[llvm] [AMDGPU] Add additional test cases to integer src mod test (PR #152692)
Chris Jackson via llvm-commits
llvm-commits at lists.llvm.org
Fri Aug 8 04:30:37 PDT 2025
https://github.com/chrisjbris created https://github.com/llvm/llvm-project/pull/152692
Adds missing 16-bit test cases to the test that src mods are not applied to integers in instructions with canonicalizing patterns.
>From 15ece8c62e11fc5933a8009aa4cf3c804d3a89f6 Mon Sep 17 00:00:00 2001
From: Chris Jackson <chris.jackson at amd.com>
Date: Fri, 8 Aug 2025 06:19:21 -0500
Subject: [PATCH] [AMDGPU] Add additional test cases to integer src mod test
Adds missing 16-bit test cases to the test that src mods are not applied
to integers in instructions with canonicalizing patterns.
---
.../integer-canonicalizing-src-modifiers.ll | 154 +++++++++++++++++-
1 file changed, 149 insertions(+), 5 deletions(-)
diff --git a/llvm/test/CodeGen/AMDGPU/integer-canonicalizing-src-modifiers.ll b/llvm/test/CodeGen/AMDGPU/integer-canonicalizing-src-modifiers.ll
index 005c8c8038c16..7b356d26d608a 100644
--- a/llvm/test/CodeGen/AMDGPU/integer-canonicalizing-src-modifiers.ll
+++ b/llvm/test/CodeGen/AMDGPU/integer-canonicalizing-src-modifiers.ll
@@ -86,8 +86,152 @@ define double @s_uitofp_i32_to_f64_neg(i32 inreg %arg0) nounwind {
%cvt = uitofp i32 %arg0.neg to double
ret double %cvt
}
-;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
-; GFX11-FAKE16: {{.*}}
-; GFX11-TRUE16: {{.*}}
-; GFX7: {{.*}}
-; GFX9: {{.*}}
+
+define half @v_uitofp_i16_to_f16_abs(i16 %arg0) nounwind {
+; GFX7-LABEL: v_uitofp_i16_to_f16_abs:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_and_b32_e32 v0, 0x7fff, v0
+; GFX7-NEXT: v_cvt_f32_u32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_uitofp_i16_to_f16_abs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_and_b32_e32 v0, 0x7fff, v0
+; GFX9-NEXT: v_cvt_f16_u16_e32 v0, v0
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-TRUE16-LABEL: v_uitofp_i16_to_f16_abs:
+; GFX11-TRUE16: ; %bb.0:
+; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-TRUE16-NEXT: v_and_b16 v0.l, 0x7fff, v0.l
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.l, v0.l
+; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-FAKE16-LABEL: v_uitofp_i16_to_f16_abs:
+; GFX11-FAKE16: ; %bb.0:
+; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 0x7fff, v0
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_cvt_f16_u16_e32 v0, v0
+; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %arg0.abs = and i16 %arg0, u0x7fff
+ %cvt = uitofp i16 %arg0.abs to half
+ ret half %cvt
+}
+
+define half @v_uitofp_i16_to_f16_neg(i16 %arg0) nounwind {
+; GFX7-LABEL: v_uitofp_i16_to_f16_neg:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: v_and_b32_e32 v0, 0x8000, v0
+; GFX7-NEXT: v_cvt_f32_u32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: v_uitofp_i16_to_f16_neg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: v_and_b32_e32 v0, 0xffff8000, v0
+; GFX9-NEXT: v_cvt_f16_u16_e32 v0, v0
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-TRUE16-LABEL: v_uitofp_i16_to_f16_neg:
+; GFX11-TRUE16: ; %bb.0:
+; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-TRUE16-NEXT: v_and_b16 v0.l, 0x8000, v0.l
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.l, v0.l
+; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-FAKE16-LABEL: v_uitofp_i16_to_f16_neg:
+; GFX11-FAKE16: ; %bb.0:
+; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-FAKE16-NEXT: v_and_b32_e32 v0, 0xffff8000, v0
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX11-FAKE16-NEXT: v_cvt_f16_u16_e32 v0, v0
+; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %arg0.neg = and i16 %arg0, u0x8000
+ %cvt = uitofp i16 %arg0.neg to half
+ ret half %cvt
+}
+
+define half @s_uitofp_i16_to_f16_abs(i16 inreg %arg0) nounwind {
+; GFX7-LABEL: s_uitofp_i16_to_f16_abs:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_and_b32 s4, s16, 0x7fff
+; GFX7-NEXT: v_cvt_f32_u32_e32 v0, s4
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_uitofp_i16_to_f16_abs:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_and_b32 s4, s16, 0x7fff
+; GFX9-NEXT: v_cvt_f16_u16_e32 v0, s4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-TRUE16-LABEL: s_uitofp_i16_to_f16_abs:
+; GFX11-TRUE16: ; %bb.0:
+; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-TRUE16-NEXT: s_and_b32 s0, s0, 0x7fff
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.l, s0
+; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-FAKE16-LABEL: s_uitofp_i16_to_f16_abs:
+; GFX11-FAKE16: ; %bb.0:
+; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-FAKE16-NEXT: s_and_b32 s0, s0, 0x7fff
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-FAKE16-NEXT: v_cvt_f16_u16_e32 v0, s0
+; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %arg0.abs = and i16 %arg0, u0x7fff
+ %cvt = uitofp i16 %arg0.abs to half
+ ret half %cvt
+}
+
+define half @s_uitofp_i16_to_f16_neg(i16 inreg %arg0) nounwind {
+; GFX7-LABEL: s_uitofp_i16_to_f16_neg:
+; GFX7: ; %bb.0:
+; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX7-NEXT: s_and_b32 s4, s16, 0x8000
+; GFX7-NEXT: v_cvt_f32_u32_e32 v0, s4
+; GFX7-NEXT: v_cvt_f16_f32_e32 v0, v0
+; GFX7-NEXT: v_cvt_f32_f16_e32 v0, v0
+; GFX7-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX9-LABEL: s_uitofp_i16_to_f16_neg:
+; GFX9: ; %bb.0:
+; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX9-NEXT: s_and_b32 s4, s16, 0x8000
+; GFX9-NEXT: v_cvt_f16_u16_e32 v0, s4
+; GFX9-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-TRUE16-LABEL: s_uitofp_i16_to_f16_neg:
+; GFX11-TRUE16: ; %bb.0:
+; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-TRUE16-NEXT: s_and_b32 s0, s0, 0x8000
+; GFX11-TRUE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-TRUE16-NEXT: v_cvt_f16_u16_e32 v0.l, s0
+; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX11-FAKE16-LABEL: s_uitofp_i16_to_f16_neg:
+; GFX11-FAKE16: ; %bb.0:
+; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GFX11-FAKE16-NEXT: s_and_b32 s0, s0, 0x8000
+; GFX11-FAKE16-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-FAKE16-NEXT: v_cvt_f16_u16_e32 v0, s0
+; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31]
+ %arg0.neg = and i16 %arg0, u0x8000
+ %cvt = uitofp i16 %arg0.neg to half
+ ret half %cvt
+}
+
More information about the llvm-commits
mailing list