[llvm] [AMDGPU] Update strict floating point tests to be more comprehensive (PR #169578)
Chinmay Deshpande via llvm-commits
llvm-commits at lists.llvm.org
Tue Nov 25 18:53:35 PST 2025
https://github.com/chinmaydd updated https://github.com/llvm/llvm-project/pull/169578
>From ae8adc5976c842936083a19ac252f58c08e120ba Mon Sep 17 00:00:00 2001
From: Chinmay Deshpande <ChinmayDiwakar.Deshpande at amd.com>
Date: Tue, 25 Nov 2025 17:13:38 -0500
Subject: [PATCH 1/3] [AMDGPU] Update strict floating point tests to be more
comprehensive
---
llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll | 792 +++++++++++++++++---
llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll | 594 ++++++++++++++-
llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll | 63 +-
llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll | 200 +++++
llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll | 221 ++++++
llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll | 58 +-
llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll | 248 ++++++
llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll | 58 +-
8 files changed, 2102 insertions(+), 132 deletions(-)
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
index a094631267e64..da2c31b374cb7 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
@@ -1,9 +1,21 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-TRUE16 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-FAKE16 %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+
; FIXME: promotion not handled without f16 insts
define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 {
@@ -13,18 +25,71 @@ define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 {
; GCN-NEXT: v_add_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_f16_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -41,18 +106,71 @@ define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 {
; GCN-NEXT: v_add_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_f16_fpexcept_ignore:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -69,18 +187,71 @@ define half @v_constained_fadd_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GCN-NEXT: v_add_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -97,19 +268,71 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_strict(<2 x half> %x, <2 x h
; GFX9-NEXT: v_pk_add_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX8-NEXT: v_or_b32_e32 v0, v0, v2
-; GFX8-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX10PLUS: ; %bb.0:
-; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2
+; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -121,19 +344,71 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h
; GFX9-NEXT: v_pk_add_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX8-NEXT: v_or_b32_e32 v0, v0, v2
-; GFX8-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX10PLUS: ; %bb.0:
-; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2
+; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x half> %val
}
@@ -145,54 +420,176 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x
; GFX9-NEXT: v_pk_add_f16 v0, v0, v1
; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX8-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX8-NEXT: v_or_b32_e32 v0, v0, v2
-; GFX8-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10PLUS-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX10PLUS: ; %bb.0:
-; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10PLUS-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+; GFX8-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v2
+; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v0, v1
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
+; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x half> %val
}
define <3 x half> @v_constained_fadd_v3f16_fpexcept_strict(<3 x half> %x, <3 x half> %y) #0 {
-; GFX9-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_pk_add_f16 v0, v0, v2
-; GFX9-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX9-NEXT: s_setpc_b64 s[30:31]
+; GFX9-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-SDAG-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX9-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX9-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v4
+; GFX8-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-GISEL-NEXT: v_add_f16_e32 v4, v0, v2
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v4, v0
+; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX10-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
;
-; GFX8-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_add_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_e32 v0, v0, v2
-; GFX8-NEXT: v_or_b32_e32 v0, v0, v4
-; GFX8-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX8-NEXT: s_setpc_b64 s[30:31]
+; GFX10-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX10-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX11-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX12-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_pk_add_f16 v0, v0, v2
; GFX10-NEXT: v_add_f16_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-TRUE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-TRUE16-NEXT: v_pk_add_f16 v0, v0, v2
; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-FAKE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -205,29 +602,127 @@ define <3 x half> @v_constained_fadd_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; FIXME: Scalarized
define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x half> %y) #0 {
-; GFX9-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
-; GFX9: ; %bb.0:
-; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX9-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX9-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX9-NEXT: v_add_f16_e32 v0, v0, v2
-; GFX9-NEXT: s_mov_b32 s4, 0x5040100
-; GFX9-NEXT: v_perm_b32 v0, v5, v0, s4
-; GFX9-NEXT: v_perm_b32 v1, v4, v1, s4
-; GFX9-NEXT: s_setpc_b64 s[30:31]
+; GFX9-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX9-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX9-SDAG-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX9-SDAG-NEXT: s_mov_b32 s4, 0x5040100
+; GFX9-SDAG-NEXT: v_perm_b32 v0, v5, v0, s4
+; GFX9-SDAG-NEXT: v_perm_b32 v1, v4, v1, s4
+; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX9-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX8-SDAG-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v0, v5
+; GFX8-SDAG-NEXT: v_or_b32_e32 v1, v1, v4
+; GFX8-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX8-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-GISEL-NEXT: v_add_f16_e32 v4, v0, v2
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v0, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_add_f16_e32 v2, v1, v3
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v1, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v4, v0
+; GFX8-GISEL-NEXT: v_or_b32_e32 v1, v2, v1
+; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX8-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX8-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX8-NEXT: v_add_f16_e32 v0, v0, v2
-; GFX8-NEXT: v_or_b32_e32 v0, v0, v5
-; GFX8-NEXT: v_or_b32_e32 v1, v1, v4
-; GFX8-NEXT: s_setpc_b64 s[30:31]
+; GFX10-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-SDAG-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
+; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX10-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX10-SDAG-NEXT: v_perm_b32 v0, v5, v0, 0x5040100
+; GFX10-SDAG-NEXT: v_perm_b32 v1, v4, v1, 0x5040100
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX10-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.h, v1.h, v3.h
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.h, v0.h, v2.h
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v2.l
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l
+; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3
+; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v2
+; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v0
+; GFX11-SDAG-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v2, v6, v5
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e32 v3, v7, v4
+; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
+; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX11-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX12-SDAG-NEXT: v_add_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v2
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-SDAG-NEXT: v_add_f16_e32 v2, v6, v5
+; GFX12-SDAG-NEXT: v_add_f16_e32 v3, v7, v4
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
+; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -238,7 +733,6 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX10-NEXT: v_perm_b32 v0, v5, v0, 0x5040100
; GFX10-NEXT: v_perm_b32 v1, v4, v1, 0x5040100
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-TRUE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -247,7 +741,6 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v2.l
; GFX11-TRUE16-NEXT: v_add_f16_e32 v1.l, v1.l, v3.l
; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-FAKE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -273,16 +766,57 @@ define amdgpu_ps half @s_constained_fadd_f16_fpexcept_strict(half inreg %x, half
; GCN-NEXT: v_add_f16_e32 v0, s2, v0
; GCN-NEXT: ; return to shader part epilog
;
+; GFX10-SDAG-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: v_add_f16_e64 v0, s2, s3
+; GFX10-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX10-GISEL-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: v_add_f16_e64 v0, s2, s3
+; GFX10-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3
+; GFX11-SDAG-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3
+; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3
+; GFX11-GISEL-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3
+; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_add_f16 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_add_f16 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
; GFX10-LABEL: s_constained_fadd_f16_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add_f16_e64 v0, s2, s3
; GFX10-NEXT: ; return to shader part epilog
-;
; GFX11-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3
; GFX11-TRUE16-NEXT: ; return to shader part epilog
-;
; GFX11-FAKE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
; GFX11-FAKE16: ; %bb.0:
; GFX11-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3
@@ -298,22 +832,64 @@ define amdgpu_ps <2 x half> @s_constained_fadd_v2f16_fpexcept_strict(<2 x half>
; GFX9-NEXT: v_pk_add_f16 v0, s2, v0
; GFX9-NEXT: ; return to shader part epilog
;
-; GFX8-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX8: ; %bb.0:
-; GFX8-NEXT: s_lshr_b32 s0, s3, 16
-; GFX8-NEXT: s_lshr_b32 s1, s2, 16
-; GFX8-NEXT: v_mov_b32_e32 v0, s0
-; GFX8-NEXT: v_mov_b32_e32 v1, s1
-; GFX8-NEXT: v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
-; GFX8-NEXT: v_mov_b32_e32 v1, s3
-; GFX8-NEXT: v_add_f16_e32 v1, s2, v1
-; GFX8-NEXT: v_or_b32_e32 v0, v1, v0
-; GFX8-NEXT: ; return to shader part epilog
-;
-; GFX10PLUS-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX10PLUS: ; %bb.0:
-; GFX10PLUS-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX10PLUS-NEXT: ; return to shader part epilog
+; GFX8-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: s_lshr_b32 s0, s3, 16
+; GFX8-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX8-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX8-SDAG-NEXT: v_mov_b32_e32 v1, s1
+; GFX8-SDAG-NEXT: v_add_f16_sdwa v0, v1, v0 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-SDAG-NEXT: v_mov_b32_e32 v1, s3
+; GFX8-SDAG-NEXT: v_add_f16_e32 v1, s2, v1
+; GFX8-SDAG-NEXT: v_or_b32_e32 v0, v1, v0
+; GFX8-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX8-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_lshr_b32 s0, s2, 16
+; GFX8-GISEL-NEXT: s_lshr_b32 s1, s3, 16
+; GFX8-GISEL-NEXT: v_mov_b32_e32 v0, s3
+; GFX8-GISEL-NEXT: v_mov_b32_e32 v1, s1
+; GFX8-GISEL-NEXT: v_mov_b32_e32 v2, s0
+; GFX8-GISEL-NEXT: v_add_f16_e32 v0, s2, v0
+; GFX8-GISEL-NEXT: v_add_f16_sdwa v1, v2, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v0, v1
+; GFX8-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX10-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX10-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX10-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX10-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-TRUE16-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX11-SDAG-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-FAKE16-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX11-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -326,4 +902,4 @@ declare <4 x half> @llvm.experimental.constrained.fadd.v4f16(<4 x half>, <4 x ha
attributes #0 = { strictfp }
attributes #1 = { inaccessiblememonly nounwind willreturn }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX11: {{.*}}
+; GFX8: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
index 2aecf5fd8753c..1efb2181a844a 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
@@ -1,7 +1,15 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-GISEL %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict:
@@ -15,6 +23,45 @@ define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -31,6 +78,45 @@ define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
@@ -47,6 +133,45 @@ define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
@@ -59,18 +184,83 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GCN-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX11-SDAG: ; %bb.0:
+; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v0, v0, v2
; GFX10-NEXT: v_add_f32_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
@@ -83,18 +273,83 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GCN-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX11-SDAG: ; %bb.0:
+; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v0, v0, v2
; GFX10-NEXT: v_add_f32_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
@@ -107,18 +362,83 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GCN-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX11-SDAG: ; %bb.0:
+; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f32_e32 v0, v0, v2
; GFX10-NEXT: v_add_f32_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
+; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
@@ -132,6 +452,57 @@ define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GCN-NEXT: v_add_f32_e32 v2, v2, v5
; GCN-NEXT: s_setpc_b64 s[30:31]
;
+; GFX10-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_add_f32_e32 v0, v0, v3
+; GFX10-SDAG-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX10-SDAG-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f32_e32 v0, v0, v3
+; GFX10-GISEL-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX10-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX11-SDAG: ; %bb.0:
+; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX11-SDAG-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX11-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX12-SDAG-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX12-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -139,13 +510,36 @@ define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX10-NEXT: v_add_f32_e32 v1, v1, v4
; GFX10-NEXT: v_add_f32_e32 v2, v2, v5
; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
; GFX11-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
; GFX11-NEXT: v_add_f32_e32 v2, v2, v5
; GFX11-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, v0, v3
+; GCN-SDAG-NEXT: v_add_f32_e32 v1, v1, v4
+; GCN-SDAG-NEXT: v_add_f32_e32 v2, v2, v5
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v3
+; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v4
+; GCN-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX12-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
@@ -161,6 +555,36 @@ define amdgpu_ps float @s_constained_fadd_f32_fpexcept_strict(float inreg %x, fl
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: v_add_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_add_f32 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_add_f32 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GCN-SDAG-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: v_mov_b32_e32 v0, s3
+; GCN-SDAG-NEXT: v_add_f32_e32 v0, s2, v0
+; GCN-SDAG-NEXT: ; return to shader part epilog
+; GCN-GISEL-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: v_mov_b32_e32 v0, s3
+; GCN-GISEL-NEXT: v_add_f32_e32 v0, s2, v0
+; GCN-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_add_f32 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -177,6 +601,45 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_add_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fadd.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -194,23 +657,118 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_add_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
define float @v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_sub_f32_e64 v0, v1, |v0|
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GCN-SDAG-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
-; GFX10PLUS: ; %bb.0:
-; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10PLUS-NEXT: v_sub_f32_e64 v0, v1, |v0|
-; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+; GCN-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GCN-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1
+; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX9-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1
+; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX10-SDAG: ; %bb.0:
+; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX10-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX10-GISEL: ; %bb.0:
+; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1
+; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX11-SDAG: ; %bb.0:
+; GFX11-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX11-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, v1, |v0|
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_add_f32_e64 v0, -|v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%neg.fabs.x = fneg float %fabs.x
%val = call float @llvm.experimental.constrained.fadd.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
@@ -223,3 +781,5 @@ declare <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float>, <2 x
declare <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float>, <3 x float>, metadata, metadata)
attributes #0 = { strictfp }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX9: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
index faa0131c88c2d..8cd25eeec2d84 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
@@ -1,7 +1,12 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 {
; GCN-LABEL: v_constained_fadd_f64_fpexcept_strict:
@@ -15,6 +20,12 @@ define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret double %val
}
@@ -31,6 +42,12 @@ define double @v_constained_fadd_f64_fpexcept_ignore(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret double %val
}
@@ -47,6 +64,12 @@ define double @v_constained_fadd_f64_fpexcept_maytrap(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret double %val
}
@@ -65,6 +88,13 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_strict(<2 x double> %x, <2
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x double> %val
}
@@ -83,6 +113,13 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_ignore(<2 x double> %x, <2
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x double> %val
}
@@ -101,6 +138,13 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_maytrap(<2 x double> %x, <
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x double> %val
}
@@ -121,6 +165,14 @@ define <3 x double> @v_constained_fadd_v3f64_fpexcept_strict(<3 x double> %x, <3
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9]
; GFX10-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fadd_v3f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[6:7]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9]
+; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x double> %val
}
@@ -137,6 +189,11 @@ define amdgpu_ps <2 x float> @s_constained_fadd_f64_fpexcept_strict(double inreg
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5]
; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX11-LABEL: s_constained_fadd_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5]
+; GFX11-NEXT: ; return to shader part epilog
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
%cast = bitcast double %val to <2 x float>
ret <2 x float> %cast
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
index eed5f016aa787..bf7b0ce80dab9 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
@@ -13,6 +13,8 @@
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-TRUE16 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
; FIXME: promotion not handled without f16 insts
@@ -52,6 +54,26 @@ define half @v_constained_fmul_f16_fpexcept_strict(half %x, half %y) #0 {
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -92,6 +114,26 @@ define half @v_constained_fmul_f16_fpexcept_ignore(half %x, half %y) #0 {
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret half %val
}
@@ -132,6 +174,26 @@ define half @v_constained_fmul_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret half %val
}
@@ -164,6 +226,26 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_strict(<2 x half> %x, <2 x h
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -196,6 +278,26 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x half> %val
}
@@ -228,6 +330,26 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x half> %val
}
@@ -299,6 +421,28 @@ define <3 x half> @v_constained_fmul_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x half> @llvm.experimental.constrained.fmul.v3f16(<3 x half> %x, <3 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x half> %val
}
@@ -394,6 +538,38 @@ define <4 x half> @v_constained_fmul_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v2
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v2, v6, v5
+; GFX12-SDAG-NEXT: v_mul_f16_e32 v3, v7, v4
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <4 x half> @llvm.experimental.constrained.fmul.v4f16(<4 x half> %x, <4 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <4 x half> %val
}
@@ -429,6 +605,20 @@ define amdgpu_ps half @s_constained_fmul_f16_fpexcept_strict(half inreg %x, half
; GFX1-GISEL-FAKE16: ; %bb.0:
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3
; GFX1-GISEL-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fmul_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_mul_f16 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fmul_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_mul_f16 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -468,6 +658,16 @@ define amdgpu_ps <2 x half> @s_constained_fmul_v2f16_fpexcept_strict(<2 x half>
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fmul_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, s2, s3
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fmul_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, s2, s3
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
index 8df2834928395..5f13eb6ae5a7d 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
@@ -1,11 +1,16 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10 %s
+
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+
define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-LABEL: v_constained_fmul_f32_fpexcept_strict:
; GCN: ; %bb.0:
@@ -18,6 +23,26 @@ define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -34,6 +59,26 @@ define float @v_constained_fmul_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
@@ -50,6 +95,26 @@ define float @v_constained_fmul_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
@@ -74,6 +139,26 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
@@ -98,6 +183,26 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
@@ -122,6 +227,26 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
@@ -149,6 +274,28 @@ define <3 x float> @v_constained_fmul_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
; GFX11-NEXT: v_mul_f32_e32 v2, v2, v5
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_v3f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
+; GFX12-SDAG-NEXT: v_mul_f32_e32 v2, v2, v5
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_v3f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
+; GFX12-GISEL-NEXT: v_mul_f32_e32 v2, v2, v5
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fmul.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
@@ -164,6 +311,20 @@ define amdgpu_ps float @s_constained_fmul_f32_fpexcept_strict(float inreg %x, fl
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fmul_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_mul_f32 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fmul_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_mul_f32 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -180,6 +341,26 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fmul.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -197,6 +378,26 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -214,6 +415,26 @@ define float @v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, -|v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%neg.fabs.x = fneg float %fabs.x
%val = call float @llvm.experimental.constrained.fmul.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll
index 8c98a662c59cc..4d2a93397e0c3 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f64.ll
@@ -1,10 +1,12 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
define double @v_constained_fmul_f64_fpexcept_strict(double %x, double %y) #0 {
; GCN-LABEL: v_constained_fmul_f64_fpexcept_strict:
@@ -18,6 +20,12 @@ define double @v_constained_fmul_f64_fpexcept_strict(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret double %val
}
@@ -34,6 +42,12 @@ define double @v_constained_fmul_f64_fpexcept_ignore(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret double %val
}
@@ -50,6 +64,12 @@ define double @v_constained_fmul_f64_fpexcept_maytrap(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret double %val
}
@@ -68,6 +88,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_strict(<2 x double> %x, <2
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x double> %val
}
@@ -86,6 +113,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_ignore(<2 x double> %x, <2
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x double> %val
}
@@ -104,6 +138,13 @@ define <2 x double> @v_constained_fmul_v2f64_fpexcept_maytrap(<2 x double> %x, <
; GFX10-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_v2f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[4:5]
+; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x double> %val
}
@@ -124,6 +165,14 @@ define <3 x double> @v_constained_fmul_v3f64_fpexcept_strict(<3 x double> %x, <3
; GFX10-NEXT: v_mul_f64 v[2:3], v[2:3], v[8:9]
; GFX10-NEXT: v_mul_f64 v[4:5], v[4:5], v[10:11]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fmul_v3f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_mul_f64 v[0:1], v[0:1], v[6:7]
+; GFX11-NEXT: v_mul_f64 v[2:3], v[2:3], v[8:9]
+; GFX11-NEXT: v_mul_f64 v[4:5], v[4:5], v[10:11]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x double> @llvm.experimental.constrained.fmul.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x double> %val
}
@@ -140,6 +189,11 @@ define amdgpu_ps <2 x float> @s_constained_fmul_f64_fpexcept_strict(double inreg
; GFX10: ; %bb.0:
; GFX10-NEXT: v_mul_f64 v[0:1], s[2:3], s[4:5]
; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX11-LABEL: s_constained_fmul_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_mul_f64 v[0:1], s[2:3], s[4:5]
+; GFX11-NEXT: ; return to shader part epilog
%val = call double @llvm.experimental.constrained.fmul.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
%cast = bitcast double %val to <2 x float>
ret <2 x float> %cast
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
index 6daea572f58c6..e044abb6bf7df 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
@@ -13,6 +13,9 @@
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-TRUE16 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+
; FIXME: promotion not handled without f16 insts
define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 {
@@ -57,6 +60,26 @@ define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -103,6 +126,26 @@ define half @v_constained_fsub_f16_fpexcept_ignore(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret half %val
}
@@ -149,6 +192,26 @@ define half @v_constained_fsub_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret half %val
}
@@ -222,6 +285,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_strict(<2 x half> %x, <2 x h
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_strict:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -309,6 +397,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_ignore:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -396,6 +509,31 @@ define <2 x half> @v_constained_fsub_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x
; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v2, 16, v1
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v3, 16, v0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v3, v2
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1 neg_lo:[0,1] neg_hi:[0,1]
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v2f16_fpexcept_maytrap:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -509,6 +647,40 @@ define <3 x half> @v_constained_fsub_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v2, 16, v0
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v2
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v0
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v2
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v5, v4
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v2
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v2
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v2, v4, v5
+; GFX12-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v3f16_fpexcept_strict:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -654,6 +826,51 @@ define <4 x half> @v_constained_fsub_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v2, 16, v0
; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v1, v3, 16, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v4, 16, v3
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v5, 16, v2
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v6, 16, v0
+; GFX12-SDAG-NEXT: v_lshrrev_b32_e32 v7, 16, v1
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v1, v1, v3
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v2
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v2, v6, v5
+; GFX12-SDAG-NEXT: v_sub_f16_e32 v3, v7, v4
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_perm_b32 v0, v2, v0, 0x5040100
+; GFX12-SDAG-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v4, 16, v0
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v5, 16, v1
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v6, 16, v2
+; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v7, 16, v3
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v2
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v2, v4, v6
+; GFX12-GISEL-NEXT: v_sub_f16_e32 v3, v5, v7
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX12-GISEL-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX12-GISEL-NEXT: v_and_b32_e32 v1, 0xffff, v1
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX12-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v0
+; GFX12-GISEL-NEXT: v_lshl_or_b32 v1, v3, 16, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
; GFX10PLUS-SDAG-LABEL: v_constained_fsub_v4f16_fpexcept_strict:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -724,6 +941,20 @@ define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half
; GFX11-GISEL-FAKE16: ; %bb.0:
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e64 v0, s2, s3
; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fsub_f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_sub_f16 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fsub_f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_sub_f16 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -808,6 +1039,23 @@ define amdgpu_ps <2 x half> @s_constained_fsub_v2f16_fpexcept_strict(<2 x half>
; GFX11-GISEL: ; %bb.0:
; GFX11-GISEL-NEXT: v_pk_add_f16 v0, s2, s3 neg_lo:[0,1] neg_hi:[0,1]
; GFX11-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fsub_v2f16_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_lshr_b32 s0, s3, 16
+; GFX12-SDAG-NEXT: s_lshr_b32 s1, s2, 16
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: s_sub_f16 s0, s1, s0
+; GFX12-SDAG-NEXT: s_sub_f16 s1, s2, s3
+; GFX12-SDAG-NEXT: s_pack_ll_b32_b16 s0, s1, s0
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fsub_v2f16_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: v_pk_add_f16 v0, s2, s3 neg_lo:[0,1] neg_hi:[0,1]
+; GFX12-GISEL-NEXT: ; return to shader part epilog
; GFX10PLUS-SDAG-LABEL: s_constained_fsub_v2f16_fpexcept_strict:
; GFX10PLUS-SDAG: ; %bb.0:
; GFX10PLUS-SDAG-NEXT: v_sub_f16_e64 v0, s2, s3
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll
index e7d136c377079..2937065b7970a 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f64.ll
@@ -1,10 +1,12 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX10 %s
+
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
define double @v_constained_fsub_f64_fpexcept_strict(double %x, double %y) #0 {
; GCN-LABEL: v_constained_fsub_f64_fpexcept_strict:
@@ -18,6 +20,12 @@ define double @v_constained_fsub_f64_fpexcept_strict(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret double %val
}
@@ -34,6 +42,12 @@ define double @v_constained_fsub_f64_fpexcept_ignore(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret double %val
}
@@ -50,6 +64,12 @@ define double @v_constained_fsub_f64_fpexcept_maytrap(double %x, double %y) #0 {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[2:3]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret double %val
}
@@ -68,6 +88,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_strict(<2 x double> %x, <2
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x double> %val
}
@@ -86,6 +113,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_ignore(<2 x double> %x, <2
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x double> %val
}
@@ -104,6 +138,13 @@ define <2 x double> @v_constained_fsub_v2f64_fpexcept_maytrap(<2 x double> %x, <
; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_v2f64_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[4:5]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[6:7]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fsub.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x double> %val
}
@@ -124,6 +165,14 @@ define <3 x double> @v_constained_fsub_v3f64_fpexcept_strict(<3 x double> %x, <3
; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], -v[8:9]
; GFX10-NEXT: v_add_f64 v[4:5], v[4:5], -v[10:11]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-LABEL: v_constained_fsub_v3f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], -v[6:7]
+; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], -v[8:9]
+; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], -v[10:11]
+; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x double> @llvm.experimental.constrained.fsub.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x double> %val
}
@@ -140,6 +189,11 @@ define amdgpu_ps <2 x float> @s_constained_fsub_f64_fpexcept_strict(double inreg
; GFX10: ; %bb.0:
; GFX10-NEXT: v_add_f64 v[0:1], s[2:3], -s[4:5]
; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX11-LABEL: s_constained_fsub_f64_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_add_f64 v[0:1], s[2:3], -s[4:5]
+; GFX11-NEXT: ; return to shader part epilog
%val = call double @llvm.experimental.constrained.fsub.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
%cast = bitcast double %val to <2 x float>
ret <2 x float> %cast
>From 0fe914e08ff4b1fa82967fbf3961c153ae8b75cb Mon Sep 17 00:00:00 2001
From: Chinmay Deshpande <ChinmayDiwakar.Deshpande at amd.com>
Date: Tue, 25 Nov 2025 17:44:16 -0500
Subject: [PATCH 2/3] [AMDGPU] Update fsub.f32 test
---
llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll | 219 ++++++++++++++++++++
1 file changed, 219 insertions(+)
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
index 23dbe21379f7f..5180f6e77e110 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
@@ -8,6 +8,9 @@
; RUN: llc -global-isel= -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+
define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-LABEL: v_constained_fsub_f32_fpexcept_strict:
; GCN: ; %bb.0:
@@ -20,6 +23,26 @@ define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -36,6 +59,26 @@ define float @v_constained_fsub_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
@@ -52,6 +95,26 @@ define float @v_constained_fsub_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
@@ -76,6 +139,26 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
@@ -100,6 +183,26 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_ignore:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_ignore:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
@@ -124,6 +227,26 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
@@ -151,6 +274,28 @@ define <3 x float> @v_constained_fsub_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
; GFX11-NEXT: v_sub_f32_e32 v2, v2, v5
; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_v3f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
+; GFX12-SDAG-NEXT: v_sub_f32_e32 v2, v2, v5
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_v3f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
+; GFX12-GISEL-NEXT: v_sub_f32_e32 v2, v2, v5
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fsub.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
@@ -166,6 +311,20 @@ define amdgpu_ps float @s_constained_fsub_f32_fpexcept_strict(float inreg %x, fl
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
+;
+; GFX12-SDAG-LABEL: s_constained_fsub_f32_fpexcept_strict:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_sub_f32 s0, s2, s3
+; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX12-GISEL-LABEL: s_constained_fsub_f32_fpexcept_strict:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_sub_f32 s0, s2, s3
+; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -182,6 +341,26 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, |v0|, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, |v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fsub.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -199,6 +378,26 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -216,6 +415,26 @@ define float @v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa
; GFX10PLUS-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, -|v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
+; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
+; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
+; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, -|v0|, v1
+; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
+; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
+; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
+; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, -|v0|, v1
+; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%neg.fabs.x = fneg float %fabs.x
%val = call float @llvm.experimental.constrained.fsub.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
>From bb312b4d4ac4e8bf083c0288939314ef26831927 Mon Sep 17 00:00:00 2001
From: Chinmay Deshpande <ChinmayDiwakar.Deshpande at amd.com>
Date: Tue, 25 Nov 2025 21:50:26 -0500
Subject: [PATCH 3/3] [AMDGPU] Update RUN lines to use common prefixes
---
llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll | 539 +++++++-------------
llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll | 482 ++++++-----------
llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll | 143 +-----
llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll | 298 +++++------
llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll | 309 ++++-------
llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll | 181 +++----
llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll | 309 ++++-------
7 files changed, 753 insertions(+), 1508 deletions(-)
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
index da2c31b374cb7..e9e4d5ebed41c 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f16.ll
@@ -1,41 +1,41 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-TRUE16 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-FAKE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-TRUE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
; FIXME: promotion not handled without f16 insts
define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 {
-; GCN-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f16_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX8-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -61,62 +61,37 @@ define half @v_constained_fadd_f16_fpexcept_strict(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX10-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX11-TRUE16: ; %bb.0:
-; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-TRUE16-NEXT: v_add_f16_e32 v0.l, v0.l, v1.l
-; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
-; GFX11-FAKE16-LABEL: v_constained_fadd_f16_fpexcept_strict:
-; GFX11-FAKE16: ; %bb.0:
-; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fadd.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 {
-; GCN-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f16_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX8-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -142,30 +117,15 @@ define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX10-LABEL: v_constained_fadd_f16_fpexcept_ignore:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f16_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_ignore:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -181,23 +141,23 @@ define half @v_constained_fadd_f16_fpexcept_ignore(half %x, half %y) #0 {
}
define half @v_constained_fadd_f16_fpexcept_maytrap(half %x, half %y) #0 {
-; GCN-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f16_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX8-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX8-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX8-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-GISEL-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -223,30 +183,15 @@ define half @v_constained_fadd_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX10-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f16_e32 v0, v0, v1
-; GFX10-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX11-TRUE16-LABEL: v_constained_fadd_f16_fpexcept_maytrap:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -284,55 +229,27 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_strict(<2 x half> %x, <2 x h
; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-SDAG-TRUE16: ; %bb.0:
-; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-SDAG-FAKE16: ; %bb.0:
-; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-GISEL: ; %bb.0:
-; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -360,55 +277,27 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h
; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX11-SDAG-TRUE16: ; %bb.0:
-; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX11-SDAG-FAKE16: ; %bb.0:
-; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX11-GISEL: ; %bb.0:
-; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x half> %val
}
@@ -436,55 +325,27 @@ define <2 x half> @v_constained_fadd_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x
; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-TRUE16-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX11-SDAG-TRUE16: ; %bb.0:
-; GFX11-SDAG-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-SDAG-FAKE16-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX11-SDAG-FAKE16: ; %bb.0:
-; GFX11-SDAG-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX11-GISEL: ; %bb.0:
-; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX11-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX11-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fadd_v2f16_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_add_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x half> %val
}
@@ -578,12 +439,6 @@ define <3 x half> @v_constained_fadd_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX10-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_pk_add_f16 v0, v0, v2
-; GFX10-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX10-NEXT: s_setpc_b64 s[30:31]
; GFX11-TRUE16-LABEL: v_constained_fadd_v3f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -723,16 +578,6 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX12-GISEL-NEXT: v_pk_add_f16 v0, v0, v2
; GFX12-GISEL-NEXT: v_pk_add_f16 v1, v1, v3
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX10-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f16_sdwa v4, v1, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT: v_add_f16_sdwa v5, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1
-; GFX10-NEXT: v_add_f16_e32 v0, v0, v2
-; GFX10-NEXT: v_add_f16_e32 v1, v1, v3
-; GFX10-NEXT: v_perm_b32 v0, v5, v0, 0x5040100
-; GFX10-NEXT: v_perm_b32 v1, v4, v1, 0x5040100
-; GFX10-NEXT: s_setpc_b64 s[30:31]
; GFX11-TRUE16-LABEL: v_constained_fadd_v4f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -760,21 +605,22 @@ define <4 x half> @v_constained_fadd_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
}
define amdgpu_ps half @s_constained_fadd_f16_fpexcept_strict(half inreg %x, half inreg %y) #0 {
-; GCN-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_mov_b32_e32 v0, s3
-; GCN-NEXT: v_add_f16_e32 v0, s2, v0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-NEXT: v_add_f16_e32 v0, s2, v0
+; GFX9-NEXT: ; return to shader part epilog
;
-; GFX10-SDAG-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: v_add_f16_e64 v0, s2, s3
-; GFX10-SDAG-NEXT: ; return to shader part epilog
+; GFX8-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: v_mov_b32_e32 v0, s3
+; GFX8-NEXT: v_add_f16_e32 v0, s2, v0
+; GFX8-NEXT: ; return to shader part epilog
;
-; GFX10-GISEL-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: v_add_f16_e64 v0, s2, s3
-; GFX10-GISEL-NEXT: ; return to shader part epilog
+; GFX10-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: v_add_f16_e64 v0, s2, s3
+; GFX10-NEXT: ; return to shader part epilog
;
; GFX11-SDAG-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -796,23 +642,12 @@ define amdgpu_ps half @s_constained_fadd_f16_fpexcept_strict(half inreg %x, half
; GFX11-GISEL-FAKE16-NEXT: v_add_f16_e64 v0, s2, s3
; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_add_f16 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_add_f16 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
-; GFX10-LABEL: s_constained_fadd_f16_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: v_add_f16_e64 v0, s2, s3
-; GFX10-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fadd_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_add_f16 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
; GFX11-TRUE16-LABEL: s_constained_fadd_f16_fpexcept_strict:
; GFX11-TRUE16: ; %bb.0:
; GFX11-TRUE16-NEXT: v_add_f16_e64 v0.l, s2, s3
@@ -856,40 +691,20 @@ define amdgpu_ps <2 x half> @s_constained_fadd_v2f16_fpexcept_strict(<2 x half>
; GFX8-GISEL-NEXT: v_or_b32_e32 v0, v0, v1
; GFX8-GISEL-NEXT: ; return to shader part epilog
;
-; GFX10-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX10-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX10-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX10-GISEL-NEXT: ; return to shader part epilog
-;
-; GFX11-SDAG-TRUE16-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-SDAG-TRUE16: ; %bb.0:
-; GFX11-SDAG-TRUE16-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX11-SDAG-TRUE16-NEXT: ; return to shader part epilog
-;
-; GFX11-SDAG-FAKE16-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-SDAG-FAKE16: ; %bb.0:
-; GFX11-SDAG-FAKE16-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog
-;
-; GFX11-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX11-GISEL: ; %bb.0:
-; GFX11-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX11-GISEL-NEXT: ; return to shader part epilog
+; GFX10-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX10-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX12-SDAG-NEXT: ; return to shader part epilog
+; GFX11-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX11-NEXT: ; return to shader part epilog
;
-; GFX12-GISEL-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: v_pk_add_f16 v0, s2, s3
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fadd_v2f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_add_f16 v0, s2, s3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.experimental.constrained.fadd.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -901,5 +716,3 @@ declare <4 x half> @llvm.experimental.constrained.fadd.v4f16(<4 x half>, <4 x ha
attributes #0 = { strictfp }
attributes #1 = { inaccessiblememonly nounwind willreturn }
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX8: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
index 1efb2181a844a..a039c2629c395 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f32.ll
@@ -1,6 +1,6 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX10-GISEL %s
@@ -8,15 +8,15 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict:
; GFX10PLUS: ; %bb.0:
@@ -24,25 +24,15 @@ define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -53,25 +43,16 @@ define float @v_constained_fadd_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_ignore:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_ignore:
; GFX10PLUS: ; %bb.0:
@@ -79,25 +60,15 @@ define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_ignore:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -108,25 +79,16 @@ define float @v_constained_fadd_f32_fpexcept_ignore(float %x, float %y) #0 {
; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_f32_fpexcept_ignore:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
; GFX10PLUS: ; %bb.0:
@@ -134,25 +96,15 @@ define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_add_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -163,26 +115,17 @@ define float @v_constained_fadd_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v1
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_f32_fpexcept_maytrap:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_add_f32_e32 v0, v0, v1
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x float> %y) #0 {
-; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v2
-; GCN-NEXT: v_add_f32_e32 v1, v1, v3
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX9-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
; GFX10-SDAG: ; %bb.0:
@@ -210,25 +153,15 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -252,26 +185,17 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_strict:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x float> %y) #0 {
-; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v2
-; GCN-NEXT: v_add_f32_e32 v1, v1, v3
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX9-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
; GFX10-SDAG: ; %bb.0:
@@ -299,25 +223,15 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -341,26 +255,17 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_ignore:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2 x float> %y) #0 {
-; GCN-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v2
-; GCN-NEXT: v_add_f32_e32 v1, v1, v3
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v2
+; GFX9-NEXT: v_add_f32_e32 v1, v1, v3
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
; GFX10-SDAG: ; %bb.0:
@@ -388,25 +293,15 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GFX11-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -430,27 +325,18 @@ define <2 x float> @v_constained_fadd_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GCN-GISEL-NEXT: v_add_f32_e32 v0, v0, v2
; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v3
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_v2f32_fpexcept_maytrap:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_dual_add_f32 v0, v0, v2 :: v_dual_add_f32 v1, v1, v3
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x float> %y) #0 {
-; GCN-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e32 v0, v0, v3
-; GCN-NEXT: v_add_f32_e32 v1, v1, v4
-; GCN-NEXT: v_add_f32_e32 v2, v2, v5
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e32 v0, v0, v3
+; GFX9-NEXT: v_add_f32_e32 v1, v1, v4
+; GFX9-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
; GFX10-SDAG: ; %bb.0:
@@ -482,27 +368,16 @@ define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX11-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
-; GFX12-SDAG-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
-; GFX12-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
+; GFX12-NEXT: v_add_f32_e32 v2, v2, v5
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX10-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -530,45 +405,28 @@ define <3 x float> @v_constained_fadd_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GCN-GISEL-NEXT: v_add_f32_e32 v1, v1, v4
; GCN-GISEL-NEXT: v_add_f32_e32 v2, v2, v5
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_v3f32_fpexcept_strict:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_dual_add_f32 v0, v0, v3 :: v_dual_add_f32 v1, v1, v4
-; GFX12-NEXT: v_add_f32_e32 v2, v2, v5
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
define amdgpu_ps float @s_constained_fadd_f32_fpexcept_strict(float inreg %x, float inreg %y) #0 {
-; GCN-LABEL: s_constained_fadd_f32_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_mov_b32_e32 v0, s3
-; GCN-NEXT: v_add_f32_e32 v0, s2, v0
-; GCN-NEXT: ; return to shader part epilog
+; GFX9-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s3
+; GFX9-NEXT: v_add_f32_e32 v0, s2, v0
+; GFX9-NEXT: ; return to shader part epilog
;
; GFX10PLUS-LABEL: s_constained_fadd_f32_fpexcept_strict:
; GFX10PLUS: ; %bb.0:
; GFX10PLUS-NEXT: v_add_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fadd_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_add_f32 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fadd_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_add_f32 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fadd_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_add_f32 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
; GCN-SDAG-LABEL: s_constained_fadd_f32_fpexcept_strict:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: v_mov_b32_e32 v0, s3
@@ -579,22 +437,16 @@ define amdgpu_ps float @s_constained_fadd_f32_fpexcept_strict(float inreg %x, fl
; GCN-GISEL-NEXT: v_mov_b32_e32 v0, s3
; GCN-GISEL-NEXT: v_add_f32_e32 v0, s2, v0
; GCN-GISEL-NEXT: ; return to shader part epilog
-; GFX12-LABEL: s_constained_fadd_f32_fpexcept_strict:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_add_f32 s0, s2, s3
-; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e64 v0, |v0|, v1
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
; GFX10PLUS: ; %bb.0:
@@ -602,25 +454,15 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: v_add_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f32_e64 v0, |v0|, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f32_e64 v0, |v0|, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e64 v0, |v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -631,26 +473,17 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-GISEL-NEXT: v_add_f32_e64 v0, |v0|, v1
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_lhs:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_add_f32_e64 v0, |v0|, v1
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fadd.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y) #0 {
-; GCN-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
-; GCN: ; %bb.0:
-; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: v_add_f32_e64 v0, v0, |v1|
-; GCN-NEXT: s_setpc_b64 s[30:31]
+; GFX9-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX9-NEXT: s_setpc_b64 s[30:31]
;
; GFX10PLUS-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
; GFX10PLUS: ; %bb.0:
@@ -658,25 +491,15 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: v_add_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_add_f32_e64 v0, v0, |v1|
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_add_f32_e64 v0, v0, |v1|
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_add_f32_e64 v0, v0, |v1|
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GCN-SDAG-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
; GCN-SDAG: ; %bb.0:
; GCN-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
@@ -687,15 +510,6 @@ define float @v_constained_fadd_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GCN-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-GISEL-NEXT: v_add_f32_e64 v0, v0, |v1|
; GCN-GISEL-NEXT: s_setpc_b64 s[30:31]
-; GFX12-LABEL: v_constained_fadd_f32_fpexcept_strict_fabs_rhs:
-; GFX12: ; %bb.0:
-; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-NEXT: s_wait_expcnt 0x0
-; GFX12-NEXT: s_wait_samplecnt 0x0
-; GFX12-NEXT: s_wait_bvhcnt 0x0
-; GFX12-NEXT: s_wait_kmcnt 0x0
-; GFX12-NEXT: v_add_f32_e64 v0, v0, |v1|
-; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fadd.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -781,5 +595,3 @@ declare <2 x float> @llvm.experimental.constrained.fadd.v2f32(<2 x float>, <2 x
declare <3 x float> @llvm.experimental.constrained.fadd.v3f32(<3 x float>, <3 x float>, metadata, metadata)
attributes #0 = { strictfp }
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX9: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
index 8cd25eeec2d84..5469fc8330971 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fadd.f64.ll
@@ -1,12 +1,12 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefix=GCN %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GCN-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GCN-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX10 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX10 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefix=GFX11 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX11 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX10PLUS,GFX11 %s
define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 {
; GCN-LABEL: v_constained_fadd_f64_fpexcept_strict:
@@ -14,18 +14,6 @@ define double @v_constained_fadd_f64_fpexcept_strict(double %x, double %y) #0 {
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_f64_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_f64_fpexcept_strict:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret double %val
}
@@ -36,18 +24,6 @@ define double @v_constained_fadd_f64_fpexcept_ignore(double %x, double %y) #0 {
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_f64_fpexcept_ignore:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_f64_fpexcept_ignore:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret double %val
}
@@ -58,18 +34,6 @@ define double @v_constained_fadd_f64_fpexcept_maytrap(double %x, double %y) #0 {
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_f64_fpexcept_maytrap:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_f64_fpexcept_maytrap:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[2:3]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret double %val
}
@@ -81,20 +45,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_strict(<2 x double> %x, <2
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_strict:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x double> %val
}
@@ -106,20 +56,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_ignore(<2 x double> %x, <2
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_ignore:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_ignore:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x double> %val
}
@@ -131,20 +67,6 @@ define <2 x double> @v_constained_fadd_v2f64_fpexcept_maytrap(<2 x double> %x, <
; GCN-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_v2f64_fpexcept_maytrap:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[4:5]
-; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[6:7]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x double> @llvm.experimental.constrained.fadd.v2f64(<2 x double> %x, <2 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x double> %val
}
@@ -157,43 +79,29 @@ define <3 x double> @v_constained_fadd_v3f64_fpexcept_strict(<3 x double> %x, <3
; GCN-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9]
; GCN-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11]
; GCN-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-LABEL: v_constained_fadd_v3f64_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT: v_add_f64 v[0:1], v[0:1], v[6:7]
-; GFX10-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9]
-; GFX10-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11]
-; GFX10-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX11-LABEL: v_constained_fadd_v3f64_fpexcept_strict:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX11-NEXT: v_add_f64 v[0:1], v[0:1], v[6:7]
-; GFX11-NEXT: v_add_f64 v[2:3], v[2:3], v[8:9]
-; GFX11-NEXT: v_add_f64 v[4:5], v[4:5], v[10:11]
-; GFX11-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double> %x, <3 x double> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x double> %val
}
define amdgpu_ps <2 x float> @s_constained_fadd_f64_fpexcept_strict(double inreg %x, double inreg %y) #0 {
-; GCN-LABEL: s_constained_fadd_f64_fpexcept_strict:
-; GCN: ; %bb.0:
-; GCN-NEXT: v_mov_b32_e32 v0, s4
-; GCN-NEXT: v_mov_b32_e32 v1, s5
-; GCN-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
-; GCN-NEXT: ; return to shader part epilog
-;
-; GFX10-LABEL: s_constained_fadd_f64_fpexcept_strict:
-; GFX10: ; %bb.0:
-; GFX10-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5]
-; GFX10-NEXT: ; return to shader part epilog
-;
-; GFX11-LABEL: s_constained_fadd_f64_fpexcept_strict:
-; GFX11: ; %bb.0:
-; GFX11-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5]
-; GFX11-NEXT: ; return to shader part epilog
+; GCN-SDAG-LABEL: s_constained_fadd_f64_fpexcept_strict:
+; GCN-SDAG: ; %bb.0:
+; GCN-SDAG-NEXT: v_mov_b32_e32 v0, s4
+; GCN-SDAG-NEXT: v_mov_b32_e32 v1, s5
+; GCN-SDAG-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GCN-SDAG-NEXT: ; return to shader part epilog
+;
+; GCN-GISEL-LABEL: s_constained_fadd_f64_fpexcept_strict:
+; GCN-GISEL: ; %bb.0:
+; GCN-GISEL-NEXT: v_mov_b32_e32 v0, s4
+; GCN-GISEL-NEXT: v_mov_b32_e32 v1, s5
+; GCN-GISEL-NEXT: v_add_f64 v[0:1], s[2:3], v[0:1]
+; GCN-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX10PLUS-LABEL: s_constained_fadd_f64_fpexcept_strict:
+; GFX10PLUS: ; %bb.0:
+; GFX10PLUS-NEXT: v_add_f64 v[0:1], s[2:3], s[4:5]
+; GFX10PLUS-NEXT: ; return to shader part epilog
%val = call double @llvm.experimental.constrained.fadd.f64(double %x, double %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
%cast = bitcast double %val to <2 x float>
ret <2 x float> %cast
@@ -205,3 +113,6 @@ declare <3 x double> @llvm.experimental.constrained.fadd.v3f64(<3 x double>, <3
attributes #0 = { strictfp }
attributes #1 = { inaccessiblememonly nounwind willreturn }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX10: {{.*}}
+; GFX11: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
index bf7b0ce80dab9..79154d0db16ec 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f16.ll
@@ -10,11 +10,11 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-TRUE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX1-GISEL,GFX1-GISEL-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
; FIXME: promotion not handled without f16 insts
@@ -43,37 +43,37 @@ define half @v_constained_fmul_f16_fpexcept_strict(half %x, half %y) #0 {
; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_strict:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_strict:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fmul_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_strict:
; GFX1-GISEL-TRUE16: ; %bb.0:
; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_strict:
; GFX1-GISEL-FAKE16: ; %bb.0:
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -103,37 +103,37 @@ define half @v_constained_fmul_f16_fpexcept_ignore(half %x, half %y) #0 {
; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_ignore:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_ignore:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fmul_f16_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_ignore:
; GFX1-GISEL-TRUE16: ; %bb.0:
; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_ignore:
; GFX1-GISEL-FAKE16: ; %bb.0:
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret half %val
}
@@ -163,37 +163,37 @@ define half @v_constained_fmul_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
+; GFX11-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
+; GFX11-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX12-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
; GFX1-GISEL-TRUE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
; GFX1-GISEL-TRUE16: ; %bb.0:
; GFX1-GISEL-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e32 v0.l, v0.l, v1.l
; GFX1-GISEL-TRUE16-NEXT: s_setpc_b64 s[30:31]
-;
; GFX1-GISEL-FAKE16-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
; GFX1-GISEL-FAKE16: ; %bb.0:
; GFX1-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e32 v0, v0, v1
; GFX1-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-SDAG-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f16_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret half %val
}
@@ -227,25 +227,15 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_strict(<2 x half> %x, <2 x h
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
@@ -279,25 +269,15 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_ignore(<2 x half> %x, <2 x h
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x half> %val
}
@@ -331,25 +311,15 @@ define <2 x half> @v_constained_fmul_v2f16_fpexcept_maytrap(<2 x half> %x, <2 x
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f16_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_pk_mul_f16 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x half> %val
}
@@ -415,12 +385,12 @@ define <3 x half> @v_constained_fmul_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e32 v1, v1, v3
; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX1-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
-; GFX1-GISEL: ; %bb.0:
-; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
-; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
-; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX11-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX11-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
; GFX12-SDAG: ; %bb.0:
@@ -443,6 +413,12 @@ define <3 x half> @v_constained_fmul_v3f16_fpexcept_strict(<3 x half> %x, <3 x h
; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX1-GISEL-LABEL: v_constained_fmul_v3f16_fpexcept_strict:
+; GFX1-GISEL: ; %bb.0:
+; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x half> @llvm.experimental.constrained.fmul.v3f16(<3 x half> %x, <3 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x half> %val
}
@@ -532,12 +508,12 @@ define <4 x half> @v_constained_fmul_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX11-SDAG-FAKE16-NEXT: v_perm_b32 v1, v3, v1, 0x5040100
; GFX11-SDAG-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX1-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
-; GFX1-GISEL: ; %bb.0:
-; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
-; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
-; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX11-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
+; GFX11-GISEL: ; %bb.0:
+; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX11-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-SDAG-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
; GFX12-SDAG: ; %bb.0:
@@ -570,6 +546,12 @@ define <4 x half> @v_constained_fmul_v4f16_fpexcept_strict(<4 x half> %x, <4 x h
; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
; GFX12-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX1-GISEL-LABEL: v_constained_fmul_v4f16_fpexcept_strict:
+; GFX1-GISEL: ; %bb.0:
+; GFX1-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX1-GISEL-NEXT: v_pk_mul_f16 v0, v0, v2
+; GFX1-GISEL-NEXT: v_pk_mul_f16 v1, v1, v3
+; GFX1-GISEL-NEXT: s_setpc_b64 s[30:31]
%val = call <4 x half> @llvm.experimental.constrained.fmul.v4f16(<4 x half> %x, <4 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <4 x half> %val
}
@@ -596,29 +578,30 @@ define amdgpu_ps half @s_constained_fmul_f16_fpexcept_strict(half inreg %x, half
; GFX11-SDAG-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3
; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog
;
+; GFX11-GISEL-TRUE16-LABEL: s_constained_fmul_f16_fpexcept_strict:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: v_mul_f16_e64 v0.l, s2, s3
+; GFX11-GISEL-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-FAKE16-LABEL: s_constained_fmul_f16_fpexcept_strict:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3
+; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX12-LABEL: s_constained_fmul_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_mul_f16 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
; GFX1-GISEL-TRUE16-LABEL: s_constained_fmul_f16_fpexcept_strict:
; GFX1-GISEL-TRUE16: ; %bb.0:
; GFX1-GISEL-TRUE16-NEXT: v_mul_f16_e64 v0.l, s2, s3
; GFX1-GISEL-TRUE16-NEXT: ; return to shader part epilog
-;
; GFX1-GISEL-FAKE16-LABEL: s_constained_fmul_f16_fpexcept_strict:
; GFX1-GISEL-FAKE16: ; %bb.0:
; GFX1-GISEL-FAKE16-NEXT: v_mul_f16_e64 v0, s2, s3
; GFX1-GISEL-FAKE16-NEXT: ; return to shader part epilog
-;
-; GFX12-SDAG-LABEL: s_constained_fmul_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_mul_f16 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fmul_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_mul_f16 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
%val = call half @llvm.experimental.constrained.fmul.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -659,15 +642,10 @@ define amdgpu_ps <2 x half> @s_constained_fmul_v2f16_fpexcept_strict(<2 x half>
; GFX10PLUS-NEXT: v_pk_mul_f16 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fmul_v2f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: v_pk_mul_f16 v0, s2, s3
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fmul_v2f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: v_pk_mul_f16 v0, s2, s3
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fmul_v2f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: v_pk_mul_f16 v0, s2, s3
+; GFX12-NEXT: ; return to shader part epilog
%val = call <2 x half> @llvm.experimental.constrained.fmul.v2f16(<2 x half> %x, <2 x half> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x half> %val
}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
index 5f13eb6ae5a7d..4c1df046a6684 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fmul.f32.ll
@@ -8,8 +8,8 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-LABEL: v_constained_fmul_f32_fpexcept_strict:
@@ -24,25 +24,15 @@ define float @v_constained_fmul_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -60,25 +50,15 @@ define float @v_constained_fmul_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
@@ -96,25 +76,15 @@ define float @v_constained_fmul_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_mul_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
@@ -140,25 +110,15 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
@@ -184,25 +144,15 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
@@ -228,25 +178,15 @@ define <2 x float> @v_constained_fmul_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GFX11-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v2f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mul_f32 v0, v0, v2 :: v_dual_mul_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
@@ -275,27 +215,16 @@ define <3 x float> @v_constained_fmul_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX11-NEXT: v_mul_f32_e32 v2, v2, v5
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_v3f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
-; GFX12-SDAG-NEXT: v_mul_f32_e32 v2, v2, v5
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_v3f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
-; GFX12-GISEL-NEXT: v_mul_f32_e32 v2, v2, v5
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_v3f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_mul_f32 v0, v0, v3 :: v_dual_mul_f32 v1, v1, v4
+; GFX12-NEXT: v_mul_f32_e32 v2, v2, v5
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fmul.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
@@ -312,19 +241,12 @@ define amdgpu_ps float @s_constained_fmul_f32_fpexcept_strict(float inreg %x, fl
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fmul_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_mul_f32 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fmul_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_mul_f32 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fmul_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_mul_f32 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -342,25 +264,15 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, |v0|, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, |v0|, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e64 v0, |v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fmul.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -379,25 +291,15 @@ define float @v_constained_fmul_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, v0, |v1|
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, v0, |v1|
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fabs_rhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e64 v0, v0, |v1|
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fmul.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -416,25 +318,15 @@ define float @v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa
; GFX10PLUS-NEXT: v_mul_f32_e64 v0, -|v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_mul_f32_e64 v0, -|v0|, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_mul_f32_e64 v0, -|v0|, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fmul_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_mul_f32_e64 v0, -|v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%neg.fabs.x = fneg float %fabs.x
%val = call float @llvm.experimental.constrained.fmul.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
@@ -447,3 +339,6 @@ declare <2 x float> @llvm.experimental.constrained.fmul.v2f32(<2 x float>, <2 x
declare <3 x float> @llvm.experimental.constrained.fmul.v3f32(<3 x float>, <3 x float>, metadata, metadata)
attributes #0 = { strictfp }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX12-GISEL: {{.*}}
+; GFX12-SDAG: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
index e044abb6bf7df..45cc77486b509 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f16.ll
@@ -5,16 +5,16 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-SDAG %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-TRUE16 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-SDAG-FAKE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-TRUE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11-GISEL,GFX11-GISEL-FAKE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-TRUE16 %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG-FAKE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=+real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -mattr=-real-true16 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
; FIXME: promotion not handled without f16 insts
@@ -25,17 +25,11 @@ define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 {
; GCN-NEXT: v_sub_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fsub_f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_strict:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -61,25 +55,15 @@ define half @v_constained_fsub_f16_fpexcept_strict(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -91,17 +75,11 @@ define half @v_constained_fsub_f16_fpexcept_ignore(half %x, half %y) #0 {
; GCN-NEXT: v_sub_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_ignore:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_ignore:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fsub_f16_fpexcept_ignore:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_ignore:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -127,25 +105,15 @@ define half @v_constained_fsub_f16_fpexcept_ignore(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f16_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret half %val
}
@@ -157,17 +125,11 @@ define half @v_constained_fsub_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GCN-NEXT: v_sub_f16_e32 v0, v0, v1
; GCN-NEXT: s_setpc_b64 s[30:31]
;
-; GFX10-SDAG-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX10-GISEL-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX10-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX10-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-SDAG-TRUE16-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -193,25 +155,15 @@ define half @v_constained_fsub_f16_fpexcept_maytrap(half %x, half %y) #0 {
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e32 v0, v0, v1
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f16_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f16_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f16_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret half %val
}
@@ -912,15 +864,10 @@ define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half
; GCN-NEXT: v_sub_f16_e32 v0, s2, v0
; GCN-NEXT: ; return to shader part epilog
;
-; GFX10-SDAG-LABEL: s_constained_fsub_f16_fpexcept_strict:
-; GFX10-SDAG: ; %bb.0:
-; GFX10-SDAG-NEXT: v_sub_f16_e64 v0, s2, s3
-; GFX10-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX10-GISEL-LABEL: s_constained_fsub_f16_fpexcept_strict:
-; GFX10-GISEL: ; %bb.0:
-; GFX10-GISEL-NEXT: v_sub_f16_e64 v0, s2, s3
-; GFX10-GISEL-NEXT: ; return to shader part epilog
+; GFX10-LABEL: s_constained_fsub_f16_fpexcept_strict:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: v_sub_f16_e64 v0, s2, s3
+; GFX10-NEXT: ; return to shader part epilog
;
; GFX11-SDAG-TRUE16-LABEL: s_constained_fsub_f16_fpexcept_strict:
; GFX11-SDAG-TRUE16: ; %bb.0:
@@ -942,19 +889,12 @@ define amdgpu_ps half @s_constained_fsub_f16_fpexcept_strict(half inreg %x, half
; GFX11-GISEL-FAKE16-NEXT: v_sub_f16_e64 v0, s2, s3
; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fsub_f16_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_sub_f16 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fsub_f16_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_sub_f16 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fsub_f16_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_sub_f16 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call half @llvm.experimental.constrained.fsub.f16(half %x, half %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret half %val
}
@@ -1081,5 +1021,6 @@ declare <4 x half> @llvm.experimental.constrained.fsub.v4f16(<4 x half>, <4 x ha
attributes #0 = { strictfp }
attributes #1 = { inaccessiblememonly nounwind willreturn }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX11: {{.*}}
; GFX8: {{.*}}
; GFX9: {{.*}}
diff --git a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
index 5180f6e77e110..3e6db4d4ac6c8 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_fsub.f32.ll
@@ -8,8 +8,8 @@
; RUN: llc -global-isel= -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GFX10PLUS,GFX11 %s
-; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12-GISEL %s
+; RUN: llc -global-isel=0 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s
+; RUN: llc -global-isel=1 -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s
define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 {
; GCN-LABEL: v_constained_fsub_f32_fpexcept_strict:
@@ -24,25 +24,15 @@ define float @v_constained_fsub_f32_fpexcept_strict(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -60,25 +50,15 @@ define float @v_constained_fsub_f32_fpexcept_ignore(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret float %val
}
@@ -96,25 +76,15 @@ define float @v_constained_fsub_f32_fpexcept_maytrap(float %x, float %y) #0 {
; GFX10PLUS-NEXT: v_sub_f32_e32 v0, v0, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e32 v0, v0, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret float %val
}
@@ -140,25 +110,15 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_strict(<2 x float> %x, <2 x
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <2 x float> %val
}
@@ -184,25 +144,15 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_ignore(<2 x float> %x, <2 x
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_ignore:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_ignore:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_ignore:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.ignore")
ret <2 x float> %val
}
@@ -228,25 +178,15 @@ define <2 x float> @v_constained_fsub_v2f32_fpexcept_maytrap(<2 x float> %x, <2
; GFX11-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_v2f32_fpexcept_maytrap:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_sub_f32 v0, v0, v2 :: v_dual_sub_f32 v1, v1, v3
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float> %x, <2 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.maytrap")
ret <2 x float> %val
}
@@ -275,27 +215,16 @@ define <3 x float> @v_constained_fsub_v3f32_fpexcept_strict(<3 x float> %x, <3 x
; GFX11-NEXT: v_sub_f32_e32 v2, v2, v5
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_v3f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
-; GFX12-SDAG-NEXT: v_sub_f32_e32 v2, v2, v5
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_v3f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
-; GFX12-GISEL-NEXT: v_sub_f32_e32 v2, v2, v5
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_v3f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_dual_sub_f32 v0, v0, v3 :: v_dual_sub_f32 v1, v1, v4
+; GFX12-NEXT: v_sub_f32_e32 v2, v2, v5
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%val = call <3 x float> @llvm.experimental.constrained.fsub.v3f32(<3 x float> %x, <3 x float> %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret <3 x float> %val
}
@@ -312,19 +241,12 @@ define amdgpu_ps float @s_constained_fsub_f32_fpexcept_strict(float inreg %x, fl
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, s2, s3
; GFX10PLUS-NEXT: ; return to shader part epilog
;
-; GFX12-SDAG-LABEL: s_constained_fsub_f32_fpexcept_strict:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_sub_f32 s0, s2, s3
-; GFX12-SDAG-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-SDAG-NEXT: ; return to shader part epilog
-;
-; GFX12-GISEL-LABEL: s_constained_fsub_f32_fpexcept_strict:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_sub_f32 s0, s2, s3
-; GFX12-GISEL-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
-; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s0
-; GFX12-GISEL-NEXT: ; return to shader part epilog
+; GFX12-LABEL: s_constained_fsub_f32_fpexcept_strict:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_sub_f32 s0, s2, s3
+; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_3)
+; GFX12-NEXT: v_mov_b32_e32 v0, s0
+; GFX12-NEXT: ; return to shader part epilog
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
}
@@ -342,25 +264,15 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_lhs(float %x, float %y)
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, |v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, |v0|, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, |v0|, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e64 v0, |v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%val = call float @llvm.experimental.constrained.fsub.f32(float %fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -379,25 +291,15 @@ define float @v_constained_fsub_f32_fpexcept_strict_fabs_rhs(float %x, float %y)
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, v0, |v1|
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, v0, |v1|
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, v0, |v1|
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fabs_rhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e64 v0, v0, |v1|
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.y = call float @llvm.fabs.f32(float %y) #0
%val = call float @llvm.experimental.constrained.fsub.f32(float %x, float %fabs.y, metadata !"round.tonearest", metadata !"fpexcept.strict")
ret float %val
@@ -416,25 +318,15 @@ define float @v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs(float %x, floa
; GFX10PLUS-NEXT: v_sub_f32_e64 v0, -|v0|, v1
; GFX10PLUS-NEXT: s_setpc_b64 s[30:31]
;
-; GFX12-SDAG-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs:
-; GFX12-SDAG: ; %bb.0:
-; GFX12-SDAG-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-SDAG-NEXT: s_wait_expcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_samplecnt 0x0
-; GFX12-SDAG-NEXT: s_wait_bvhcnt 0x0
-; GFX12-SDAG-NEXT: s_wait_kmcnt 0x0
-; GFX12-SDAG-NEXT: v_sub_f32_e64 v0, -|v0|, v1
-; GFX12-SDAG-NEXT: s_setpc_b64 s[30:31]
-;
-; GFX12-GISEL-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs:
-; GFX12-GISEL: ; %bb.0:
-; GFX12-GISEL-NEXT: s_wait_loadcnt_dscnt 0x0
-; GFX12-GISEL-NEXT: s_wait_expcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0
-; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0
-; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0
-; GFX12-GISEL-NEXT: v_sub_f32_e64 v0, -|v0|, v1
-; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31]
+; GFX12-LABEL: v_constained_fsub_f32_fpexcept_strict_fneg_fabs_lhs:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX12-NEXT: s_wait_expcnt 0x0
+; GFX12-NEXT: s_wait_samplecnt 0x0
+; GFX12-NEXT: s_wait_bvhcnt 0x0
+; GFX12-NEXT: s_wait_kmcnt 0x0
+; GFX12-NEXT: v_sub_f32_e64 v0, -|v0|, v1
+; GFX12-NEXT: s_setpc_b64 s[30:31]
%fabs.x = call float @llvm.fabs.f32(float %x) #0
%neg.fabs.x = fneg float %fabs.x
%val = call float @llvm.experimental.constrained.fsub.f32(float %neg.fabs.x, float %y, metadata !"round.tonearest", metadata !"fpexcept.strict")
@@ -447,3 +339,6 @@ declare <2 x float> @llvm.experimental.constrained.fsub.v2f32(<2 x float>, <2 x
declare <3 x float> @llvm.experimental.constrained.fsub.v3f32(<3 x float>, <3 x float>, metadata, metadata)
attributes #0 = { strictfp }
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GFX12-GISEL: {{.*}}
+; GFX12-SDAG: {{.*}}
More information about the llvm-commits
mailing list