[llvm] [AMDGPU][GlobalISel] Add RegBankLegalize support for G_STRICT_FLDEXP (PR #177525)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jan 22 20:56:32 PST 2026
https://github.com/vangthao95 created https://github.com/llvm/llvm-project/pull/177525
None
>From d4a08e27d358b987608caee4f2f0c7d9614aea57 Mon Sep 17 00:00:00 2001
From: Vang Thao <vang.thao at amd.com>
Date: Thu, 22 Jan 2026 20:55:48 -0800
Subject: [PATCH] [AMDGPU][GlobalISel] Add RegBankLegalize support for
G_STRICT_FLDEXP
---
.../AMDGPU/AMDGPURegBankLegalizeRules.cpp | 2 +-
llvm/test/CodeGen/AMDGPU/strict_ldexp.f16.ll | 102 +++++++++++++++---
llvm/test/CodeGen/AMDGPU/strict_ldexp.f32.ll | 35 +++++-
llvm/test/CodeGen/AMDGPU/strict_ldexp.f64.ll | 48 +++++++--
4 files changed, 159 insertions(+), 28 deletions(-)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
index 7441846dc3e34..48a865f730b49 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp
@@ -1050,7 +1050,7 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
.Uni(S32, {{UniInVgprS32}, {Vgpr32, Vgpr32, Vgpr32}})
.Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32, Vgpr32}});
- addRulesForGOpcs({G_FLDEXP}, Standard)
+ addRulesForGOpcs({G_FLDEXP, G_STRICT_FLDEXP}, Standard)
.Uni(S32, {{UniInVgprS32}, {Vgpr32, Vgpr32}})
.Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}})
.Uni(S16, {{UniInVgprS16}, {Vgpr16, Vgpr16}})
diff --git a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f16.ll b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f16.ll
index 98d0a6281054a..fd313a76fc675 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f16.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f16.ll
@@ -6,11 +6,11 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s
-; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s
+; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s
; define half @test_ldexp_f16_i16(ptr addrspace(1) %out, half %a, i16 %b) #0 {
; %result = call half @llvm.experimental.constrained.ldexp.f16.i16(half %a, i16 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
@@ -287,11 +287,12 @@ define <3 x half> @test_ldexp_v3f16_v3i32(ptr addrspace(1) %out, <3 x half> %a,
; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, 0x7fff
; GFX9-GISEL-NEXT: v_med3_i32 v4, v4, v0, v1
; GFX9-GISEL-NEXT: v_med3_i32 v5, v5, v0, v1
+; GFX9-GISEL-NEXT: v_med3_i32 v0, v6, v0, v1
; GFX9-GISEL-NEXT: v_ldexp_f16_e32 v4, v2, v4
; GFX9-GISEL-NEXT: v_ldexp_f16_sdwa v2, v2, v5 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-GISEL-NEXT: v_med3_i32 v0, v6, v0, v1
; GFX9-GISEL-NEXT: v_ldexp_f16_e32 v1, v3, v0
; GFX9-GISEL-NEXT: v_lshl_or_b32 v0, v2, 16, v4
+; GFX9-GISEL-NEXT: v_lshl_or_b32 v1, s4, 16, v1
; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-GISEL-TRUE16-LABEL: test_ldexp_v3f16_v3i32:
@@ -312,18 +313,21 @@ define <3 x half> @test_ldexp_v3f16_v3i32(ptr addrspace(1) %out, <3 x half> %a,
; GFX11-GISEL-FAKE16: ; %bb.0:
; GFX11-GISEL-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-GISEL-FAKE16-NEXT: v_mov_b32_e32 v0, 0x7fff
-; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_3)
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_4)
; GFX11-GISEL-FAKE16-NEXT: v_med3_i32 v1, 0xffff8000, v4, v0
-; GFX11-GISEL-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2
-; GFX11-GISEL-FAKE16-NEXT: v_med3_i32 v5, 0xffff8000, v5, v0
-; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v1, v2, v1
-; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_3)
-; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v2, v4, v5
; GFX11-GISEL-FAKE16-NEXT: v_med3_i32 v4, 0xffff8000, v6, v0
+; GFX11-GISEL-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2
+; GFX11-GISEL-FAKE16-NEXT: v_med3_i32 v0, 0xffff8000, v5, v0
+; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v1, v2, v1
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v2, v3, v4
+; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v0, v6, v0
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v1, 0xffff, v1
-; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v2, 16, v1
-; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e32 v1, v3, v4
+; GFX11-GISEL-FAKE16-NEXT: v_and_b32_e32 v2, 0xffff, v2
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v0, v0, 16, v1
+; GFX11-GISEL-FAKE16-NEXT: v_lshl_or_b32 v1, s0, 16, v2
; GFX11-GISEL-FAKE16-NEXT: s_setpc_b64 s[30:31]
%result = call <3 x half> @llvm.experimental.constrained.ldexp.v3f16.v3i32(<3 x half> %a, <3 x i32> %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
ret <3 x half> %result
@@ -482,6 +486,74 @@ define <4 x half> @test_ldexp_v4f16_v4i32(ptr addrspace(1) %out, <4 x half> %a,
ret <4 x half> %result
}
+define amdgpu_ps half @s_test_ldexp_f16_i32(half inreg %a, i32 inreg %b) #0 {
+; GFX8-SDAG-LABEL: s_test_ldexp_f16_i32:
+; GFX8-SDAG: ; %bb.0:
+; GFX8-SDAG-NEXT: v_mov_b32_e32 v0, 0xffff8000
+; GFX8-SDAG-NEXT: v_mov_b32_e32 v1, 0x7fff
+; GFX8-SDAG-NEXT: v_med3_i32 v0, s1, v0, v1
+; GFX8-SDAG-NEXT: v_ldexp_f16_e32 v0, s0, v0
+; GFX8-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX9-SDAG-LABEL: s_test_ldexp_f16_i32:
+; GFX9-SDAG: ; %bb.0:
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, 0xffff8000
+; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, 0x7fff
+; GFX9-SDAG-NEXT: v_med3_i32 v0, s1, v0, v1
+; GFX9-SDAG-NEXT: v_ldexp_f16_e32 v0, s0, v0
+; GFX9-SDAG-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-TRUE16-LABEL: s_test_ldexp_f16_i32:
+; GFX11-SDAG-TRUE16: ; %bb.0:
+; GFX11-SDAG-TRUE16-NEXT: v_mov_b32_e32 v0, 0x7fff
+; GFX11-SDAG-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-SDAG-TRUE16-NEXT: v_med3_i32 v0, 0xffff8000, s1, v0
+; GFX11-SDAG-TRUE16-NEXT: v_ldexp_f16_e32 v0.l, s0, v0.l
+; GFX11-SDAG-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-SDAG-FAKE16-LABEL: s_test_ldexp_f16_i32:
+; GFX11-SDAG-FAKE16: ; %bb.0:
+; GFX11-SDAG-FAKE16-NEXT: v_mov_b32_e32 v0, 0x7fff
+; GFX11-SDAG-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX11-SDAG-FAKE16-NEXT: v_med3_i32 v0, 0xffff8000, s1, v0
+; GFX11-SDAG-FAKE16-NEXT: v_ldexp_f16_e32 v0, s0, v0
+; GFX11-SDAG-FAKE16-NEXT: ; return to shader part epilog
+;
+; GFX8-GISEL-LABEL: s_test_ldexp_f16_i32:
+; GFX8-GISEL: ; %bb.0:
+; GFX8-GISEL-NEXT: s_max_i32 s1, s1, 0xffff8000
+; GFX8-GISEL-NEXT: s_min_i32 s1, s1, 0x7fff
+; GFX8-GISEL-NEXT: v_mov_b32_e32 v0, s1
+; GFX8-GISEL-NEXT: v_ldexp_f16_e32 v0, s0, v0
+; GFX8-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX9-GISEL-LABEL: s_test_ldexp_f16_i32:
+; GFX9-GISEL: ; %bb.0:
+; GFX9-GISEL-NEXT: s_max_i32 s1, s1, 0xffff8000
+; GFX9-GISEL-NEXT: s_min_i32 s1, s1, 0x7fff
+; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-GISEL-NEXT: v_ldexp_f16_e32 v0, s0, v0
+; GFX9-GISEL-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-TRUE16-LABEL: s_test_ldexp_f16_i32:
+; GFX11-GISEL-TRUE16: ; %bb.0:
+; GFX11-GISEL-TRUE16-NEXT: s_max_i32 s1, s1, 0xffff8000
+; GFX11-GISEL-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-GISEL-TRUE16-NEXT: s_min_i32 s1, s1, 0x7fff
+; GFX11-GISEL-TRUE16-NEXT: v_ldexp_f16_e64 v0.l, s0, s1
+; GFX11-GISEL-TRUE16-NEXT: ; return to shader part epilog
+;
+; GFX11-GISEL-FAKE16-LABEL: s_test_ldexp_f16_i32:
+; GFX11-GISEL-FAKE16: ; %bb.0:
+; GFX11-GISEL-FAKE16-NEXT: s_max_i32 s1, s1, 0xffff8000
+; GFX11-GISEL-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
+; GFX11-GISEL-FAKE16-NEXT: s_min_i32 s1, s1, 0x7fff
+; GFX11-GISEL-FAKE16-NEXT: v_ldexp_f16_e64 v0, s0, s1
+; GFX11-GISEL-FAKE16-NEXT: ; return to shader part epilog
+ %result = call half @llvm.experimental.constrained.ldexp.f16.i32(half %a, i32 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+ ret half %result
+}
+
declare half @llvm.experimental.constrained.ldexp.f16.i16(half, i16, metadata, metadata) #1
declare half @llvm.experimental.constrained.ldexp.f16.i32(half, i32, metadata, metadata) #1
declare <2 x half> @llvm.experimental.constrained.ldexp.v2f16.v2i16(<2 x half>, <2 x i16>, metadata, metadata) #1
diff --git a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f32.ll b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f32.ll
index 456f0bd2434fb..7bf8e8954bd1b 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f32.ll
@@ -4,10 +4,10 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL %s
; define float @test_ldexp_f32_i16(ptr addrspace(1) %out, float %a, i16 %b) #0 {
; %result = call float @llvm.experimental.constrained.ldexp.f32.i16(float %a, i16 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
@@ -236,6 +236,33 @@ define <4 x float> @test_ldexp_v4f32_v4i32(ptr addrspace(1) %out, <4 x float> %a
ret <4 x float> %result
}
+define amdgpu_ps float @s_test_ldexp_f32_i32(float inreg %a, i32 inreg %b) #0 {
+; GFX6-LABEL: s_test_ldexp_f32_i32:
+; GFX6: ; %bb.0:
+; GFX6-NEXT: v_mov_b32_e32 v0, s1
+; GFX6-NEXT: v_ldexp_f32_e32 v0, s0, v0
+; GFX6-NEXT: ; return to shader part epilog
+;
+; GFX8-LABEL: s_test_ldexp_f32_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: v_mov_b32_e32 v0, s1
+; GFX8-NEXT: v_ldexp_f32 v0, s0, v0
+; GFX8-NEXT: ; return to shader part epilog
+;
+; GFX9-LABEL: s_test_ldexp_f32_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s1
+; GFX9-NEXT: v_ldexp_f32 v0, s0, v0
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX11-LABEL: s_test_ldexp_f32_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_ldexp_f32 v0, s0, s1
+; GFX11-NEXT: ; return to shader part epilog
+ %result = call float @llvm.experimental.constrained.ldexp.f32.i32(float %a, i32 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+ ret float %result
+}
+
declare float @llvm.experimental.constrained.ldexp.f32.i16(float, i16, metadata, metadata) #1
declare float @llvm.experimental.constrained.ldexp.f32.i32(float, i32, metadata, metadata) #1
declare <2 x float> @llvm.experimental.constrained.ldexp.v2f32.v2i16(<2 x float>, <2 x i16>, metadata, metadata) #1
diff --git a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f64.ll b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f64.ll
index b31625f8f7073..e72083e4b3569 100644
--- a/llvm/test/CodeGen/AMDGPU/strict_ldexp.f64.ll
+++ b/llvm/test/CodeGen/AMDGPU/strict_ldexp.f64.ll
@@ -4,10 +4,10 @@
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-SDAG %s
; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-SDAG %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
-; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX6,GFX6-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX8,GFX8-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9,GFX9-GISEL %s
+; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11,GFX11-GISEL %s
; define double @test_ldexp_f64_i16(ptr addrspace(1) %out, double %a, i16 %b) #0 {
; %result = call double @llvm.experimental.constrained.ldexp.f64.i16(double %a, i16 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
@@ -65,6 +65,42 @@ define <4 x double> @test_ldexp_v4f64_v4i32(ptr addrspace(1) %out, <4 x double>
ret <4 x double> %result
}
+define amdgpu_ps double @s_test_ldexp_f64_i32(double inreg %a, i32 inreg %b) #0 {
+; GFX6-LABEL: s_test_ldexp_f64_i32:
+; GFX6: ; %bb.0:
+; GFX6-NEXT: v_mov_b32_e32 v0, s2
+; GFX6-NEXT: v_ldexp_f64 v[0:1], s[0:1], v0
+; GFX6-NEXT: v_readfirstlane_b32 s0, v0
+; GFX6-NEXT: v_readfirstlane_b32 s1, v1
+; GFX6-NEXT: ; return to shader part epilog
+;
+; GFX8-LABEL: s_test_ldexp_f64_i32:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: v_mov_b32_e32 v0, s2
+; GFX8-NEXT: v_ldexp_f64 v[0:1], s[0:1], v0
+; GFX8-NEXT: v_readfirstlane_b32 s0, v0
+; GFX8-NEXT: v_readfirstlane_b32 s1, v1
+; GFX8-NEXT: ; return to shader part epilog
+;
+; GFX9-LABEL: s_test_ldexp_f64_i32:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: v_mov_b32_e32 v0, s2
+; GFX9-NEXT: v_ldexp_f64 v[0:1], s[0:1], v0
+; GFX9-NEXT: v_readfirstlane_b32 s0, v0
+; GFX9-NEXT: v_readfirstlane_b32 s1, v1
+; GFX9-NEXT: ; return to shader part epilog
+;
+; GFX11-LABEL: s_test_ldexp_f64_i32:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: v_ldexp_f64 v[0:1], s[0:1], s2
+; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX11-NEXT: v_readfirstlane_b32 s0, v0
+; GFX11-NEXT: v_readfirstlane_b32 s1, v1
+; GFX11-NEXT: ; return to shader part epilog
+ %result = call double @llvm.experimental.constrained.ldexp.f64.i32(double %a, i32 %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+ ret double %result
+}
+
declare double @llvm.experimental.constrained.ldexp.f64.i16(double, i16, metadata, metadata) #1
declare double @llvm.experimental.constrained.ldexp.f64.i32(double, i32, metadata, metadata) #1
declare <2 x double> @llvm.experimental.constrained.ldexp.v2f64.v2i16(<2 x double>, <2 x i16>, metadata, metadata) #1
@@ -75,15 +111,11 @@ declare <4 x double> @llvm.experimental.constrained.ldexp.v4f64.v4i32(<4 x doubl
attributes #0 = { strictfp }
attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: readwrite) }
;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX11: {{.*}}
; GFX11-GISEL: {{.*}}
; GFX11-SDAG: {{.*}}
-; GFX6: {{.*}}
; GFX6-GISEL: {{.*}}
; GFX6-SDAG: {{.*}}
-; GFX8: {{.*}}
; GFX8-GISEL: {{.*}}
; GFX8-SDAG: {{.*}}
-; GFX9: {{.*}}
; GFX9-GISEL: {{.*}}
; GFX9-SDAG: {{.*}}
More information about the llvm-commits
mailing list