Skip to content

[AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12 #77927

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

Jump to bottom
Merged
merged 4 commits into from
Jan 17, 2024
Merged

[AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12 #77927

merged 4 commits into from
Jan 17, 2024

Conversation

jayfoad
Copy link
Contributor

@jayfoad jayfoad commented Jan 12, 2024

  • Precommit extra GFX12 test coverage
  • [AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12

@llvmbot
Copy link
Member

llvmbot commented Jan 12, 2024 •
edited
Loading

@llvm/pr-subscribers-llvm-globalisel

@llvm/pr-subscribers-backend-amdgpu

Author: Jay Foad (jayfoad)

Changes
  • Precommit extra GFX12 test coverage
  • [AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12

Patch is 46.03 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/77927.diff

9 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/AMDGPU.td (-1)
  • (modified) llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-mad_64_32.mir (+21)
  • (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll (+7-8)
  • (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll (+18-22)
  • (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll (+7-8)
  • (modified) llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll (+7-8)
  • (modified) llvm/test/CodeGen/AMDGPU/llvm.mulo.ll (+163)
  • (modified) llvm/test/CodeGen/AMDGPU/mad_64_32.ll (+191)
  • (modified) llvm/test/CodeGen/AMDGPU/mul.ll (+22-21)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index b27edb1e9e14bb..023a4260d76a37 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1501,7 +1501,6 @@ def FeatureISAVersion12 : FeatureSet<
    FeaturePseudoScalarTrans,
    FeatureHasRestrictedSOffset,
    FeatureVGPRSingleUseHintInsts,
-   FeatureMADIntraFwdBug,
    FeatureScalarDwordx3Loads]>;
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-mad_64_32.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-mad_64_32.mir
index 698281caca245e..59f6114ca5cd3a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-mad_64_32.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-mad_64_32.mir
@@ -1,6 +1,7 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
 # RUN: llc -march=amdgcn -mcpu=gfx1030 -run-pass=instruction-select -global-isel-abort=2 -pass-remarks-missed='gisel*' -verify-machineinstrs %s -o - 2>%t | FileCheck -check-prefix=GFX10 %s
 # RUN: llc -march=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -global-isel-abort=2 -pass-remarks-missed='gisel*' -verify-machineinstrs %s -o - 2>%t | FileCheck -check-prefix=GFX11 %s
+# RUN: llc -march=amdgcn -mcpu=gfx1200 -run-pass=instruction-select -global-isel-abort=2 -pass-remarks-missed='gisel*' -verify-machineinstrs %s -o - 2>%t | FileCheck -check-prefix=GFX12 %s
 
 ---
 name: mad_u64_u32_vvv
@@ -18,6 +19,7 @@ body: |
     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
     ; GFX10-NEXT: [[V_MAD_U64_U32_e64_:%[0-9]+]]:vreg_64, [[V_MAD_U64_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_U64_U32_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
     ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAD_U64_U32_e64_]], implicit [[V_MAD_U64_U32_e64_1]]
+    ;
     ; GFX11-LABEL: name: mad_u64_u32_vvv
     ; GFX11: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
     ; GFX11-NEXT: {{  $}}
@@ -26,6 +28,15 @@ body: |
     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
     ; GFX11-NEXT: [[V_MAD_U64_U32_gfx11_e64_:%[0-9]+]]:vreg_64, [[V_MAD_U64_U32_gfx11_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_U64_U32_gfx11_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
     ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAD_U64_U32_gfx11_e64_]], implicit [[V_MAD_U64_U32_gfx11_e64_1]]
+    ;
+    ; GFX12-LABEL: name: mad_u64_u32_vvv
+    ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; GFX12-NEXT: {{  $}}
+    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
+    ; GFX12-NEXT: [[V_MAD_U64_U32_e64_:%[0-9]+]]:vreg_64, [[V_MAD_U64_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_U64_U32_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
+    ; GFX12-NEXT: S_ENDPGM 0, implicit [[V_MAD_U64_U32_e64_]], implicit [[V_MAD_U64_U32_e64_1]]
     %0:vgpr(s32) = COPY $vgpr0
     %1:vgpr(s32) = COPY $vgpr1
     %2:vgpr(s32) = COPY $vgpr2
@@ -51,6 +62,7 @@ body: |
     ; GFX10-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
     ; GFX10-NEXT: [[V_MAD_I64_I32_e64_:%[0-9]+]]:vreg_64, [[V_MAD_I64_I32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_I64_I32_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
     ; GFX10-NEXT: S_ENDPGM 0, implicit [[V_MAD_I64_I32_e64_]], implicit [[V_MAD_I64_I32_e64_1]]
+    ;
     ; GFX11-LABEL: name: mad_i64_i32_vvv
     ; GFX11: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
     ; GFX11-NEXT: {{  $}}
@@ -59,6 +71,15 @@ body: |
     ; GFX11-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
     ; GFX11-NEXT: [[V_MAD_I64_I32_gfx11_e64_:%[0-9]+]]:vreg_64, [[V_MAD_I64_I32_gfx11_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_I64_I32_gfx11_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
     ; GFX11-NEXT: S_ENDPGM 0, implicit [[V_MAD_I64_I32_gfx11_e64_]], implicit [[V_MAD_I64_I32_gfx11_e64_1]]
+    ;
+    ; GFX12-LABEL: name: mad_i64_i32_vvv
+    ; GFX12: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; GFX12-NEXT: {{  $}}
+    ; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GFX12-NEXT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr3
+    ; GFX12-NEXT: [[V_MAD_I64_I32_e64_:%[0-9]+]]:vreg_64, [[V_MAD_I64_I32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_MAD_I64_I32_e64 [[COPY]], [[COPY1]], [[COPY2]], 0, implicit $exec
+    ; GFX12-NEXT: S_ENDPGM 0, implicit [[V_MAD_I64_I32_e64_]], implicit [[V_MAD_I64_I32_e64_1]]
     %0:vgpr(s32) = COPY $vgpr0
     %1:vgpr(s32) = COPY $vgpr1
     %2:vgpr(s32) = COPY $vgpr2
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
index e3d2ecefbda30d..b63a50dc4e6e7a 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_buffer.ll
@@ -520,13 +520,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W64-NEXT:  .LBB1_2:
 ; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W64-NEXT:    v_mul_lo_u32 v0, s6, v0
 ; GFX12W64-NEXT:    s_waitcnt vmcnt(0)
 ; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12W64-NEXT:    v_mad_co_u64_u32 v[0:1], null, s6, v0, s[2:3]
 ; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
-; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX12W64-NEXT:    v_add_nc_u32_e32 v0, s2, v0
 ; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W64-NEXT:    s_nop 0
 ; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -553,12 +552,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W32-NEXT:  .LBB1_2:
 ; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
 ; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W32-NEXT:    v_mul_lo_u32 v0, s2, v0
 ; GFX12W32-NEXT:    s_waitcnt vmcnt(0)
-; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W32-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12W32-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_add_nc_u32 v0, s2, v0
+; GFX12W32-NEXT:    v_mad_co_u64_u32 v[0:1], null, s2, v0, s[4:5]
+; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W32-NEXT:    s_nop 0
 ; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
index b4c8da44337ae5..42bef4faf8a49b 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_global_pointer.ll
@@ -572,13 +572,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1264-NEXT:    global_inv scope:SCOPE_DEV
 ; GFX1264-NEXT:  .LBB1_2:
 ; GFX1264-NEXT:    s_or_b64 exec, exec, s[0:1]
-; GFX1264-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1264-NEXT:    v_mul_lo_u32 v0, s8, v0
 ; GFX1264-NEXT:    v_readfirstlane_b32 s0, v1
+; GFX1264-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1264-NEXT:    s_mov_b32 s6, -1
 ; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1264-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX1264-NEXT:    v_mad_co_u64_u32 v[0:1], null, s8, v0, s[0:1]
 ; GFX1264-NEXT:    buffer_store_b32 v0, off, s[4:7], null
 ; GFX1264-NEXT:    s_nop 0
 ; GFX1264-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -610,13 +609,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1232-NEXT:    global_inv scope:SCOPE_DEV
 ; GFX1232-NEXT:  .LBB1_2:
 ; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s1
+; GFX1232-NEXT:    v_readfirstlane_b32 s2, v1
 ; GFX1232-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1232-NEXT:    v_mul_lo_u32 v0, s0, v0
-; GFX1232-NEXT:    v_readfirstlane_b32 s0, v1
 ; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1232-NEXT:    s_mov_b32 s6, -1
 ; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1232-NEXT:    v_add_nc_u32_e32 v0, s0, v0
+; GFX1232-NEXT:    v_mad_co_u64_u32 v[0:1], null, s0, v0, s[2:3]
 ; GFX1232-NEXT:    buffer_store_b32 v0, off, s[4:7], null
 ; GFX1232-NEXT:    s_nop 0
 ; GFX1232-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -1671,12 +1669,11 @@ define amdgpu_kernel void @add_i64_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1264-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX1264-NEXT:    v_readfirstlane_b32 s3, v1
 ; GFX1264-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1264-NEXT:    v_mul_lo_u32 v3, s1, v2
 ; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1264-NEXT:    s_mov_b32 s6, -1
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1264-NEXT:    v_mad_co_u64_u32 v[0:1], null, s0, v2, s[2:3]
-; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1264-NEXT:    v_add_nc_u32_e32 v1, v3, v1
+; GFX1264-NEXT:    v_mad_co_u64_u32 v[1:2], null, s1, v2, v[1:2]
 ; GFX1264-NEXT:    buffer_store_b64 v[0:1], off, s[4:7], null
 ; GFX1264-NEXT:    s_nop 0
 ; GFX1264-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -1712,12 +1709,11 @@ define amdgpu_kernel void @add_i64_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1232-NEXT:    v_readfirstlane_b32 s2, v0
 ; GFX1232-NEXT:    v_readfirstlane_b32 s3, v1
 ; GFX1232-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1232-NEXT:    v_mul_lo_u32 v3, s1, v2
 ; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1232-NEXT:    s_mov_b32 s6, -1
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
 ; GFX1232-NEXT:    v_mad_co_u64_u32 v[0:1], null, s0, v2, s[2:3]
-; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX1232-NEXT:    v_add_nc_u32_e32 v1, v3, v1
+; GFX1232-NEXT:    v_mad_co_u64_u32 v[1:2], null, s1, v2, v[1:2]
 ; GFX1232-NEXT:    buffer_store_b64 v[0:1], off, s[4:7], null
 ; GFX1232-NEXT:    s_nop 0
 ; GFX1232-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -3608,16 +3604,16 @@ define amdgpu_kernel void @sub_i64_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1264-NEXT:  .LBB10_2:
 ; GFX1264-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX1264-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1264-NEXT:    v_mul_lo_u32 v5, s1, v2
 ; GFX1264-NEXT:    v_mad_co_u64_u32 v[3:4], null, s0, v2, 0
 ; GFX1264-NEXT:    v_readfirstlane_b32 s0, v0
-; GFX1264-NEXT:    v_readfirstlane_b32 s1, v1
 ; GFX1264-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1264-NEXT:    s_mov_b32 s6, -1
-; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1264-NEXT:    v_add_nc_u32_e32 v1, v4, v5
+; GFX1264-NEXT:    s_wait_alu 0xfff
+; GFX1264-NEXT:    v_mad_co_u64_u32 v[4:5], null, s1, v2, v[4:5]
+; GFX1264-NEXT:    v_readfirstlane_b32 s1, v1
 ; GFX1264-NEXT:    v_sub_co_u32 v0, vcc, s0, v3
-; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1264-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1264-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1264-NEXT:    v_sub_co_ci_u32_e32 v1, vcc, s1, v1, vcc
 ; GFX1264-NEXT:    buffer_store_b64 v[0:1], off, s[4:7], null
 ; GFX1264-NEXT:    s_nop 0
@@ -3652,16 +3648,16 @@ define amdgpu_kernel void @sub_i64_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX1232-NEXT:  .LBB10_2:
 ; GFX1232-NEXT:    s_or_b32 exec_lo, exec_lo, s8
 ; GFX1232-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX1232-NEXT:    v_mul_lo_u32 v5, s1, v2
 ; GFX1232-NEXT:    v_mad_co_u64_u32 v[3:4], null, s0, v2, 0
 ; GFX1232-NEXT:    v_readfirstlane_b32 s0, v0
-; GFX1232-NEXT:    v_readfirstlane_b32 s1, v1
 ; GFX1232-NEXT:    s_mov_b32 s7, 0x31016000
 ; GFX1232-NEXT:    s_mov_b32 s6, -1
-; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_4)
-; GFX1232-NEXT:    v_add_nc_u32_e32 v1, v4, v5
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_4)
+; GFX1232-NEXT:    v_mad_co_u64_u32 v[4:5], null, s1, v2, v[4:5]
+; GFX1232-NEXT:    v_readfirstlane_b32 s1, v1
 ; GFX1232-NEXT:    v_sub_co_u32 v0, vcc_lo, s0, v3
-; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_2)
+; GFX1232-NEXT:    s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1232-NEXT:    v_mov_b32_e32 v1, v4
 ; GFX1232-NEXT:    v_sub_co_ci_u32_e32 v1, vcc_lo, s1, v1, vcc_lo
 ; GFX1232-NEXT:    buffer_store_b64 v[0:1], off, s[4:7], null
 ; GFX1232-NEXT:    s_nop 0
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll
index 79f8b3a1d5d84c..280b3c13f410db 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_raw_buffer.ll
@@ -519,13 +519,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W64-NEXT:  .LBB1_2:
 ; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W64-NEXT:    v_mul_lo_u32 v0, s6, v0
 ; GFX12W64-NEXT:    s_waitcnt vmcnt(0)
 ; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12W64-NEXT:    v_mad_co_u64_u32 v[0:1], null, s6, v0, s[2:3]
 ; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
-; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX12W64-NEXT:    v_add_nc_u32_e32 v0, s2, v0
 ; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W64-NEXT:    s_nop 0
 ; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -552,12 +551,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W32-NEXT:  .LBB1_2:
 ; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
 ; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W32-NEXT:    v_mul_lo_u32 v0, s2, v0
 ; GFX12W32-NEXT:    s_waitcnt vmcnt(0)
-; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W32-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12W32-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_add_nc_u32 v0, s2, v0
+; GFX12W32-NEXT:    v_mad_co_u64_u32 v[0:1], null, s2, v0, s[4:5]
+; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W32-NEXT:    s_nop 0
 ; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll
index edf6fbadf1a60a..f2eea9d2218417 100644
--- a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll
+++ b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_struct_buffer.ll
@@ -535,13 +535,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W64-NEXT:  .LBB1_2:
 ; GFX12W64-NEXT:    s_or_b64 exec, exec, s[2:3]
 ; GFX12W64-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W64-NEXT:    v_mul_lo_u32 v0, s6, v0
 ; GFX12W64-NEXT:    s_waitcnt vmcnt(0)
 ; GFX12W64-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W64-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GFX12W64-NEXT:    v_mad_co_u64_u32 v[0:1], null, s6, v0, s[2:3]
 ; GFX12W64-NEXT:    v_mov_b32_e32 v1, 0
-; GFX12W64-NEXT:    s_delay_alu instid0(VALU_DEP_2)
-; GFX12W64-NEXT:    v_add_nc_u32_e32 v0, s2, v0
 ; GFX12W64-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W64-NEXT:    s_nop 0
 ; GFX12W64-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
@@ -568,12 +567,12 @@ define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, ptr addrspace(
 ; GFX12W32-NEXT:  .LBB1_2:
 ; GFX12W32-NEXT:    s_or_b32 exec_lo, exec_lo, s3
 ; GFX12W32-NEXT:    s_load_b64 s[0:1], s[0:1], 0x24
-; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
-; GFX12W32-NEXT:    v_mul_lo_u32 v0, s2, v0
 ; GFX12W32-NEXT:    s_waitcnt vmcnt(0)
-; GFX12W32-NEXT:    v_readfirstlane_b32 s2, v1
+; GFX12W32-NEXT:    v_readfirstlane_b32 s4, v1
+; GFX12W32-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX12W32-NEXT:    s_delay_alu instid0(VALU_DEP_1)
-; GFX12W32-NEXT:    v_dual_mov_b32 v1, 0 :: v_dual_add_nc_u32 v0, s2, v0
+; GFX12W32-NEXT:    v_mad_co_u64_u32 v[0:1], null, s2, v0, s[4:5]
+; GFX12W32-NEXT:    v_mov_b32_e32 v1, 0
 ; GFX12W32-NEXT:    global_store_b32 v1, v0, s[0:1]
 ; GFX12W32-NEXT:    s_nop 0
 ; GFX12W32-NEXT:    s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.mulo.ll b/llvm/test/CodeGen/AMDGPU/llvm.mulo.ll
index 249acec639540b..9afe3e0f97551f 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.mulo.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.mulo.ll
@@ -3,6 +3,7 @@
 ; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX9 %s
 ; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX10 %s
 ; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX11 %s
+; RUN: llc -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck --check-prefix=GFX12 %s
 
 define { i64, i1 } @umulo_i64_v_v(i64 %x, i64 %y) {
 ; SI-LABEL: umulo_i64_v_v:
@@ -97,6 +98,32 @@ define { i64, i1 } @umulo_i64_v_v(i64 %x, i64 %y) {
 ; GFX11-NEXT:    v_cmp_ne_u64_e32 vcc_lo, 0, v[2:3]
 ; GFX11-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc_lo
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: umulo_i64_v_v:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX12-NEXT:    v_dual_mov_b32 v4, v0 :: v_dual_mov_b32 v5, v1
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX12-NEXT:    v_mad_co_u64_u32 v[0:1], null, v4, v2, 0
+; GFX12-NEXT:    v_mad_co_u64_u32 v[6:7], null, v4, v3, 0
+; GFX12-NEXT:    v_mad_co_u64_u32 v[8:9], null, v5, v2, 0
+; GFX12-NEXT:    v_mad_co_u64_u32 v[2:3], null, v5, v3, 0
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-NEXT:    v_mov_b32_e32 v4, v1
+; GFX12-NEXT:    v_add3_u32 v1, v1, v6, v8
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-NEXT:    v_add_co_u32 v4, vcc_lo, v4, v6
+; GFX12-NEXT:    v_add_co_ci_u32_e32 v5, vcc_lo, 0, v7, vcc_lo
+; GFX12-NEXT:    v_add_co_u32 v4, vcc_lo, v4, v8
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-NEXT:    v_add_co_ci_u32_e32 v4, vcc_lo, v5, v9, vcc_lo
+; GFX12-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX12-NEXT:    v_add_co_u32 v2, vcc_lo, v4, v2
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-NEXT:    v_add_co_ci_u32_e32 v3, vcc_lo, 0, v3, vcc_lo
+; GFX12-NEXT:    v_cmp_ne_u64_e32 vcc_lo, 0, v[2:3]
+; GFX12-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc_lo
+; GFX12-NEXT:    s_setpc_b64 s[30:31]
 bb:
   %umulo = tail call { i64, i1 } @llvm.umul.with.overflow.i64(i64 %x, i64 %y)
   ret { i64, i1 } %umulo
@@ -248,6 +275,47 @@ define { i64, i1 } @smulo_i64_v_v(i64 %x, i64 %y) {
 ; GFX11-NEXT:    v_cmp_ne_u64_e32 vcc_lo, v[4:5], v[2:3]
 ; GFX11-NEXT:    v_cndmask_b32_e64 v2, 0, 1, vcc_lo
 ; GFX11-NEXT:    s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: smulo_i64_v_v:
+; GFX12:       ; %bb.0: ; %bb
+; GFX12-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX12-NEXT:    v_dual_mov_b32 v4, v0 :: v_dual_mov_b32 v5, v1
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_3)
+; GFX12-NEXT:    v_mad_co_u64_u32 v[0:1], null, v4, v2, 0
+; GFX12-NEXT:    v_mad_co_u64_u32 v[6:7], null, v4, v3, 0
+; GFX12-NEXT:    v_mad_co_u64_u32 v[8:9], null, v5, v2, 0
+; GFX12-NEXT:    v_mad_co_i64_i32 v[10:11], null, v5, v3, 0
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-NEXT:    v_mov_b32_e32 v12, v1
+; GFX12-NEXT:    v_add3_u32 v1, v1, v6, v8
+; GFX12-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-NEXT:    v_add_co_u32 v12, vcc_lo, v12, v6
+; GFX12-NEXT:    v_add_co_ci_u32_e32 v7, vcc_lo, 0, v7, vcc_lo
+; GFX12-NEXT:    v_add_co_u...
[truncated]

@jayfoad jayfoad changed the title gfx12 mad intra fwd [AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12 Jan 12, 2024
@jayfoad jayfoad merged commit e9e9d1b into llvm:main Jan 17, 2024
@jayfoad jayfoad deleted the gfx12-mad-intra-fwd branch January 17, 2024 11:52
ampandey-1995 pushed a commit to ampandey-1995/llvm-project that referenced this pull request Jan 19, 2024
@jayfoad @ampandey-1995
[AMDGPU] Disable V_MAD_U64_U32/V_MAD_I64_I32 workaround for GFX12 (ll…
00e2b90 
…vm#77927)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm arsenm approved these changes

Assignees
No one assigned
Labels
backend:AMDGPU llvm:globalisel
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@jayfoad @llvmbot @arsenm