[AMDGPU] SIPeepholeSDWA: Disable on existing SDWA instructions #124131

frederik-h · 2025-01-23T15:22:10Z

This PR reapplies the changes from PR #123942 which had to be reverted because of a test failure. The test has been adjusted.

…123942) This is meant as a short-term workaround for an invalid conversion in this pass that occurs because existing SDWA selections are not correctly taken into account during the conversion. See the draft PR llvm#123221 for an attempt to fix the actual issue.

llvmbot · 2025-01-23T15:22:46Z

@llvm/pr-subscribers-llvm-globalisel

@llvm/pr-subscribers-backend-amdgpu

Author: Frederik Harwath (frederik-h)

Changes

This PR reapplies the changes from PR #123942 which had to be reverted because of a test failure. The test has been adjusted.

Patch is 77.66 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/124131.diff

20 Files Affected:

(modified) llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp (+5-2)
(modified) llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll (+10-5)
(modified) llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll (+10-5)
(modified) llvm/test/CodeGen/AMDGPU/GlobalISel/uaddsat.ll (+16-10)
(modified) llvm/test/CodeGen/AMDGPU/GlobalISel/usubsat.ll (+16-10)
(modified) llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll (+21-7)
(modified) llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fadd.ll (+36-12)
(modified) llvm/test/CodeGen/AMDGPU/flat-atomicrmw-fsub.ll (+24-8)
(modified) llvm/test/CodeGen/AMDGPU/global-atomicrmw-fadd.ll (+42-14)
(modified) llvm/test/CodeGen/AMDGPU/global-atomicrmw-fsub.ll (+24-8)
(modified) llvm/test/CodeGen/AMDGPU/idot4u.ll (+12-10)
(modified) llvm/test/CodeGen/AMDGPU/local-atomicrmw-fadd.ll (+12-4)
(modified) llvm/test/CodeGen/AMDGPU/local-atomicrmw-fsub.ll (+12-4)
(modified) llvm/test/CodeGen/AMDGPU/permute_i8.ll (+2-1)
(added) llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel.ll (+87)
(added) llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel.mir (+56)
(modified) llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-gfx10.mir (+2-1)
(modified) llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr.mir (+4-3)
(modified) llvm/test/CodeGen/AMDGPU/sdwa-preserve.mir (+9-6)
(modified) llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll (+4-2)

diff --git a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
index 467f042892cebe..bdd164a2f01312 100644
--- a/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
+++ b/llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp
@@ -962,8 +962,11 @@ bool isConvertibleToSDWA(MachineInstr &MI,
                          const SIInstrInfo* TII) {
   // Check if this is already an SDWA instruction
   unsigned Opc = MI.getOpcode();
-  if (TII->isSDWA(Opc))
-    return true;
+  if (TII->isSDWA(Opc)) {
+    // FIXME: Reenable after fixing selection handling.
+    // Cf. llvm/test/CodeGen/AMDGPU/sdwa-peephole-instr-combine-sel.ll
+    return false;
+  }
 
   // Check if this instruction has opcode that supports SDWA
   if (AMDGPU::getSDWAOp(Opc) == -1)
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
index e289ee759da158..2d9e8969fdbb52 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
@@ -280,8 +280,9 @@ define i16 @v_saddsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX8-NEXT:    v_min_i16_e32 v1, v2, v1
 ; GFX8-NEXT:    v_add_u16_e32 v1, v3, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v2, 0xff
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v1), v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_and_b32_sdwa v0, sext(v0), v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v1), v2 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -299,7 +300,8 @@ define i16 @v_saddsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX9-NEXT:    v_pk_add_i16 v0, v0, v1 clamp
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -439,7 +441,8 @@ define amdgpu_ps i16 @s_saddsat_v2i8(i16 inreg %lhs.arg, i16 inreg %rhs.arg) {
 ; GFX9-NEXT:    v_pk_add_i16 v0, s0, v0 clamp
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -609,9 +612,11 @@ define i32 @v_saddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
 ; GFX8-NEXT:    v_and_b32_sdwa v0, sext(v0), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v2), v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v2), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v3), v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v3), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
index 43ebe156eb2a28..a98b305c15f75c 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
@@ -281,8 +281,9 @@ define i16 @v_ssubsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX8-NEXT:    v_min_i16_e32 v1, v1, v4
 ; GFX8-NEXT:    v_sub_u16_e32 v1, v3, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v2, 0xff
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v1), v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_and_b32_sdwa v0, sext(v0), v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v1), v2 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -300,7 +301,8 @@ define i16 @v_ssubsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX9-NEXT:    v_pk_sub_i16 v0, v0, v1 clamp
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -440,7 +442,8 @@ define amdgpu_ps i16 @s_ssubsat_v2i8(i16 inreg %lhs.arg, i16 inreg %rhs.arg) {
 ; GFX9-NEXT:    v_pk_sub_i16 v0, s0, v0 clamp
 ; GFX9-NEXT:    v_pk_ashrrev_i16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -610,9 +613,11 @@ define i32 @v_ssubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
 ; GFX8-NEXT:    v_and_b32_sdwa v0, sext(v0), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v2), v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v2), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v3), v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, sext(v3), v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/uaddsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/uaddsat.ll
index 788692c94b0cfa..3d7fec9a5986cd 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/uaddsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/uaddsat.ll
@@ -224,7 +224,8 @@ define i16 @v_uaddsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX9-NEXT:    v_pk_add_u16 v0, v0, v1 clamp
 ; GFX9-NEXT:    v_pk_lshrrev_b16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -329,7 +330,8 @@ define amdgpu_ps i16 @s_uaddsat_v2i8(i16 inreg %lhs.arg, i16 inreg %rhs.arg) {
 ; GFX9-NEXT:    v_pk_add_u16 v0, s0, v0 clamp
 ; GFX9-NEXT:    v_pk_lshrrev_b16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -451,9 +453,11 @@ define i32 @v_uaddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
 ; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -618,18 +622,20 @@ define amdgpu_ps i32 @s_uaddsat_v4i8(i32 inreg %lhs.arg, i32 inreg %rhs.arg) {
 ; GFX8-NEXT:    v_mov_b32_e32 v4, 0xff
 ; GFX8-NEXT:    s_lshl_b32 s0, s3, 8
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s1
-; GFX8-NEXT:    s_lshl_b32 s1, s7, 8
 ; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_add_u16_e64 v2, s0, v2 clamp
-; GFX8-NEXT:    s_lshl_b32 s0, s4, 8
-; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s7, 8
 ; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
-; GFX8-NEXT:    v_add_u16_e64 v3, s0, v3 clamp
+; GFX8-NEXT:    s_lshl_b32 s0, s4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_add_u16_e64 v3, s0, v3 clamp
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX8-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/usubsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/usubsat.ll
index 0042d34e235d17..0ab16d95b191d9 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/usubsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/usubsat.ll
@@ -218,7 +218,8 @@ define i16 @v_usubsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
 ; GFX9-NEXT:    v_pk_sub_u16 v0, v0, v1 clamp
 ; GFX9-NEXT:    v_pk_lshrrev_b16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -321,7 +322,8 @@ define amdgpu_ps i16 @s_usubsat_v2i8(i16 inreg %lhs.arg, i16 inreg %rhs.arg) {
 ; GFX9-NEXT:    v_pk_sub_u16 v0, s0, v0 clamp
 ; GFX9-NEXT:    v_pk_lshrrev_b16 v0, 8, v0 op_sel_hi:[0,1]
 ; GFX9-NEXT:    v_mov_b32_e32 v1, 0xff
-; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:BYTE_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_and_b32_sdwa v1, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT:    v_lshlrev_b16_e32 v1, 8, v1
 ; GFX9-NEXT:    v_or_b32_sdwa v0, v0, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
 ; GFX9-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX9-NEXT:    ; return to shader part epilog
@@ -439,9 +441,11 @@ define i32 @v_usubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
 ; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    s_setpc_b64 s[30:31]
 ;
@@ -602,18 +606,20 @@ define amdgpu_ps i32 @s_usubsat_v4i8(i32 inreg %lhs.arg, i32 inreg %rhs.arg) {
 ; GFX8-NEXT:    v_mov_b32_e32 v4, 0xff
 ; GFX8-NEXT:    s_lshl_b32 s0, s3, 8
 ; GFX8-NEXT:    v_mov_b32_e32 v2, s1
-; GFX8-NEXT:    s_lshl_b32 s1, s7, 8
 ; GFX8-NEXT:    v_and_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_sub_u16_e64 v2, s0, v2 clamp
-; GFX8-NEXT:    s_lshl_b32 s0, s4, 8
-; GFX8-NEXT:    v_mov_b32_e32 v3, s1
+; GFX8-NEXT:    s_lshl_b32 s1, s7, 8
 ; GFX8-NEXT:    v_and_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
 ; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 8, v1
-; GFX8-NEXT:    v_sub_u16_e64 v3, s0, v3 clamp
+; GFX8-NEXT:    s_lshl_b32 s0, s4, 8
+; GFX8-NEXT:    v_mov_b32_e32 v3, s1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_sub_u16_e64 v3, s0, v3 clamp
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
-; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:BYTE_3 dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_and_b32_sdwa v1, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_1 src1_sel:DWORD
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 24, v1
 ; GFX8-NEXT:    v_or_b32_e32 v0, v0, v1
 ; GFX8-NEXT:    v_readfirstlane_b32 s0, v0
 ; GFX8-NEXT:    ; return to shader part epilog
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll
index e8f1619c5d418c..a969e3d4f4f79b 100644
--- a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll
+++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointer-atomicrmw-fadd.ll
@@ -6398,8 +6398,10 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fadd_ret_v2f16__offset__amdgpu_no
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v5, v0
-; GFX8-NEXT:    v_add_f16_sdwa v0, v5, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX8-NEXT:    v_add_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v1, v5, v2
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX8-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, v5
@@ -6625,8 +6627,10 @@ define void @buffer_fat_ptr_agent_atomic_fadd_noret_v2f16__offset__amdgpu_no_fin
 ; GFX8-NEXT:  .LBB20_1: ; %atomicrmw.start
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_add_f16_sdwa v1, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX8-NEXT:    v_add_f16_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v4, v2, v0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v1, v4, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v5, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v4, v1
@@ -7044,7 +7048,9 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fadd_ret_v2f16__offset__waterfall
 ; GFX8-NEXT:    ; =>This Loop Header: Depth=1
 ; GFX8-NEXT:    ; Child Loop BB21_4 Depth 2
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_add_f16_sdwa v4, v8, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v4, 16, v8
+; GFX8-NEXT:    v_add_f16_sdwa v4, v4, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; GFX8-NEXT:    v_lshlrev_b32_e32 v4, 16, v4
 ; GFX8-NEXT:    v_add_f16_e32 v6, v8, v5
 ; GFX8-NEXT:    v_or_b32_e32 v7, v6, v4
 ; GFX8-NEXT:    v_mov_b32_e32 v6, v7
@@ -7390,8 +7396,10 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fadd_ret_v2f16__offset(ptr addrsp
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v5, v0
-; GFX8-NEXT:    v_add_f16_sdwa v0, v5, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX8-NEXT:    v_add_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v1, v5, v2
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX8-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, v5
@@ -7650,8 +7658,10 @@ define void @buffer_fat_ptr_agent_atomic_fadd_noret_v2f16__offset(ptr addrspace(
 ; GFX8-NEXT:  .LBB23_1: ; %atomicrmw.start
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_add_f16_sdwa v1, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX8-NEXT:    v_add_f16_sdwa v1, v1, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v4, v2, v0
+; GFX8-NEXT:    v_lshlrev_b32_e32 v1, 16, v1
 ; GFX8-NEXT:    v_or_b32_e32 v1, v4, v1
 ; GFX8-NEXT:    v_mov_b32_e32 v5, v2
 ; GFX8-NEXT:    v_mov_b32_e32 v4, v1
@@ -7915,8 +7925,10 @@ define <2 x half> @buffer_fat_ptr_agent_atomic_fadd_ret_v2f16__offset__amdgpu_no
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
 ; GFX8-NEXT:    v_mov_b32_e32 v5, v0
-; GFX8-NEXT:    v_add_f16_sdwa v0, v5, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v0, 16, v5
+; GFX8-NEXT:    v_add_f16_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; GFX8-NEXT:    v_add_f16_e32 v1, v5, v2
+; GFX8-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
 ; GFX8-NEXT:    v_or_b32_e32 v4, v1, v0
 ; GFX8-NEXT:    v_mov_b32_e32 v0, v4
 ; GFX8-NEXT:    v_mov_b32_e32 v1, v5
@@ -8175,8 +8187,10 @@ define void @buffer_fat_ptr_agent_atomic_fadd_noret_v2f16__offset__amdgpu_no_rem
 ; GFX8-NEXT:  .LBB25_1: ; %atomicrmw.start
 ; GFX8-NEXT:    ; =>This Inner Loop Header: Depth=1
 ; GFX8-NEXT:    s_waitcnt vmcnt(0)
-; GFX8-NEXT:    v_add_f16_sdwa v1, v2, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-NEXT:    v_lshrrev_b32_e32 v1, 16, v2
+; GFX8-NEXT...
[truncated]

…124131) This PR reapplies the changes from PR llvm#123942 which had to be reverted because of a test failure. The test has been adjusted. (cherry picked from commit bfd9bc2)

frederik-h and others added 2 commits January 23, 2025 10:06

[AMDGPU] Adjust test to SIPeepholeSDWA changes

109746e

frederik-h requested a review from arsenm January 23, 2025 15:22

llvmbot added backend:AMDGPU llvm:globalisel labels Jan 23, 2025

arsenm approved these changes Jan 23, 2025

View reviewed changes

frederik-h merged commit bfd9bc2 into llvm:main Jan 24, 2025
11 checks passed

frederik-h deleted the SIPeepholeSDWA-SkipPreexistingSDWA2 branch January 24, 2025 08:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[AMDGPU] SIPeepholeSDWA: Disable on existing SDWA instructions #124131

[AMDGPU] SIPeepholeSDWA: Disable on existing SDWA instructions #124131

frederik-h commented Jan 23, 2025

llvmbot commented Jan 23, 2025 •

edited

Loading

[AMDGPU] SIPeepholeSDWA: Disable on existing SDWA instructions #124131

[AMDGPU] SIPeepholeSDWA: Disable on existing SDWA instructions #124131

Conversation

frederik-h commented Jan 23, 2025

llvmbot commented Jan 23, 2025 • edited Loading

llvmbot commented Jan 23, 2025 •

edited

Loading