AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123

arsenm · 2025-02-21T04:11:12Z

No description provided.

arsenm · 2025-02-21T04:11:30Z

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123 👈 (View in Graphite)
AMDGPU: Widen f16 minimum/maximum to v2f16 on gfx950 #128121
main

This stack of pull requests is managed by Graphite. Learn more about stacking.

llvmbot · 2025-02-21T04:11:45Z

@llvm/pr-subscribers-backend-amdgpu

Author: Matt Arsenault (arsenm)

Changes

Patch is 47.60 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/128123.diff

4 Files Affected:

(modified) llvm/lib/Target/AMDGPU/SIISelLowering.cpp (+2-1)
(modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+2-2)
(modified) llvm/test/CodeGen/AMDGPU/fmaximum3.ll (+54-121)
(modified) llvm/test/CodeGen/AMDGPU/fminimum3.ll (+54-121)

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index 0b13a53a0c989..6ed09253c51e1 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -13515,7 +13515,8 @@ static bool supportsMin3Max3(const GCNSubtarget &Subtarget, unsigned Opc,
   case ISD::FMINIMUM:
   case ISD::FMAXIMUM:
     return (VT == MVT::f32 && Subtarget.hasMinimum3Maximum3F32()) ||
-           (VT == MVT::f16 && Subtarget.hasMinimum3Maximum3F16());
+           (VT == MVT::f16 && Subtarget.hasMinimum3Maximum3F16()) ||
+           (VT == MVT::v2f16 && Subtarget.hasMinimum3Maximum3PKF16());
   case ISD::SMAX:
   case ISD::SMIN:
   case ISD::UMAX:
diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index d5c6e8af109f4..85c047167f1e1 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -145,8 +145,8 @@ def : VOP3PSatPat<ssubsat, V_PK_SUB_I16>;
 } // End SubtargetPredicate = HasVOP3PInsts
 
 let SubtargetPredicate = HasMinimum3Maximum3PKF16, FPDPRounding = 1 in {
-defm V_PK_MINIMUM3_F16 : VOP3PInst<"v_pk_minimum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
-defm V_PK_MAXIMUM3_F16 : VOP3PInst<"v_pk_maximum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>>;
+defm V_PK_MINIMUM3_F16 : VOP3PInst<"v_pk_minimum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, AMDGPUfminimum3>;
+defm V_PK_MAXIMUM3_F16 : VOP3PInst<"v_pk_maximum3_f16", VOP3P_Profile<VOP_V2F16_V2F16_V2F16_V2F16>, AMDGPUfmaximum3>;
 }
 
 // TODO: Make sure we're doing the right thing with denormals. Note
diff --git a/llvm/test/CodeGen/AMDGPU/fmaximum3.ll b/llvm/test/CodeGen/AMDGPU/fmaximum3.ll
index f228824ff750e..2a372dffce650 100644
--- a/llvm/test/CodeGen/AMDGPU/fmaximum3.ll
+++ b/llvm/test/CodeGen/AMDGPU/fmaximum3.ll
@@ -1269,9 +1269,7 @@ define half @v_fmaximum3_f16(half %a, half %b, half %c) {
 ; GFX950-LABEL: v_fmaximum3_f16:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
   %max1 = call half @llvm.maximum.f16(half %max0, half %c)
@@ -1306,9 +1304,7 @@ define half @v_fmaximum3_f16_commute(half %a, half %b, half %c) {
 ; GFX950-LABEL: v_fmaximum3_f16_commute:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v2, v0, v0
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v2, v0, v1
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
   %max1 = call half @llvm.maximum.f16(half %c, half %max0)
@@ -1346,10 +1342,9 @@ define amdgpu_ps i32 @s_fmaximum3_f16(half inreg %a, half inreg %b, half inreg %
 ;
 ; GFX950-LABEL: s_fmaximum3_f16:
 ; GFX950:       ; %bb.0:
-; GFX950-NEXT:    v_mov_b32_e32 v0, s0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s1, s1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s2, s2
+; GFX950-NEXT:    v_mov_b32_e32 v0, s1
+; GFX950-NEXT:    v_mov_b32_e32 v1, s2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, s0, v0, v1
 ; GFX950-NEXT:    s_nop 0
 ; GFX950-NEXT:    v_and_b32_e32 v0, 0xffff, v0
 ; GFX950-NEXT:    s_nop 0
@@ -1392,9 +1387,7 @@ define half @v_fmaximum3_f16_fabs0(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fabs = call half @llvm.fabs.f16(half %a)
   %max0 = call half @llvm.maximum.f16(half %a.fabs, half %b)
@@ -1431,9 +1424,7 @@ define half @v_fmaximum3_f16_fabs1(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_and_b32_e32 v1, 0x7fff, v1
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %b.fabs = call half @llvm.fabs.f16(half %b)
   %max0 = call half @llvm.maximum.f16(half %a, half %b.fabs)
@@ -1470,9 +1461,7 @@ define half @v_fmaximum3_f16_fabs2(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_and_b32_e32 v2, 0x7fff, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %c.fabs = call half @llvm.fabs.f16(half %c)
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
@@ -1511,9 +1500,7 @@ define half @v_fmaximum3_f16_fabs_all(half %a, half %b, half %c) {
 ; GFX950-NEXT:    v_and_b32_e32 v0, 0x7fff, v0
 ; GFX950-NEXT:    v_and_b32_e32 v1, 0x7fff, v1
 ; GFX950-NEXT:    v_and_b32_e32 v2, 0x7fff, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fabs = call half @llvm.fabs.f16(half %a)
   %b.fabs = call half @llvm.fabs.f16(half %b)
@@ -1554,9 +1541,7 @@ define half @v_fmaximum3_f16_fneg_all(half %a, half %b, half %c) {
 ; GFX950-NEXT:    v_xor_b32_e32 v0, 0x8000, v0
 ; GFX950-NEXT:    v_xor_b32_e32 v1, 0x8000, v1
 ; GFX950-NEXT:    v_xor_b32_e32 v2, 0x8000, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fneg = fneg half %a
   %b.fneg = fneg half %b
@@ -1597,9 +1582,7 @@ define half @v_fmaximum3_f16_fneg_fabs_all(half %a, half %b, half %c) {
 ; GFX950-NEXT:    v_or_b32_e32 v0, 0x8000, v0
 ; GFX950-NEXT:    v_or_b32_e32 v1, 0x8000, v1
 ; GFX950-NEXT:    v_or_b32_e32 v2, 0x8000, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fabs = call half @llvm.fabs.f16(half %a)
   %b.fabs = call half @llvm.fabs.f16(half %b)
@@ -1641,9 +1624,7 @@ define half @v_fmaximum3_f16_fneg0(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_xor_b32_e32 v0, 0x8000, v0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fneg = fneg half %a
   %max0 = call half @llvm.maximum.f16(half %a.fneg, half %b)
@@ -1680,9 +1661,7 @@ define half @v_fmaximum3_f16_fneg1(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_xor_b32_e32 v1, 0x8000, v1
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %b.fneg = fneg half %b
   %max0 = call half @llvm.maximum.f16(half %a, half %b.fneg)
@@ -1719,9 +1698,7 @@ define half @v_fmaximum3_f16_fneg2(half %a, half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    v_xor_b32_e32 v2, 0x8000, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %c.fneg = fneg half %c
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
@@ -1758,9 +1735,7 @@ define half @v_fmaximum3_f16_const0(half %b, half %c) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    s_movk_i32 s0, 0x4800
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, s0
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, v1
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half 8.0, half %b)
   %max1 = call half @llvm.maximum.f16(half %max0, half %c)
@@ -1795,9 +1770,8 @@ define half @v_fmaximum3_f16__const2(half %a, half %b) {
 ; GFX950-LABEL: v_fmaximum3_f16__const2:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
 ; GFX950-NEXT:    s_movk_i32 s0, 0x4800
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, s0
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, s0
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
   %max1 = call half @llvm.maximum.f16(half %max0, half 8.0)
@@ -1832,9 +1806,7 @@ define half @v_fmaximum3_f16_inlineimm0(half %b, half %c) {
 ; GFX950-LABEL: v_fmaximum3_f16_inlineimm0:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 4.0, 4.0
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 4.0, v1
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half 4.0, half %b)
   %max1 = call half @llvm.maximum.f16(half %max0, half %c)
@@ -1869,9 +1841,7 @@ define half @v_fmaximum3_f16__inlineimm(half %a, half %b) {
 ; GFX950-LABEL: v_fmaximum3_f16__inlineimm:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 4.0, 4.0
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, 4.0
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half %a, half %b)
   %max1 = call half @llvm.maximum.f16(half %max0, half 4.0)
@@ -1909,9 +1879,8 @@ define half @v_fmaximum3_f16_const1_const2(half %a) {
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX950-NEXT:    s_movk_i32 s0, 0x4800
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, s0
-; GFX950-NEXT:    s_movk_i32 s0, 0x4c00
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, s0
+; GFX950-NEXT:    v_mov_b32_e32 v1, 0x4c00
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, s0, v1
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call half @llvm.maximum.f16(half %a, half 8.0)
   %max1 = call half @llvm.maximum.f16(half %max0, half 16.0)
@@ -1959,9 +1928,7 @@ define <2 x half> @v_fmaximum3_v2f16(<2 x half> %a, <2 x half> %b, <2 x half> %c
 ; GFX950-LABEL: v_fmaximum3_v2f16:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v2, v0, v0
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v2, v0, v1
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)
   %max1 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %c, <2 x half> %max0)
@@ -2009,9 +1976,7 @@ define <2 x half> @v_fmaximum3_v2f16_commute(<2 x half> %a, <2 x half> %b, <2 x
 ; GFX950-LABEL: v_fmaximum3_v2f16_commute:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)
   %max1 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %max0, <2 x half> %c)
@@ -2067,9 +2032,7 @@ define <2 x half> @v_fmaximum3_v2f16__fabs_all(<2 x half> %a, <2 x half> %b, <2
 ; GFX950-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
 ; GFX950-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
 ; GFX950-NEXT:    v_and_b32_e32 v2, 0x7fff7fff, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %a)
   %b.fabs = call <2 x half> @llvm.fabs.v2f16(<2 x half> %b)
@@ -2120,9 +2083,7 @@ define <2 x half> @v_fmaximum3_v2f16__fneg_all(<2 x half> %a, <2 x half> %b, <2
 ; GFX950-LABEL: v_fmaximum3_v2f16__fneg_all:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1 neg_lo:[1,1,1] neg_hi:[1,1,1]
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2 neg_lo:[0,1,1] neg_hi:[0,1,1]
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v2 neg_lo:[1,1,1] neg_hi:[1,1,1]
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fneg = fneg <2 x half> %a
   %b.fneg = fneg <2 x half> %b
@@ -2173,9 +2134,7 @@ define <2 x half> @v_fmaximum3_v2f16__inlineimm1(<2 x half> %a, <2 x half> %c) {
 ; GFX950-LABEL: v_fmaximum3_v2f16__inlineimm1:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 2.0, 2.0 op_sel_hi:[1,0,0]
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 2.0, v1 op_sel_hi:[1,0,1]
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> <half 2.0, half 2.0>)
   %max1 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %max0, <2 x half> %c)
@@ -2223,9 +2182,7 @@ define <2 x half> @v_fmaximum3_v2f16__inlineimm2(<2 x half> %a, <2 x half> %b) {
 ; GFX950-LABEL: v_fmaximum3_v2f16__inlineimm2:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, v1
-; GFX950-NEXT:    s_nop 0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 4.0, 4.0 op_sel_hi:[1,0,0]
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v1, 4.0 op_sel_hi:[1,1,0]
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %a, <2 x half> %b)
   %max1 = call <2 x half> @llvm.maximum.v2f16(<2 x half> %max0, <2 x half> <half 4.0, half 4.0>)
@@ -2287,10 +2244,8 @@ define <3 x half> @v_fmaximum3_v3f16(<3 x half> %a, <3 x half> %b, <3 x half> %c
 ; GFX950-LABEL: v_fmaximum3_v3f16:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v5, v1, v1
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v4, v0, v0
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v4, v0, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v5, v1, v3
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %a, <3 x half> %b)
   %max1 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %c, <3 x half> %max0)
@@ -2352,10 +2307,8 @@ define <3 x half> @v_fmaximum3_v3f16_commute(<3 x half> %a, <3 x half> %b, <3 x
 ; GFX950-LABEL: v_fmaximum3_v3f16_commute:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v5, v5
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v4, v4
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v4
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v5
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %a, <3 x half> %b)
   %max1 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %max0, <3 x half> %c)
@@ -2428,16 +2381,14 @@ define <3 x half> @v_fmaximum3_v3f16__fabs_all(<3 x half> %a, <3 x half> %b, <3
 ; GFX950-LABEL: v_fmaximum3_v3f16__fabs_all:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
 ; GFX950-NEXT:    v_and_b32_e32 v1, 0x7fff7fff, v1
-; GFX950-NEXT:    v_and_b32_e32 v2, 0x7fff7fff, v2
+; GFX950-NEXT:    v_and_b32_e32 v0, 0x7fff7fff, v0
 ; GFX950-NEXT:    v_and_b32_e32 v3, 0x7fff7fff, v3
+; GFX950-NEXT:    v_and_b32_e32 v2, 0x7fff7fff, v2
 ; GFX950-NEXT:    v_and_b32_e32 v5, 0x7fff7fff, v5
 ; GFX950-NEXT:    v_and_b32_e32 v4, 0x7fff7fff, v4
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v5, v5
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v4, v4
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v4
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v5
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fabs = call <3 x half> @llvm.fabs.v3f16(<3 x half> %a)
   %b.fabs = call <3 x half> @llvm.fabs.v3f16(<3 x half> %b)
@@ -2502,10 +2453,8 @@ define <3 x half> @v_fmaximum3_v3f16__fneg_all(<3 x half> %a, <3 x half> %b, <3
 ; GFX950-LABEL: v_fmaximum3_v3f16__fneg_all:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3 neg_lo:[1,1,1] neg_hi:[1,1,1]
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2 neg_lo:[1,1,1] neg_hi:[1,1,1]
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v5, v5 neg_lo:[0,1,1] neg_hi:[0,1,1]
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v4, v4 neg_lo:[0,1,1] neg_hi:[0,1,1]
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v4 neg_lo:[1,1,1] neg_hi:[1,1,1]
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v5 neg_lo:[1,1,1] neg_hi:[1,1,1]
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %a.fneg = fneg <3 x half> %a
   %b.fneg = fneg <3 x half> %b
@@ -2567,10 +2516,8 @@ define <3 x half> @v_fmaximum3_v3f16__inlineimm1(<3 x half> %a, <3 x half> %c) {
 ; GFX950-LABEL: v_fmaximum3_v3f16__inlineimm1:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, 2.0, 2.0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 2.0, 2.0 op_sel_hi:[1,0,0]
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 2.0, v2 op_sel_hi:[1,0,1]
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, 2.0, v3
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %a, <3 x half> <half 2.0, half 2.0, half 2.0>)
   %max1 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %max0, <3 x half> %c)
@@ -2632,10 +2579,8 @@ define <3 x half> @v_fmaximum3_v3f16__inlineimm2(<3 x half> %a, <3 x half> %b) {
 ; GFX950-LABEL: v_fmaximum3_v3f16__inlineimm2:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, 4.0, 4.0
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, 4.0, 4.0 op_sel_hi:[1,0,0]
+; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, 4.0 op_sel_hi:[1,1,0]
+; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, 4.0
 ; GFX950-NEXT:    s_setpc_b64 s[30:31]
   %max0 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %a, <3 x half> %b)
   %max1 = call <3 x half> @llvm.maximum.v3f16(<3 x half> %max0, <3 x half> <half 4.0, half 4.0, half 4.0>)
@@ -2703,10 +2648,8 @@ define <4 x half> @v_fmaximum3_v4f16(<4 x half> %a, <4 x half> %b, <4 x half> %c
 ; GFX950-LABEL: v_fmaximum3_v4f16:
 ; GFX950:       ; %bb.0:
 ; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v1, v3, v3
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v0, v2, v2
-; GFX950-NEXT:    v_pk_maximum3_f16 v1, v5, v1, v1
-; GFX950-NEXT:    v_pk_maximum3_f16 v0, v4, v0, v...
[truncated]

arsenm · 2025-02-21T04:12:15Z

Missed in initial gfx950 upstreaming

arsenm · 2025-02-21T05:02:23Z

Merge activity

Feb 21, 12:02 AM EST: A user started a stack merge that includes this pull request via Graphite.
Feb 21, 12:09 AM EST: Graphite rebased this pull request as part of a merge.
Feb 21, 12:11 AM EST: A user merged this pull request with Graphite.

llvm-ci · 2025-02-21T05:23:56Z

LLVM Buildbot has detected a new failure on builder clangd-ubuntu-tsan running on clangd-ubuntu-clang while building llvm at step 6 "test-build-clangd-clangd-index-server-clangd-in...".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/134/builds/13798

Here is the relevant piece of the build log for the reference

Step 6 (test-build-clangd-clangd-index-server-clangd-in...) failure: test (failure)
******************** TEST 'Clangd :: target_info.test' FAILED ********************
Exit Code: 66

Command Output (stderr):
--
RUN: at line 5: rm -rf /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir && mkdir -p /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir
+ rm -rf /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir
+ mkdir -p /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir
RUN: at line 7: echo '[{"directory": "/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir", "command": "/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir/armv7-clang -x c++ the-file.cpp -v", "file": "the-file.cpp"}]' > /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir/compile_commands.json
+ echo '[{"directory": "/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir", "command": "/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir/armv7-clang -x c++ the-file.cpp -v", "file": "the-file.cpp"}]'
RUN: at line 9: sed -e "s|INPUT_DIR|/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir|g" /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/llvm-project/clang-tools-extra/clangd/test/target_info.test > /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test.1
+ sed -e 's|INPUT_DIR|/vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.dir|g' /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/llvm-project/clang-tools-extra/clangd/test/target_info.test
RUN: at line 12: sed -E -e 's|"file://([A-Z]):/|"file:///\1:/|g' /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test.1 > /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test
+ sed -E -e 's|"file://([A-Z]):/|"file:///\1:/|g' /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test.1
RUN: at line 14: clangd -lit-test < /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test 2>&1 | /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/bin/FileCheck -strict-whitespace /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test
+ clangd -lit-test
+ /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/bin/FileCheck -strict-whitespace /vol/worker/clangd-ubuntu-clang/clangd-ubuntu-tsan/build/tools/clang/tools/extra/clangd/test/Output/target_info.test.tmp.test

--

********************

arsenm added the backend:AMDGPU label Feb 21, 2025 — with Graphite App

arsenm requested review from jayfoad, mbrkusanin, pravinjagtap, rampitec and shiltian February 21, 2025 04:11

arsenm marked this pull request as ready for review February 21, 2025 04:11

arsenm mentioned this pull request Feb 21, 2025

AMDGPU: Widen f16 minimum/maximum to v2f16 on gfx950 #128121

Merged

arsenm added this to the LLVM 20.X Release milestone Feb 21, 2025

shiltian approved these changes Feb 21, 2025

View reviewed changes

arsenm force-pushed the users/arsenm/gfx950/widen-f16-fminimum-fmaximum branch 2 times, most recently from 8b17104 to e77fd5b Compare February 21, 2025 05:05

Base automatically changed from users/arsenm/gfx950/widen-f16-fminimum-fmaximum to main February 21, 2025 05:08

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950

3263aaf

arsenm force-pushed the users/arsenm/gfx950/form-v2f16-minimum3-maximum3 branch from da446d3 to 3263aaf Compare February 21, 2025 05:09

arsenm merged commit cc46d00 into main Feb 21, 2025
6 of 10 checks passed

arsenm deleted the users/arsenm/gfx950/form-v2f16-minimum3-maximum3 branch February 21, 2025 05:11

searlmc1 pushed a commit to ROCm/llvm-project that referenced this pull request Mar 13, 2025

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 (llvm#128123)

9bd2174

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123

Uh oh!

arsenm commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025 •

edited

Loading

Uh oh!

llvmbot commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025 •

edited

Loading

Uh oh!

Uh oh!

llvm-ci commented Feb 21, 2025

Uh oh!

Uh oh!

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123

AMDGPU: Form v2f16 minimum3/maximum3 on gfx950 #128123

Uh oh!

Conversation

arsenm commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025

Uh oh!

arsenm commented Feb 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merge activity

Uh oh!

Uh oh!

llvm-ci commented Feb 21, 2025

Uh oh!

Uh oh!

arsenm commented Feb 21, 2025 •

edited

Loading

arsenm commented Feb 21, 2025 •

edited

Loading