From e267ce24e37b9b7f98921f75abc150c1463b1d6d Mon Sep 17 00:00:00 2001
From: jsmall-nvidia <jsmall@nvidia.com>
Date: Fri, 27 Mar 2020 16:16:27 -0400
Subject: Adds WaveShuffle intrinsic (#1298)

* Support for WaveReadLaneAt with dynamic (but uniform across Wave) on Vk by enabling VK1.4.
Fixed wave-lane-at.slang test to test with laneId that is uniform across the Wave.

* Added WaveShuffle intrinsic.
Test for WaveShuffle intrinsic.

* Added some documentation on WaveShuffle

* Fix that version required for subgroupBroadcast to be non constexpr is actually 1.5
---
 tests/hlsl-intrinsic/wave-lane-at-vk.slang         |  2 +-
 tests/hlsl-intrinsic/wave-lane-at.slang            |  2 +-
 tests/hlsl-intrinsic/wave-shuffle-vk.slang         | 33 +++++++++++++++++
 .../wave-shuffle-vk.slang.expected.txt             |  4 +++
 tests/hlsl-intrinsic/wave-shuffle.slang            | 42 ++++++++++++++++++++++
 .../hlsl-intrinsic/wave-shuffle.slang.expected.txt |  4 +++
 6 files changed, 85 insertions(+), 2 deletions(-)
 create mode 100644 tests/hlsl-intrinsic/wave-shuffle-vk.slang
 create mode 100644 tests/hlsl-intrinsic/wave-shuffle-vk.slang.expected.txt
 create mode 100644 tests/hlsl-intrinsic/wave-shuffle.slang
 create mode 100644 tests/hlsl-intrinsic/wave-shuffle.slang.expected.txt

(limited to 'tests')

diff --git a/tests/hlsl-intrinsic/wave-lane-at-vk.slang b/tests/hlsl-intrinsic/wave-lane-at-vk.slang
index 137632f16..0d52f781e 100644
--- a/tests/hlsl-intrinsic/wave-lane-at-vk.slang
+++ b/tests/hlsl-intrinsic/wave-lane-at-vk.slang
@@ -9,7 +9,7 @@
 RWStructuredBuffer<int> outputBuffer;
 
 // `The input lane index must be uniform across the wave.`. 
-// The same restriction applies to glsl/SPIR-V 1.4
+// The same restriction applies to glsl/SPIR-V 1.5
 // So we are going to use the input buffer to achieve this.
 
 //TEST_INPUT:ubuffer(data=[1 2 3 0], stride=4):name inputBuffer
diff --git a/tests/hlsl-intrinsic/wave-lane-at.slang b/tests/hlsl-intrinsic/wave-lane-at.slang
index 8661dbc55..c3caaa4e8 100644
--- a/tests/hlsl-intrinsic/wave-lane-at.slang
+++ b/tests/hlsl-intrinsic/wave-lane-at.slang
@@ -9,7 +9,7 @@
 RWStructuredBuffer<int> outputBuffer;
 
 // Note from HLSL: `The input lane index must be uniform across the wave.`. 
-// The same restriction applies to glsl/SPIR-V 1.4
+// The same restriction applies to glsl/SPIR-V 1.5
 // So we are going to use the input buffer to achieve this.
 
 //TEST_INPUT:ubuffer(data=[1 2 3 0], stride=4):name inputBuffer
diff --git a/tests/hlsl-intrinsic/wave-shuffle-vk.slang b/tests/hlsl-intrinsic/wave-shuffle-vk.slang
new file mode 100644
index 000000000..01fb59155
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-shuffle-vk.slang
@@ -0,0 +1,33 @@
+// Disabled because main tests is wave-shuffle.slang, this just tests VK 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+
+//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<int> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = int(dispatchThreadID.x);
+    
+    int value = 0;
+    
+    // Scalar
+    
+    value += WaveShuffle(idx, (idx + 1) & 3);
+    
+    // vector
+    
+    {
+        float2 v = float2(idx + 1, idx + 2);
+        float2 readValue = WaveShuffle(v, (idx - 1) & 3);
+        
+        value += int(readValue[0] + readValue[1]);
+    }
+    
+    outputBuffer[idx] = value;
+}
\ No newline at end of file
diff --git a/tests/hlsl-intrinsic/wave-shuffle-vk.slang.expected.txt b/tests/hlsl-intrinsic/wave-shuffle-vk.slang.expected.txt
new file mode 100644
index 000000000..b20444fc5
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-shuffle-vk.slang.expected.txt
@@ -0,0 +1,4 @@
+A
+5
+8
+7
diff --git a/tests/hlsl-intrinsic/wave-shuffle.slang b/tests/hlsl-intrinsic/wave-shuffle.slang
new file mode 100644
index 000000000..093babcce
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-shuffle.slang
@@ -0,0 +1,42 @@
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
+//Disabled on D3D, because in general WaveShuffle requires hardware that doesn't have the 'uniform laneId across Wave' restriction. 
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -use-dxil -profile cs_6_0
+// Disabled because vk doesn't currently support matrix types. See wave-shuffle-vk.slang
+//DISABLE_TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute
+//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
+
+//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer
+RWStructuredBuffer<int> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    int idx = int(dispatchThreadID.x);
+    
+    int value = 0;
+    
+    // Scalar
+    
+    value += WaveShuffle(idx, (idx + 1) & 3);
+    
+    // vector
+    
+    {
+        float2 v = float2(idx + 1, idx + 2);
+        float2 readValue = WaveShuffle(v, (idx - 1) & 3);
+        
+        value += int(readValue[0] + readValue[1]);
+    }
+        
+    // matrix
+    {
+        matrix<int, 2, 2> v = matrix<int, 2, 2>(idx, idx - 1, idx * 3, idx - 2);
+        
+        matrix<int, 2, 2> readValue = WaveShuffle(v, (idx - 1) & 3);
+        
+        value += int(readValue[0][0] + readValue[0][1] + readValue[1][0] + readValue[1][1]);
+    }
+    
+    outputBuffer[idx] = value;
+}
\ No newline at end of file
diff --git a/tests/hlsl-intrinsic/wave-shuffle.slang.expected.txt b/tests/hlsl-intrinsic/wave-shuffle.slang.expected.txt
new file mode 100644
index 000000000..a327b0804
--- /dev/null
+++ b/tests/hlsl-intrinsic/wave-shuffle.slang.expected.txt
@@ -0,0 +1,4 @@
+19
+2
+B
+10
-- 
cgit v1.2.3