
 
-let OtherPredicates = [HasPackedD16VMem] in {
+let OtherPredicates = [HasPackedD16VMem, HasFormattedMUBUFInsts] in {
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, f16, "BUFFER_LOAD_FORMAT_D16_X">;
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, i16, "BUFFER_LOAD_FORMAT_D16_X">;
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, i32, "BUFFER_LOAD_FORMAT_D16_X">;
@@ -1399,7 +1407,7 @@ let OtherPredicates = [HasPackedD16VMem] in {
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, v4i16, "BUFFER_LOAD_FORMAT_D16_XYZ", v3i16>;
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, v4f16, "BUFFER_LOAD_FORMAT_D16_XYZW">;
   defm : MUBUF_LoadIntrinsicPat<SIbuffer_load_format_d16, v4i16, "BUFFER_LOAD_FORMAT_D16_XYZW">;
-} // End HasPackedD16VMem.
+} // End OtherPredicates = [HasPackedD16VMem, HasFormattedMUBUFInsts].
 
 foreach vt = Reg32Types.types in {
 defm : MUBUF_LoadIntrinsicPat<SIbuffer_load, vt, "BUFFER_LOAD_DWORD">;
@@ -1475,6 +1483,7 @@ multiclass MUBUF_StoreIntrinsicPat<SDPatternOperator name, ValueType vt,
   defm : MUBUF_StoreIntrinsicPat_Common<name, vt, opcode # "_VBUFFER", memoryVt>;
 }
 
+let OtherPredicates = [HasFormattedMUBUFInsts] in {
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, f32, "BUFFER_STORE_FORMAT_X">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, i32, "BUFFER_STORE_FORMAT_X">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v2f32, "BUFFER_STORE_FORMAT_XY">;
@@ -1484,17 +1493,18 @@ defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v3f32, "BUFFER_STORE_FORMA
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v3i32, "BUFFER_STORE_FORMAT_XYZ">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v4f32, "BUFFER_STORE_FORMAT_XYZW">;
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format, v4i32, "BUFFER_STORE_FORMAT_XYZW">;
+} // End OtherPredicates = [HasFormattedMUBUFInsts].
 
-let OtherPredicates = [HasUnpackedD16VMem] in {
+let OtherPredicates = [HasUnpackedD16VMem, HasFormattedMUBUFInsts] in {
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, f16, "BUFFER_STORE_FORMAT_D16_X_gfx80">;
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, i16, "BUFFER_STORE_FORMAT_D16_X_gfx80">;
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, i32, "BUFFER_STORE_FORMAT_D16_X_gfx80">;
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, v2i32, "BUFFER_STORE_FORMAT_D16_XY_gfx80">;
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, v3i32, "BUFFER_STORE_FORMAT_D16_XYZ_gfx80">;
   defm : MUBUF_StoreIntrinsicPat_Common<SIbuffer_store_format_d16, v4i32, "BUFFER_STORE_FORMAT_D16_XYZW_gfx80">;
-} // End HasUnpackedD16VMem.
+} // End OtherPredicates = [HasUnpackedD16VMem, HasFormattedMUBUFInsts].
 
-let OtherPredicates = [HasPackedD16VMem] in {
+let OtherPredicates = [HasPackedD16VMem, HasFormattedMUBUFInsts] in {
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, f16, "BUFFER_STORE_FORMAT_D16_X">;
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, i16, "BUFFER_STORE_FORMAT_D16_X">;
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, i32, "BUFFER_STORE_FORMAT_D16_X">;
@@ -1504,7 +1514,7 @@ let OtherPredicates = [HasPackedD16VMem] in {
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, v4i16, "BUFFER_STORE_FORMAT_D16_XYZ", v3i16>;
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, v4f16, "BUFFER_STORE_FORMAT_D16_XYZW">;
   defm : MUBUF_StoreIntrinsicPat<SIbuffer_store_format_d16, v4i16, "BUFFER_STORE_FORMAT_D16_XYZW">;
-} // End HasPackedD16VMem.
+} // End OtherPredicates = [HasPackedD16VMem, HasFormattedMUBUFInsts].
 
 foreach vt = Reg32Types.types in {
 defm : MUBUF_StoreIntrinsicPat<SIbuffer_store, vt, "BUFFER_STORE_DWORD">;
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index 3ca5e35bcf065..eaac9b0cacf2d 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -691,6 +691,8 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
 
   bool hasMTBUFInsts() const { return !hasGFX1250Insts(); }
 
+  bool hasFormattedMUBUFInsts() const { return !hasGFX1250Insts(); }
+
   bool hasExportInsts() const {
     return !hasGFX940Insts();
   }
diff --git a/llvm/test/MC/AMDGPU/gfx1250_asm_unsupported.s b/llvm/test/MC/AMDGPU/gfx1250_asm_unsupported.s
index 7b94a04c48923..bccd2d55e1c29 100644
--- a/llvm/test/MC/AMDGPU/gfx1250_asm_unsupported.s
+++ b/llvm/test/MC/AMDGPU/gfx1250_asm_unsupported.s
@@ -1,5 +1,7 @@
 ; RUN: not llvm-mc -triple=amdgcn -mcpu=gfx1250 -show-encoding %s 2>&1 | FileCheck --check-prefix=GFX1250-ERR --implicit-check-not=error: --strict-whitespace %s
 
+;; All "TBUFFER" ops, and BUFFER_LOAD/STORE_FORMAT ops.
+
 tbuffer_load_d16_format_x v4, off, s[8:11], s3 format:[BUF_FMT_8_UNORM] offset:8388607
 // GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
 
@@ -47,3 +49,87 @@ tbuffer_store_format_xyz v[4:6], off, s[8:11], s3 format:[BUF_FMT_32_FLOAT] offs
 
 tbuffer_store_format_xyzw v[4:7], off, s[8:11], s3 format:[BUF_FMT_2_10_10_10_SNORM] offset:8388607
 // GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_d16_format_x v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_d16_format_xy v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_d16_format_xyz v[5:6], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_d16_format_xyzw v[5:6], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_d16_hi_format_x v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_x v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_xy v[5:6], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_xyz v[5:7], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_xyzw v[5:8], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_d16_format_x v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_d16_format_xy v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_d16_format_xyz v[1:2], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_d16_format_xyzw v[1:2], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_d16_hi_format_x v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_x v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_xy v[1:2], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_xyz v[1:3], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_xyzw v[1:4], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_d16_x v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_d16_xy v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_d16_xyz v[5:6], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_d16_xyzw v[5:6], off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_load_format_d16_hi_x v5, off, s[8:11], s3 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_d16_x v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_d16_xy v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_d16_xyz v[1:2], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_d16_xyzw v[1:2], off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
+
+buffer_store_format_d16_hi_x v1, off, s[12:15], s4 offset:8388607
+// GFX1250-ERR: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU