Avoid atomic ops on vector members

* This doesn't translate well to metalsl and is in general not very useful - we
  were only doing it to avoid packing issues which in some cases are moot
  anyway. E.g. the histogram buffer doesn't have to be std140 and packed like an
  array of vectors, it can be std430 and packed like an array of uints.

renderdoc/data/glsl/histogram.comp

@@ -24,9 +24,9 @@
 
 //#include "texsample.h" // while includes aren't supported in glslang, this will be added in code
 
-layout(binding=0, std140) buffer minmaxresultdest
+layout(binding=0, std430) buffer minmaxresultdest
 {
-	uvec4 result[HGRAM_NUM_BUCKETS];
+	uint result[HGRAM_NUM_BUCKETS];
 } dest;
 
 layout (local_size_x = HGRAM_TILES_PER_BLOCK, local_size_y = HGRAM_TILES_PER_BLOCK) in;
@@ -183,8 +183,8 @@ void main()
 			}
 #endif
 
-			if(bucketIdx >= 0 && bucketIdx < HGRAM_NUM_BUCKETS)
-				atomicAdd(dest.result[bucketIdx].x, 1U);
+			if(bucketIdx < HGRAM_NUM_BUCKETS)
+				atomicAdd(dest.result[bucketIdx], 1U);
 		}
 	}
 }

renderdoc/data/glsl/mesh.comp

@@ -34,7 +34,12 @@ layout(binding = 2, std430) readonly buffer index_data
 
 layout(binding = 3, std140) buffer pickresult_buffer
 {
-	uvec4 counter;
+	uint counter;
+	// individual padding to prevent uint/uint3 packing woes
+	uint pad0;
+	uint pad1;
+	uint pad2;
+
 	uvec4 results[];
 } pickresult;
 
@@ -162,7 +167,7 @@ void trianglePath(uint threadID)
 		float dist1 = distance(pos1.xyz/pos1.w, hitPosition);
 		float dist2 = distance(pos2.xyz/pos2.w, hitPosition);
 
-		uint result_idx = atomicAdd(pickresult.counter.x, 1u);
+		uint result_idx = atomicAdd(pickresult.counter, 1u);
 
 		uint meshVert = vertid0;
 		if (dist1 < dist0 && dist1 < dist2)
@@ -208,7 +213,7 @@ void defaultPath(uint threadID)
 	float len = length(scr - meshpick.coords);
 	if(len < 35.0f)
 	{
-		uint result_idx = atomicAdd(pickresult.counter.x, 1u);
+		uint result_idx = atomicAdd(pickresult.counter, 1u);
 		pickresult.results[result_idx] = uvec4(vertid, idx, floatBitsToUint(len), floatBitsToUint(wpos.z));
 	}
 }

renderdoc/driver/gl/gl_debug.cpp

@@ -500,7 +500,7 @@ void GLReplay::InitDebugData()
 
     drv.glNamedBufferDataEXT(DebugData.minmaxTileResult, byteSize, NULL, eGL_DYNAMIC_DRAW);
     drv.glNamedBufferDataEXT(DebugData.minmaxResult, sizeof(Vec4f) * 2, NULL, eGL_DYNAMIC_READ);
-    drv.glNamedBufferDataEXT(DebugData.histogramBuf, sizeof(uint32_t) * 4 * HGRAM_NUM_BUCKETS, NULL,
+    drv.glNamedBufferDataEXT(DebugData.histogramBuf, sizeof(uint32_t) * HGRAM_NUM_BUCKETS, NULL,
                              eGL_DYNAMIC_READ);
   }
 
@@ -1168,17 +1168,10 @@ bool GLReplay::GetHistogram(ResourceId texid, uint32_t sliceFace, uint32_t mip,
   GL.glMemoryBarrier(GL_SHADER_STORAGE_BARRIER_BIT);
 
   histogram.clear();
-  histogram.resize(HGRAM_NUM_BUCKETS * 4);
+  histogram.resize(HGRAM_NUM_BUCKETS);
 
   GL.glBindBuffer(eGL_COPY_READ_BUFFER, DebugData.histogramBuf);
-  GL.glGetBufferSubData(eGL_COPY_READ_BUFFER, 0, sizeof(uint32_t) * 4 * HGRAM_NUM_BUCKETS,
-                        &histogram[0]);
-
-  // compress down from uvec4, then resize down
-  for(size_t i = 1; i < HGRAM_NUM_BUCKETS; i++)
-    histogram[i] = histogram[i * 4];
-
-  histogram.resize(HGRAM_NUM_BUCKETS);
+  GL.glGetBufferSubData(eGL_COPY_READ_BUFFER, 0, sizeof(uint32_t) * HGRAM_NUM_BUCKETS, &histogram[0]);
 
   if(maxlevel >= 0)
     GL.glTextureParameterivEXT(texname, target, eGL_TEXTURE_MAX_LEVEL, (GLint *)&maxlevel);

renderdoc/driver/vulkan/vk_debug.cpp

@@ -2399,9 +2399,9 @@ void VulkanReplay::HistogramMinMax::Init(WrappedVulkan *driver, VkDescriptorPool
   m_MinMaxResult.Create(driver, driver->GetDev(), sizeof(Vec4f) * 2, 1, GPUBuffer::eGPUBufferSSBO);
   m_MinMaxReadback.Create(driver, driver->GetDev(), sizeof(Vec4f) * 2, 1,
                           GPUBuffer::eGPUBufferReadback);
-  m_HistogramBuf.Create(driver, driver->GetDev(), sizeof(uint32_t) * 4 * HGRAM_NUM_BUCKETS, 1,
+  m_HistogramBuf.Create(driver, driver->GetDev(), sizeof(uint32_t) * HGRAM_NUM_BUCKETS, 1,
                         GPUBuffer::eGPUBufferSSBO);
-  m_HistogramReadback.Create(driver, driver->GetDev(), sizeof(uint32_t) * 4 * HGRAM_NUM_BUCKETS, 1,
+  m_HistogramReadback.Create(driver, driver->GetDev(), sizeof(uint32_t) * HGRAM_NUM_BUCKETS, 1,
                              GPUBuffer::eGPUBufferReadback);
 
   // don't need to ring this, as we hard-sync for readback anyway

renderdoc/driver/vulkan/vk_replay.cpp

@@ -2033,9 +2033,7 @@ bool VulkanReplay::GetHistogram(ResourceId texid, uint32_t sliceFace, uint32_t m
 
   uint32_t *buckets = (uint32_t *)m_Histogram.m_HistogramReadback.Map(NULL);
 
-  histogram.resize(HGRAM_NUM_BUCKETS);
-  for(size_t i = 0; i < HGRAM_NUM_BUCKETS; i++)
-    histogram[i] = buckets[i * 4];
+  histogram.assign(buckets, buckets + HGRAM_NUM_BUCKETS);
 
   m_Histogram.m_HistogramReadback.Unmap();